Abstract

Despite recent media stories about both labor unions and the potential revitalization of U.S. manufacturing, most current policy discussions about improving business climate to foster manufacturing neglect the role of unions. This, plus the continued decline in U.S. union membership, might lead one to believe that unions matter little for new investment decisions.

This essay argues that, in fact, unions remain an extremely significant factor in decisions by U.S. manufacturers about where they will or will not make new investments. Both unions and manufacturing are discussed in an analysis that distinguishes between new investment at new plants and at existing plants. Two central arguments are presented:

(1) Union success (or lack thereof) in organizing new plants is a reflection, in part, of an intentional strategy by firms to choose locations that have historically not been receptive to unions, in the South and in rural areas. This well-established historical process continues today. That is, unions still make a difference for new investment in manufacturing because they influence where firms decide to open new plants.

(2) Unions also remain relevant for corporate decisions about new investment at existing plants. Many such facilities are hubs of interaction between unionized blue-collar workers and nonunion white-collar workers, including researchers and engineers in research and development labs. To continue this valued interaction at a new nonunion plant, the firm would have to shift white-collar workers, at potentially high cost. The firm might instead consider adding new investment to an existing facility. In this way, the new investment keeps alive a union established long ago.

Through its influence on the ease of labor organizing, policy can therefore influence both the location and the amount of new investment in U.S. manufacturing.

Introduction

With the decline of labor union membership in the United States over recent decades, discussions of policy toward unions usually show up in the back pages of newspapers, if at all. But recently, labor union policy has been front-page news. One major story is that in 2011, the National Labor Relations Board (NLRB) began proceedings to block Boeing, the largest manufacturing exporter in the United States, from opening a billion-dollar plant in South Carolina, due to alleged labor law violations. That the (now-resolved) dispute even made headlines is significant news in an era of supposed union irrelevance.

Another major labor story involves efforts in several states to pass “right-to-work” laws, anti-union statutes that prohibit making union membership a requirement of employment. In December 2012, Michigan, a traditional center of union power, enacted a right-to-work law, joining the ranks of anti-union states in the South that passed such laws over 50 years ago. Indiana did so in February 2012, and Wisconsin enacted a related law for public sector unions in 2011.

Also on the front page are discussions of a potential revitalization of American manufacturing. The automobile industry has been in recovery since the 2009 crisis. General Electric’s (GE’s) “reverse offshoring” of water heater production from China back to Kentucky got substantial media attention, as did Boeing’s rollout of the new fuel-efficient 787 Dreamliner, which it hopes will be a key source of competitive advantage for years to come.

Despite these stories, overall gains in manufacturing have been meager relative to the broad decline of U.S. manufacturing since the 1970s, and many, including President Barack Obama, argue that it should be an important policy priority to promote U.S. manufacturing. For example, a recent presidential report urges improvements in the business climate for manufacturing (President’s Council of Advisors on Science and Technology, July 2012).

It’s telling, perhaps, that this report doesn’t mention unions in its discussion of business climate, consistent with a view that unions are largely irrelevant to corporate decisions about investment in manufacturing. The share of the manufacturing workforce in unions has been in free fall for many years, only 9.6 percent in 2012, compared with 38.9 percent in 1973.¹ This statistic actually understates current union weakness, because factories that are unionized today, to a remarkable degree, are the legacy of union victories over 50 years ago.

These facts might lead one to the view that while unions were relevant to old investment decisions (in locations where unions were established decades ago under a more favorable environment), they matter little for new investment decisions. In discussions of business climate, policymakers and businesspeople think about new investments, of course, not investments made years ago. Given today’s small union membership numbers, it might seem sensible to leave unions out of the discussion about current business climate and new investment.

I believe that this conclusion is ill founded, and in this essay, I will argue that in fact unions remain an extremely significant factor in decisions by U.S. manufacturers about where they will or will not make new investments. To make this point, I discuss both unions and manufacturing, and I present an analysis that distinguishes between new investment at new plants and at existing plants.

I argue first that low union success in organizing new plants is not an accident, but rather a reflection, in part, of an intentional strategy by firms to choose locations that have historically not been receptive to unions, in the South and in rural areas. True, this is an old story, a process that has been going on for decades.² My point is that this process continues today. That is, unions still make a difference for new investment in manufacturing because they influence where firms decide to open new plants.

Second, I argue that unions remain quite relevant for corporate decisions about new investment being considered at existing plants. Many such facilities are significant hubs of interaction between unionized blue-collar workers and nonunion white-collar workers, including researchers and engineers in research and development (R&D) labs. These facilities are old (in some cases 100 years or more!), and unions at them were generally organized just after the National Labor Relations Act of 1935 provided a favorable environment to do so.

If a firm with such a facility were to shift production workers to a new nonunion plant, it would have to shift the white-collar workers as well, if it wants to continue the interactions. It might be costly to break up an existing successful research center, and so the firm might instead consider adding new investment to an existing facility.³ In this way, the new investment keeps alive a union established long ago. Public policy that affects such a firm’s interactions with the incumbent union and its bargaining strength then potentially affects the business climate in which the decision about new investment is made.

To illustrate the continuing relevance of unions to investment decisions, consider again GE’s decision to bring production back from China to its appliance plant in Kentucky. The Kentucky plant is old and has long been union. It is also the headquarters for GE’s appliance business and the R&D center. In public statements, including comments by CEO Jeff Immelt, GE makes explicit the high value it places on having innovation and production at the same location.⁴ GE sustained this co-location by choosing to add new investment to its already unionized plant. However, it is important to emphasize the role recent weakness of unions potentially played in providing a favorable climate for the investment. As part of the deal, the union made a concession that the new workers be paid $10 less per hour than existing workers.⁵ This kind of two-tiered wage structure is anathema to union solidarity, and a concession like this was rarely made in earlier periods when unions were strong.

Consider also the NLRB’s 2011 case against Boeing. Historically, Boeing’s base of production is its heavily unionized facilities in Washington state. (I say “heavily” because even engineers there are in a union.) Boeing has had a rocky relationship with its unions over the years, and strikes are a regular occurrence. In 2010, Boeing began opening a second Dreamliner production line in a South Carolina nonunion plant; “only the third site in the world to assemble and deliver twin-aisle commercial airplanes,” according to Boeing.⁶ CEO Jim McNerney explained that Boeing was doing this because the company was tired of “strikes happening every three to four years in Puget Sound.”⁷ Based on these remarks and others like it, the NLRB filed its case accusing Boeing of an illegal labor practice regarding threats firms can make about how they might respond to strikes.

I offer the Boeing CEO’s expressed motivation for moving to South Carolina as “Exhibit A” for my case that big manufacturers even today are choosing locations to avoid unions. However, company officials have to be very careful about public statements on this issue because these statements have legal ramifications. Hence, for the analysis I will focus on what firms do, rather than on what their officials say. By observing the choices firms make when they decide where to make new investments, I can draw inferences about what matters to them most.

The main work of this paper is an analysis of recent investment behavior by GE, which will serve as “Exhibit B.” Putting GE under the microscope reveals a picture with a great deal of clarity. In the recent period that I look at, whenever GE has built a brand new plant, it has picked a location unlikely to be unionized. And when GE has invested in an existing unionized facility, for the vast majority of new jobs involved, the facility was one with significant R&D presence, and new workers were hired at a lower wage tier than existing employees.

This is a case study of two firms. While these are two very important firms—the two largest manufacturing exporters in the U.S.—as in any case study, there is always an issue of the broader applicability of the results. I believe the insights of this analysis hold more broadly for large U.S. companies in heavy industry, and I give two quick examples to back this up. First, Caterpillar, the construction-equipment manufacturer, is another firm high on the list of top exporters. Union avoidance in this firm’s investment decisions has been very much in the recent news.⁸ Second, if I had included the auto industry in this study (and, in particular, the site-selection decisions of foreign-owned firms), I expect that many of the conclusions would be similar. Foreign automakers in every case have chosen plant locations where they have been able to remain nonunion.

Background

Several key points about firms and unions will aid discussion of the case studies that follow.

1. Unions are organized at the plant level; once established, they seldom disappear.
   Generally speaking, union organization takes place at the plant level, involving a representation election supervised by the NLRB. Once a union gets in a plant and, in particular, is able to negotiate a first contract, it becomes entrenched over time. An NLRB mechanism for decertifying a union does exist, but it affects only a trivial number of cases. In 2005, for instance, unions representing 11,000 workers were decertified, but out of a base of 9 million represented private sector workers, this is a decertification rate of only 0.13 percent.⁹ Hence, once a union becomes entrenched at a plant, it is generally there for good, until the plant shuts down.
2. Unions spread to neighboring establishments, so firms often build new plants in distant areas.
   Unions tend to spill out of organized plants into nearby businesses; that is, to some degree unions are “contagious.” In Holmes (2006), I provide evidence on this point, showing how unions in steel mills, auto plants and coal mines found their way into neighboring grocery stores and health care facilities. If a union can spread from an auto plant to a nursing home down the street, it can likely extend to a neighboring auto plant. Aware of this, firms understand that starting a new nonunion plant generally requires geographic separation from existing unionized plants.
3. Manufacturers may augment existing unionized plants if benefits outweigh costs.
   If a manufacturer invests and adds production worker jobs to an existing unionized plant, the new workers usually join the current union. The manufacturer may make this decision, rather than open a new nonunion facility elsewhere, if the initial site has advantages, like proximity to R&D labs, that offset the disadvantage of being unionized. In this way, a unionization event from many years ago is kept alive.

General Electric

With that as background, I’ll now turn to the meat of the essay where I analyze what key manufacturers are doing. I focus on GE, but I also come back to Boeing.

GE is one of most influential U.S. companies. It is the second largest U.S. manufacturing exporter (after Boeing). It is the third most innovative U.S. firm, measured in terms of patent counts (after IBM and Microsoft).¹⁰ It is at the center of discussions about revitalization of U.S. manufacturing. Immelt is highly visible in this discussion and serves on the Council on Jobs and Competitiveness set up by Obama.

GE is also interesting for my purposes because it has a long history of having both union and nonunion operations. It has long held a reputation of taking a tough stance in dealing with unions. (See the discussion in Meyer (2001), for example.) Here, I take a look at its recent behavior regarding plant openings and new investment.

GE publicizes its new plant openings and investments in an internet series called “GE Reports,” under the category “jobs.”¹¹ I reviewed all announcements in the series published over the four-year period Jan. 1, 2009, to Dec. 31, 2012, and created a data set of new plant openings and expansions. I restricted attention to announcements in which new jobs were added and excluded announcements for GE Capital and GE Corporate, in order to focus on the manufacturing divisions. When multiple expansions occurred at the same location—for example, the appliance factory in Louisville, Ky., mentioned in the introduction had three expansions during this period—I combined the records. After going through 93 announcements and combining information this way, I found 24 locations in which new investment and job growth were announced over the four-year period, with a total of 8,344 new jobs. The 24 locations are listed in Tables 1, 2 and 3.

In constructing the tables, I first categorize locations as new or existing.¹² In my definition of existing, I include brand new buildings and facilities that are part of a larger preexisting GE campus. For example, there is battery factory in Schenectady, N.Y., that was described as new in the announcement, but I classified it as preexisting because it was added to GE’s main campus in Schenectady, which serves as its headquarters location and the site of a number of existing facilities.¹³ Using this classification system, I determined that of the 24 locations receiving new investment, eight were new locations and 16 were existing locations. Table 1 lists the new facilities.

Take a look at the locations of the eight new plants. With one exception, a plant in Michigan discussed below, they are all in locations where unions are weak: two aviation plants in Mississippi, a locomotive plant in Texas, other locations in the South. A partial exception: a non-South location in Colorado, a state where unions are relatively weak. The full exception: GE’s new facility in Michigan, in the Detroit area, a center of union power. But this, as it turns out, is an R&D center, with only white-collar labor;¹⁴ unionization is thus a nonissue.

Of course, union avoidance is only one of many factors considered in a plant location decision. For example, states in the South getting the new plants may have offered better tax incentives than other potential sites in northern states. In fact, GE’s CEO is on record as saying that tax incentives matter in site selection.¹⁵ But this is why GE’s choice to put the R&D center in the Detroit area is interesting. If taxes are the primary consideration and taxes are lower in the South, I might expect the R&D center to be put in the South as well. With a case study of only eight data points, I cannot draw definitive conclusions. Nonetheless, it is striking that a simple theory that GE picks nonunion locations when unions matter gets it right eight out of eight tries. Along with the other evidence from Boeing, it suggests a pattern of behavior.

I next turn to new investment at the 16 locations where GE already had facilities. I classify these plants as “union” or “nonunion” depending on whether the location has workers represented by a union (based on various public sources).¹⁶ The nine nonunion facilities are listed in Table 2, and the seven union plants are listed in Table 3.

Two comments about the nine nonunion facilities. Note first, there is a nonunion GE aviation plant in Michigan. As this is a production facility with blue-collar workers, it might be surprising that it has remained nonunion in Michigan. However, the plant is in western Michigan, where unions are not as strong. Next, note the nonunion GE transportation facility in Grove City, Pa. The plant makes engines for a locomotive plant in Erie, listed in Table 3 in the “union” category. The Erie locomotive plant dates from 1913 and has been a union plant since 1940.¹⁷ The engine plant in Grove City dates from 1971 and has remained nonunion, despite the connection with the union plant in Erie.¹⁸ Apparently, the 85-mile distance between the two locations has been enough to keep the union in Erie out of the Grove City plant.

I now turn to the seven union plants that received new investment, listed in Table 3. The plants are sorted from the highest number of new jobs to the lowest, and I focus on the top three, highlighted in bold. These are the GE energy facility at Schenectady, with 1,200 new jobs, the GE appliance facility in Louisville, with 1,130 new jobs, and the GE transportation facility in Erie, with 610 new jobs. Together they account for the vast majority of new jobs in union plants, 2,940 out of 3,518.

The last three columns of Table 3 reveal interesting facts about these three facilities. First, each is the respective headquarters for its division. Second, each of these three locations has an R&D lab on site.¹⁹ Third, each of the three locations is a successful producer of a large number of patents. I base this on calculations with publicly available U.S. patent data. I extracted all granted patents assigned to GE over the period 2000-11. In the data for each patent, the location of each inventor is provided. The last column of Table 3 reports the number of GE patents over this period with at least one inventor in each of the given locations. ²⁰ Schenectady, the overall GE headquarters, has 4,348 granted patents over the period, while Louisville has 280 and Erie has 351. This is an impressive amount of innovative output.

Earlier, I argued that if production workers are unionized at a facility, the location disadvantage for new investment of the existing union could potentially be offset by beneficial co-location with R&D activity and other white-collar work. I see evidence for this claim in GE’s investment behavior. The vast majority of new investment in unionized facilities has occurred in plants with significant R&D and connections to headquarters.

I consider one last issue for the seven union plants receiving new investment: What is happening to the net number of union jobs at each of the facilities? The “GE Reports” series mentions expansions leading to new jobs to publicize GE’s contribution to U.S. employment, but it doesn’t publicize job cuts through efficiencies or outsourcing. To look at the net effect on union jobs, I use data from the Department of Labor on union membership for each of the union locals at the respective plants.²¹ Membership by local and year are reported in Table 4, and the bottom row tabulates the sum across all seven union plants receiving new investment. Membership at these seven facilities between 2010 and 2011 increased from 7,592 to 8,710 workers, consistent with GE’s message that it is increasing production worker employment at these plants.

However, the recent gain is not enough to offset the fall from 2007. Moreover, these are the selection of union plants getting new investment. I have looked at some of the other large unionized plants not getting new investment, and membership is falling in these plants. One takeaway point is that even though GE is putting some new investment in unionized plants that for historical reasons are connected to headquarters and R&D facilities, this force is not strong enough to offset continual decline of the unionized workforce at GE.

Boeing

Let’s get back to the earlier story about Boeing, where I noted that the NLRB had filed a complaint against Boeing in 2011. The complaint alleged that Boeing had engaged in an unlawful labor practice, by making public statements that it was moving production to a nonunion facility to avoid strikes.²² As a remedy, the acting general counsel sought a court order that Boeing be forced to open the second production line in a union facility in the Washington state area instead of the nonunion facility in South Carolina.

In the end, the issue was resolved by Boeing agreeing to add additional union jobs in Washington state in return for the union dropping the charges, enabling Boeing to go ahead with the South Carolina plant.²³ The story illustrates both kinds of investment highlighted in this essay. First, there is new investment at a location where unions are weak, at a site where Boeing did not have a previous facility. ²⁴ Second, there is new investment at an existing unionized facility, at a site close to Boeing’s R&D infrastructure and other white-collar activity.

The story has two epilogues. In January 2012, Boeing announced that it was closing its entire operations in Wichita, Kan., a unionized facility. (Kansas is not known as a strong union state, but the facility in question dates to 1927, and old facilities in heavy industry are generally union, no matter where they are located.) Many of the jobs were shifted to nonunion facilities in Texas and Oklahoma, some to union facilities in Washington state and some cut altogether. Various news articles report cutbacks in defense spending as the driving factor behind this closure.²⁵ Even so, it is also clear that this decision has implications for the “chess game” of labor management relations going forward, with a longstanding union outpost eliminated and nonunion activity expanded.

The second epilogue is that Boeing’s main union is currently trying to unionize the South Carolina plant.²⁶ Clearly, Boeing has an incentive to try to keep the workers happy enough that they won’t want the union. And it is reasonable to expect that the workers would be familiar with earlier statements by company officials that a nonunion workforce is why Boeing came in the first place. (Public officials in South Carolina have actually reminded the workers on this point. ²⁷) If the South Carolina workers were to vote in the union, they will be giving up the competitive advantage they hold over union workers in Washington state in future competition for new plant investment. Obviously, this situation puts the union in a weak position.

Remarks about Labor Relations Policy

Public policy affects the extent of unions. For example, the 1935 passage of the National Labor Relations Act was followed by a huge surge in the share of unionized workers. (See Freeman 1998.) Think of there being a policy lever, where how high the lever is pushed determines how easy it is for unions to organize in a workplace. For example, in 2009 at the beginning of Obama’s first term, when the Democrats controlled both houses of Congress, there was discussion of the “Employee Free Choice Act,” a bill to allow unions to substitute the secret ballot in an NLRB-supervised election with a system where union organizers collect signed cards from workers.

This policy, called “card check,” would be a significant upward push on the policy lever. (With the new Congress, it is currently not under consideration.) The NLRB recently made administrative rule changes to speed up union representation elections.²⁸ This is an upward push on the lever, because employers have less time to respond. The right-to-work laws recently enacted in Michigan and Indiana push the lever down. In addition to the direct negative effect on unions in these two states, there will likely be a broader negative effect on unions throughout the country. These laws make it harder to collect union dues, and this can potentially lessen the resources available for organizing in other states. For example, when the autoworkers union conducts organizing drives at nonunion auto plants in the South, they are funded by autoworkers’ dues in states like Michigan and Indiana.

Suppose the pro-union organizing policy lever gets pushed up so high that the union gets into Boeing’s new South Carolina plant, and Boeing expects that this will be true for other new plants it might open in the South. Based on the findings above, how will this policy change affect new manufacturing investment?

The analysis above presents evidence that even today, big firms like Boeing and GE are selecting locations to avoid unions. If Boeing were to get a union even in South Carolina, it will have less incentive to shift production from Washington state to South Carolina. Thus, an increase in the policy lever potentially affects where new investment goes within the United States.

In addition to where, the policy lever can potentially affect how much overall new investment there is in this country. If one accepts the proposition that firms choose locations within the United States to avoid unions, then one has to consider the possibility that a change in policy might lead the firm to not invest in the United States. That is, if policy changes so that the firm gets a union no matter where in this country it goes, it might consider investing abroad or not investing at all. In the NLRB case referred to above, the NLRB notes that Boeing has experienced strikes by production workers in 1977, 1989, 1995, 2005 and 2008. In December 2012, Boeing’s engineers union leaders in Seattle said that “the likelihood of a strike is very high,” and though negotiations continued in early 2013, prospects for settlement on a contract remained distant.²⁹ Dealing with strikes on a regular basis can only make Boeing less competitive in the world marketplace, diminishing the returns to new investment.

Large Image

Large Image

Large Image

Large Image

Endnotes

1 Statistics on union membership share are based on the Current Population Survey conducted by the U.S. Census Bureau and were obtained from tabulations published at unionstats.com.

2 Fuchs (1962) is an early work arguing for the important role of unions in the migration of industry to the South. See also Holmes (1998) for a discussion of the role of anti-union policies pursued in Southern states.

3 I note two costs in particular. First, key researchers might be unwilling to move. Second, there is much discussion in the economics literature for how R&D centers potentially benefit from knowledge spillovers from other researchers in the vicinity. If the R&D center is moved, it might lose access to these beneficial spillovers.

4 See, in particular, Immelt’s comments in the Harvard Business Review, Immelt (2012). He highlights the Kentucky appliance plant and writes, “Our success on the factory floor rests on human innovation and technical innovation.” He adds, “Engineering and manufacturing are hands-on and interactive … at a time when speed to market is everything, separating design and development from manufacturing didn’t make sense.”

5 In discussing GE’s decision to invest in the Kentucky plant, Immelt writes, “The third element in human innovation is a new model for labor relations. ... The union accepted a lower wage for new hires, we pledged to create new jobs” Immelt (2012). For more on the story, see “G.E. to Add Two New U.S. Plants as Unions Agree on Cost Controls,” New York Times, Aug. 6, 2009.

6 This is how Boeing’s website describes the South Carolina facility. The other two are the Boeing facility in Everett, Wash., and the airbus facility in Toulouse, France.

7 The CEO is quoted in the case document, NLRB Case 19-CA-32431, dated April 20, 2011. The brief also quotes similar comments made by other company officials.

8 For a story about Caterpillar closing a union plant in Ontario and transferring jobs to a nonunion plant in newly right-to-work Indiana, see “As Unions Lose Their Grip, Indiana Lures Manufacturing Jobs,” Wall Street Journal, March 18, 2012.

9 This statistic is based on the author’s calculations with the raw NLRB election data. The statistic includes cases where unions were decertified and replaced with an alternative union. Dickens and Leonard (1984) report an analogous estimate with earlier data that is the same order of magnitude.

10 The patent count figure is as reported for 2012 by IFI CLAIMS. The claim about exporting is one regularly made by GE. See, for example, GE Reports.

11 See GE Reports.

12 I use GE’s records in the Million Dollar Directory of Dun and Bradstreet to build a database of GE’s manufacturing plants. I merge this with plant information over the 1987-2010 period in the Toxic Release Inventory published by the Environmental Protection Agency, which can be used to determine when a plant is emitting pollution and is therefore in operation. I combined these data with the GE announcement information to distinguish new and existing plants.

13 See “New York powers up with new GE battery plant,” GE Reports, May 12, 2009.

14 See “GE to bring research center and 1,100 jobs to Michigan,” GE Reports, June 26, 2009.

15 See comments in Immelt (2012).

16 The master 2007-11 GE contract lists all facilities party to the contract that were represented by IUE-CWA, the largest union at GE. I also used government data from the Federal Mediation and Conciliation Service, which publishes information about the location of facilities with expiring union contracts. I resolved ambiguous cases through web searches, including inspection of various websites of local and national unions.

17 See A Brief History of UE Bargaining with GE: Seventy Years of Struggle, United Electrical, Radio and Machine Workers of America (undated manuscript), and GE Transportation BusinessWire news release, “GE Transportation Celebrates 40 Years in Grove City,” Aug. 6, 2011.

18 The age of the plant is based on “GE Transportation Celebrates 40 Years in Grove City.” Given Pennsylvania’s tradition of strong unions, the fact that GE has a nonunion plant there may come as a surprise. Two points are worth noting in addition to the geographic separation with the Erie plant noted in the text. First, the Grove City plant did not exist in the 1940-70 era when labor organizing at plants was easier. Second, it is in a rural area away from other unions.

19 Specifically, each location cited is listed in the Directory of American Research and Technology, 23rd ed., R. R. Bowker, Reed Elsevier, New Providence, N.J., 1998.

20 The patent data report the city and state of a given inventor, but generally not the address. Table 2 reports the count of patents with at least one inventor in the given city and state.

21 The data are the LM Filing Data, published by the Office of Labor-Management Standards at its website. For all but two exceptions, I used the disaggregated membership information in the file, which is useful for separating out membership in the local not in a GE bargaining unit. For the Bucyrus and Madisonville units, only total local membership is available, but this should not be a problem because both appear to represent only GE employees.

22 The complaint is NLRB Case 19-CA-32431, dated April 20, 2011.

23 See, “Union Seeks to Dismiss Complaint Against Boeing,” New York Times, Dec. 9, 2011.

24 For brevity, I am glossing over details. In July 2009, Boeing purchased a supplier plant in South Carolina that already had a union. The plant workers voted to decertify the union in September, and subsequently in October Boeing announced it was going to build the second line in South Carolina.

25 See “Boeing to Shut Wichita Plant, Citing Cuts at Pentagon,” New York Times, Jan. 4, 2012.

26 See “Boeing faces union drive at 787 plant in South Carolina,” Reuters, Oct. 12, 2012.

27 Seattle Times, Oct. 22, 2012, quoted Sen. Jim DeMint, R-S.C., as saying, “It would blow me away if the employees of Boeing here were so foolish as to unionize when that was one of the key reasons that this plant was built.”

28 See “Labor Board Adopts Rules to Speed Unionization Votes,” New York Times, Dec. 11, 2011.

29 See “Boeing’s engineer unions says strike is likely, prepares workers,” Reuters, Dec. 10, 2012. See also “Boeing, engineers set to resume contract talks Wednesday,” Reuters, Jan. 14, 2013.

References

Dickens, William T., and Jonathan S. Leonard. 1984. “Accounting for the Decline in Union Membership, 1950-1980.” Industrial and Labor Relations Review 38: 323.

Freeman, Richard B. 1998. “Spurts in Union Growth: Defining Moments and Social Processes,” in The Defining Moment: The Great Depression and the American Economy in the Twentieth Century.” Michael D. Bordo, Claudia Goldin and Eugene N. White, eds. Chicago: University of Chicago Press, pp. 265-95.

Fuchs, Victor R. 1962. Changes in the Location of Manufacturing in the United States since 1929. Yale University Press.

Holmes, Thomas J. 1998. “The Effects of State Policies on the Location of Industry: Evidence from State Borders.” Journal of Political Economy 106 (August): 667-705.

Holmes, Thomas J. 2006. “Geographic Spillover of Unionism.” Research Department Staff Report 368. Federal Reserve Bank of Minneapolis.

Immelt, Jeffrey R. 2012. “The CEO of General Electric on Sparking an American Manufacturing Renewal.” Harvard Business Review 90 (March): 43-46.

Meyer, Douglas. 2001. “Building Union Power in the Global Economy: A Case Study of the Coordinated Bargaining Committee of General Electric Unions (CBC).” Labor Studies Journal (Spring): 60-75.

President’s Council of Advisors on Science and Technology. 2012. “Report to the President on Capturing Domestic Competitive Advantage in Advanced Manufacturing.” Manuscript.

This Economic Policy Paper was originally published in the December 2012 issue of The Region.

Abstract

Banks are prone to panic-induced runs due to their traditional structure of short-term, unconditional liabilities and long-term, illiquid assets. To avoid systemic crises caused by such panics, governments tend to bail out failing banks. Traditional banking systems thus impose external costs. Three major theoretical benefits are often used to justify a banking system that relies on short-term debt despite these costs: (1) maturity transformation, (2) efficient monitoring of bank managers and (3) facilitation of financial transactions. In a previous paper, we argued that the first two justifications, while seemingly compelling, actually suggest financial arrangements very different from our current system. In this paper, we examine the third justification, that a banking system reliant on short-term debt is essential for the facilitation of transactions. We find, in fact, that this reliance is more costly than generally recognized and, moreover, that socially beneficial financial transactions can and should be provided at less cost and risk by both restricting and broadening the payments system. Transactions should be restricted to institutions that continuously mark to market the value of their assets and issue equity claims to owners. Such accounts should also be broadened to include financial vehicles that are readily available, thanks to advances in information and communication technologies, and possibly quite different from current banks.

Introduction¹

The financial crisis of 2007-08 and consequent Great Recession generated substantial discussion and debate over future banking regulation. Largely absent, however, has been a careful reexamination of whether the beneficial services provided by traditional banks outweigh the inherent financial fragilities of those banks and their associated costs to society.

Three major benefits are usually said to justify traditional bank reliance on short-term debt, the source of their inherent fragility. In a previous article, we assessed—and found wanting—two of these proposed rationales: (1) the benefit of maturity transformation, or creation of long-term financial assets from shorter-term assets and (2) the benefit of efficient monitoring of bank managers, through appropriate alignment of investor incentives. (See Chari and Phelan 2012a.)

Here we discuss the third justification, that traditional banks are beneficial and necessary because they provide payments services essential to the efficient function of modern economies. We conclude that while this rationale was compelling in an earlier historical era—prior to modern advances in information and communication technology that facilitate transactions of all sorts—the necessary services can now be provided through existing financial vehicles that do not rely on traditionally structured, inherently fragile banks.

We begin by briefly reviewing the structural source of traditional bank fragility and proceed to consideration of the necessity of banks, despite this fragility. We then address the main topic of this paper: the argument that banks as currently structured are necessary because their demand deposits facilitate financial transactions. We conclude that the current structure of banks is unduly costly to society and that essential payments services can, with modern information and communication technologies, be provided with less fragile and more efficient financial institutions.

The inherent fragility of banks

In what sense are banks and similar financial institutions fragile? Our previous paper discusses this question in detail; here we provide a synopsis, referring interested readers to the earlier discussion.

The assets of financial institutions are, by and large, financial assets, and claims on them are primarily financial liabilities. Their financial assets consist mainly of conditional promises to deliver dollars at future dates. These assets, such as home mortgages, are often long term and illiquid. Their financial liabilities consist mostly of a variety of obligations to deliver dollars at particular dates, under certain circumstances. Banks in particular have liabilities that are mostly short term and unconditional, such as demand deposits and certificates of deposit.

This reliance on short-term debt makes banks fragile in that they are particularly vulnerable to the risks of insolvency and the possibility of confidence crises. Since bank assets are much longer term and illiquid than their liabilities and because the value of these assets fluctuates, a bank’s net worth also fluctuates a great deal.

The illiquidity of banks’ assets and the demandable structure of their liabilities thus expose banks to crises of confidence. Since a bank typically will not be able to meet the demands of all depositors within a short period of time should they all choose to withdraw, banks are vulnerable to self-fulfilling panics in which depositors withdraw their funds simply because they believe other depositors will do so. This panic is an entirely rational response even if the bank is solvent (though illiquid).

Governments have a strong incentive to intervene to bail out debt holders of banks in order to prevent the entire financial system from failing. Paradoxically, expectations of such bailouts can increase the incidence and depth of financial crises. Once depositors believe that their deposits will be protected in the event of systemic failure, they have less incentive to monitor bank managers, who, in turn, have increased incentive to take on risk, knowing their failures are implicitly insured by taxpayers.

In this way, expectations of bailouts can lead financial systems to rely excessively—from a social perspective—on short-term debt to fund long-term assets. Fragile banking systems thus impose external costs, and regulation may therefore be socially desirable.

Are banks necessary?

The fragility of the banking system together with the reality that such fragility may well lead to occasional massive bailouts compel us to ask why societies would choose regulatory systems that allow financial institutions to fund illiquid assets whose value can fluctuate rapidly with short-term debt and demand deposits.

One could perhaps argue that banks were necessary prior to the electronic information age because no other forms of financial intermediation were feasible. With the advent of high-speed computers and modern communications, however, alternative financial institutions can provide similar services with far less potential for crises. We discuss such alternatives later in this paper.

We now examine the possible social benefits of a financial system in which illiquid assets with volatile values are funded by demand deposits and short-term debt. This cost-benefit analysis facilitates the design of a better regulatory system for banks, clearly a matter of considerable importance.

The previous paper examined two of the three major theoretical justifications for the reliance of the banking system on short-term debt: (1) demand deposits allow banks to engage in socially useful maturity transformation and (2) demand deposits allow for efficient monitoring of bank managers. This paper considers the third major justification: (3) demand deposits facilitate financial transactions.

To anticipate our conclusion, we believe that while all three justifications are compelling, they point us to a financial system very different from the one currently in place. The first two justifications suggest that it is important to have institutions that finance long-term assets with short-term debt, but we have argued that the assets that are so funded should not have close substitutes in publicly traded markets. In this paper, we will argue two main points regarding the usefulness of banks in facilitating transactions. First, we argue that regardless of technology, the social benefit to using banks to facilitate transactions is lower than the private benefit, thus potentially explaining why the historical ubiquity of bank-facilitated transactions does not imply their efficiency. Second, we argue that the necessity of bank-facilitated transactions is much less obvious than it was a century ago, before advances in information and communication technologies allowed us to create very different institutions than we currently have to facilitate transactions.

Our analysis will suggest a framework for thinking about regulatory policy for institutions that facilitate payments. The economic case for regulating such institutions is convincing, given that the failure of the payments system imposes significant external costs. We argue that institutions that facilitate payments should primarily issue equity-like claims such as those issued by standard mutual funds. Current practice hopelessly conflates these two economic cases into a single institution called “banks” and exposes the economy to unnecessary risks and recurrent costly bailouts.

Assessment of the transactions facilitation view

The most obvious service that banks provide now, and have offered throughout their ubiquitous existence, is payments services. Historically, banks have allowed individuals and firms to pay for goods and services through their provision of bank checks and other widely accepted claims. Therefore, those individuals and firms haven’t had to resort to costly barter or specie trade.

Here, we raise the possibility that banks exist because they provide a privately useful function—the facilitation of transactions in a form that pays households interest—but the social usefulness is less than the private usefulness.

The starting point of our assessment involves the central bank and monetary policy. The central bank creates money, which, for simplicity, we will call “cash.” Cash typically earns no interest. Our first key point is that, to the extent that monetary policy is conducted so as to keep inflation—and thus the (nominal) interest rate—inefficiently high, private agents have strong incentives to develop private payments systems to economize on the use of cash. Interest-bearing demand deposits (checking accounts) at banks are one example of such a private payments system. Because of the interest received in such accounts, households and firms will find it advantageous to switch from cash to these private deposits as their means of payment. Clearly, then, there would be private benefits to the introduction of payments systems like checking accounts.

But do these private benefits imply equivalent social benefits? If one household’s use of demand deposits imposed no costs on other households, the answer would be yes. But if use of such demand deposits does indeed impose costs on other households, the net social benefit of demand deposits will be lower and can, in fact, be negative. In the appendix, we present an example economy where these net social benefits from demand deposits are indeed negative, even though each household finds it in its interest to use them (since the private benefits are positive). In Chari and Phelan (2012b), we present a more general model where the net social benefits from interest-bearing means of payments can be either positive or negative, but are nevertheless always less than the private benefits.

The reason one household’s use of demand deposits imposes costs on other households is as follows: Introducing bank-provided payments leads to an expansion of the “means-of-payment” supply, now defined to include both cash and the amount of demand deposits. This higher means-of-payment supply leads to higher prices in the aggregate economy, which reduces the purchasing power of other households’ deposits and cash—but individual firms or households do not take this into account when they choose to use demand deposits over cash. This pecuniary externality (that is, an external cost imposed through prices rather than real resources) can cause households to use deposits instead of cash in cases where they wouldn’t, had they internalized this cost imposed on other households, and this externality implies that net social benefits of demand deposits are lower than net private benefits.

With net private benefits of banking exceeding net social benefits, it is clear that the banking system will be inefficiently large. In the model presented in the appendix, because the net social benefits are negative, not only is the banking system inefficiently large, the optimal size of banks is zero.

The model in the appendix is but a simple example, and the implications from it seem unrealistic. However, we would argue that recent developments in communication technologies and financial innovations may in fact make the model’s implications more than just a hypothetical scenario.

Historically, communication costs and limited development of financial markets have led to the use of systems in which only a fraction of a household’s financial wealth could be used for payments. With improvements in communication and financial markets, however, we can conceive of a world in which each individual can instantaneously access all of his or her financial wealth to make payments. We can also imagine a world in which settlement of transactions is instantaneous. In this world, cash becomes unnecessary, and precisely because cash is unnecessary, there is little or no need for payments systems that arise from the need to economize on cash, that is, arise because monetary policy is setting the inflation rate too high.

In the 1800s, it would have been inconceivable to pay for groceries, for example, by using a debit card associated with one’s mutual fund or stock portfolio (and in doing so, stocks were immediately sold, and the grocery received its settlement while the shopper was still at the counter). But today, this scenario is not far-fetched. In a world with virtually costless communication, banks as specialized providers of transactions services would simply be obsolete. These observations lead us to conclude that the actual importance of banks in the payments system is likely small today and will likely become even smaller in the near future. This is the third and final key point in our assessment of the transactions facilitation view.

What should “banks” look like, if not the traditional but fragile demand-deposit bank? As mentioned in the introduction, alternative financial institutions can provide similar services to the transactions facilitation services that traditional banks offer with far less potential for crises. One such example is the open-end mutual fund. These funds do not owe their shareholders a fixed dollar amount, but instead only the value of their percentage of the fund on the day the shareholder wishes to withdraw. If an unexpected surge of withdrawals occurs, the fund simply sells a sufficient quantity of the fund’s assets and gives the proceeds to the withdrawing shareholders. After this, the remaining shareholders still hold exactly the same assets per share as before. No shareholder gains by being earlier in line than other shareholders. Therefore, a belief that a run will occur cannot cause a run for a mutual fund—the self-fulfilling nature of runs that afflicts banks with demand deposits is thus avoided.

Implications for policy

Banks have been a durable part of the economic landscape for many centuries, and economic theory does explain why it might be efficient to set up institutions that fund long-term assets with short-term debt. Theory also illustrates that it might be optimal for private agents, but undesirable for society at large, to establish such institutions. These competing lessons from economic theory also provide guidance for regulation of such institutions.

As discussed in the earlier paper, both the
maturity transformation and the efficient monitoring views suggest that, given the costs imposed by crises and attendant bailouts, it may be desirable to allow financial institutions to issue short-term debt only if their assets do not have close publicly traded substitutes. Further, to minimize the incentive of governments to bail out institutions if a crisis occurs, such institutions should be separated from the payments system.

Any regulatory system must also take seriously the central role that banks have long played in the payments system. We have argued that this role may well be an artifact of a bygone era. Advances in information and communication technology make it feasible to access a wide array of assets, from stocks in public firms to portfolios of home equity loans, to undertake transactions. We have also argued that payments systems that require the use of demand deposits expose the economy to confidence crises and that it is possible to devise payments systems that do not require the use of debt-like claims, but instead use equity-like claims for transactions purposes.

These considerations suggest that the payments system should be both restricted and broadened. Transactions accounts should be restricted to institutions that mark the value of their assets to market continuously and that issue mutual-fund-like equity claims to owners. Such accounts should be broadened to institutions that are possibly very different from modern-day banks to include institutions such as stock and bond mutual fund companies.

We emphasize that the money market mutual fund as currently structured resembles a bank more than it does a mutual fund and therefore should not be allowed to issue transactions accounts. So, for example, Vanguard’s money market mutual fund (as currently structured) would no longer be allowed to serve as a transactions account, but Vanguard’s 500 Index Fund would.

The framework for regulatory policy implied by our analysis would lead to a banking system that is radically different from the one we currently have. Institutions that issue large amounts of short-term debt relative to their assets would be regulated and required to hold relatively little of their assets in publicly traded securities. The liabilities of such institutions would not serve as means of payment. The payments system would consist of institutions that issue equity claims.

Economic theory tells us that we do need banks. Theory also points us to constructive ways in which we can reform the financial system to make it more efficient and to ensure that crises that affect particular financial institutions do not spill over into the rest of the economy.

Endnotes

References

Chari, V. V., and Christopher Phelan. 2012a. What Assets Should Banks Be Allowed to Hold? Economic Policy Paper 12-3, (May) Federal Reserve Bank of Minneapolis.

Chari, V. V., and Christopher Phelan. 2012b. On the Social Usefulness of Fractional Reserve Banking. Working paper. University of Minnesota.

Fifty-eight years ago, Harberger (1954) estimated that the costs of monopoly, which resulted from misallocation of resources across industries, were trivial. Others showed that the same was true for tariffs. This research soon led to the consensus that monopoly costs are of little significance—a consensus that persists to this day.

This paper reports on a new literature that takes a different approach to the costs of monopoly. It examines the costs of monopoly and tariffs within industries. In particular, it examines the histories of industries in which a monopoly is destroyed (or tariffs greatly reduced) and the industry transitions quickly from monopoly to competition. If there are costs of monopoly and high tariffs within industries, it should be possible to see those costs whittled away as the monopoly is destroyed.

In contrast to the prevailing consensus, this new research has identified significant costs of monopoly. Monopoly (and high tariffs) is shown to significantly lower productivity within establishments. It also leads to misallocation within industries: Resources are transferred from high- to low-productivity establishments.

From these histories, a common theme (or theory) emerges as to why monopoly is costly. When a monopoly is created, “rents” are created. Conflict emerges among shareholders, managers and employees of the monopoly as they negotiate how to divide these rents. Mechanisms are set up to split the rents. These mechanisms are often means to reduce competition among members of the monopoly. Although the mechanisms divide rents, they also destroy them (by leading to low productivity and misallocation).

Introduction¹

In standard economic theory, monopoly leads to a welfare loss. This loss stems from a misallocation of resources across industries: Too few goods are produced by the monopolist; too many in other industries. Economic theory had long suggested that this welfare loss exacted high costs from the economy. But modern understanding took a turn when, in a landmark 1954 paper, Arnold Harberger analyzed the quantitative significance of monopoly costs in the United States. Were these costs as high as conventional economic theory suggested? The clear but surprising answer that Harberger provided was no.

Harberger estimated that, contrary to his expectation and to standard theory, the costs of monopoly were quite trivial. “We come to the conclusion that monopoly misallocations entail a welfare loss of no more than a thirteenth of a per cent of the national income. Or, in present values, no more than about $1.40 per capita,” he wrote. “I must confess that I was amazed at this result. ... Monopoly does not seem to affect aggregate welfare very seriously through its effect on resource allocation” (Harberger 1954, pp. 85, 86, 87).

Other economists extended Harberger’s work to estimate costs associated with tariffs, and here, too, the costs were trivial. A consensus quickly developed that Harberger’s conclusion was indeed valid.

Recently, a new literature has taken a different approach to understanding the costs of monopoly. Looking within industries, it examines the histories of industries in which a monopoly is destroyed and the industry transitions quickly from monopoly to competition, as well as the histories of industries that rapidly moved the opposite way, from competition to monopoly. If there are costs of monopoly, those costs should be whittled away as the monopoly is destroyed. Likewise, if an industry is monopolized, costs should be created. In both cases, costs should be apparent when comparing the industry before and after monopolization.

Several industries have been studied with this method, including transportation in the United States and U.S. manufacturing of sugar, iron ore and cement. The historical records of these disparate industries show that there are costs of monopoly and tariffs within industries. In these industries, this new literature has shown that monopoly led to, among other costs, the following:

1. Low productivity at each factory. That is, for any given amount of inputs, monopoly meant that less output was produced than under competition.
2. Misallocation of resources between high- and low-productivity factories. That is, monopoly led to resources (capital, labor, etc.) being transferred from productive factories to unproductive factories. Again, this misallocation occurs within an industry and is different from the misallocation that Harberger (1954) studied.

In sharp contrast to Harberger’s finding, these studies show that the welfare costs associated with monopoly and tariffs are not small. The consequence of cases (1) and (2) above is that industry output could have been produced with fewer inputs. One way to measure the loss, then, is to calculate the value of the “wasted” inputs. The histories of these industries show that as monopoly was destroyed in each, productivity at each factory soared. Doubling of productivities in a few years was common. The value of the wasted inputs was as much as 20 percent to 30 percent of industry value added.

A common theme (or theory) emerges from the histories as to why monopoly led to these costs. When a monopoly is created, “rents” are created.² Conflict emerges among shareholders, managers and employees of the monopoly as they negotiate how to divide these rents among themselves—or, more colloquially, how to “split the spoils.” Mechanisms are set up to split the rents. Although they divide rents, they also destroy them (by leading to low productivity and to misallocation).

As used in this paper, the term “monopoly” means more than the strict definition: an industry with a single producer. One industry mentioned later in the paper was a cartel for 40 years. Conflict over rents emerged between groups in the cartel, firms, workers and managers. In some industries, there were high tariffs (and other forms of protection). This high protection led to strong incentives among groups in the domestic industry to form monopolies. Firms attempted to collude, and workers formed industrywide unions (i.e., monopolies). So, the statement that “tariffs led to large welfare losses” means that tariffs led to incentives to form monopolies and then to actual monopolies, and these monopolies then led to large welfare losses.

A body of literature in the 1960s and 1970s argued that the costs of monopoly and tariffs were not trivial, saying (in essence) that there were costs within industries. This theoretical literature, and why it did little to dent the “Harberger consensus,” is briefly reviewed in Minneapolis Fed Staff Report 468 (Schmitz 2012), on which this policy paper is based. In this policy paper, I discuss historical studies that look at the collapse of monopoly. I describe how the monopolies emerged and how they were destroyed. Then I discuss the mechanisms that were used to split rents and why these mechanisms led to welfare losses.

Monopoly: Its Creation and Destruction

In this section, I introduce some of the industries that have been studied, discussing how monopolies were created in the industries and how they were destroyed. In the section that follows, I discuss the costs of these monopolies.

I discuss four industries: U.S. sugar manufacturing, in particular, sugar manufacturing using sugar beets (Bridgman, Qi and Schmitz 2009, 2012); the U.S. iron ore manufacturing industry (Galdón-Sánchez and Schmitz 2002 and Schmitz 2005); the U.S. cement manufacturing industry (Dunne, Klimek and Schmitz 2010); and U.S. freight transportation by water in the 19th century (Holmes and Schmitz 2001). These papers can be consulted for details that are only sketched here. Other industries will be briefly discussed.

When a monopoly is created, the government often has a hand in the process. This is the case in most of the industries studied, to greater or lesser degrees. In U.S. sugar manufacturing, the government played a central role in creating monopoly. During the Great Depression, sugar manufacturers were permitted, indeed encouraged, by U.S. law to form a cartel.

Many U.S. cartels were created during the Depression (as part of the New Deal), but the New Deal sugar cartel survived much longer than most. For 40 years, from 1934 to 1974, the industry was repeatedly able to renew the U.S. laws that enabled it to operate as a cartel. Soaring world sugar prices in 1974 resulted in the cartel losing political support, and the laws permitting it to operate as a cartel were not renewed.

To describe the government’s role in creating monopoly in the other industries, a useful approach is to first sketch a very simple model, one in the spirit of that in Holmes and Schmitz (1995). Consider an industry where transportation costs are large relative to production costs. If the domestic price is initially set equal to the cost of domestic production, then domestic producers will have a strong incentive to push their price up to the sum of foreign production cost plus the cost of transportation (or tariff) involved in bringing the foreign product to domestic markets.

The incentive to do so is great in this “industry” because, by assumption, transport costs are large relative to production costs. A very large tariff will be an incentive to increase prices, just as a large transportation cost would.

If the transportation or tariff cost is large, then, as in Holmes and Schmitz (1995), assume that groups will make investments to form monopolies. Firms will attempt to collude, and workers to form strong unions. Some groups may succeed. If later on protection is cut, the incentives to make these investments will fall, and the monopolies will weaken (or disappear).

This same logic applies if, rather than a transportation cost advantage, local firms have a production cost advantage.

This simple abstraction is a good representation of both the iron ore and cement manufacturing industries. In the early 1950s, U.S. producers had production cost advantages over foreign producers, and the industries received significant protection.³ Groups invested in creating monopolies. At various times, firms in these industries were charged with trying to collude. The U.S. government investigated the industries for antitrust violations. It is unnecessary to enter into the argument as to how effective collusion was; there is little doubt that very strong, industrywide unions emerged in these industries. Although antitrust laws in the United States made firm collusion difficult, building monopoly unions was easier. Collective bargaining laws enacted by the U.S. government allowed unions to organize all workers in an industry and not be bound by antitrust laws (see Meltzer 1963 and Winter 1963).

The monopolies in these industries—in particular, the strong monopoly unions—lasted for many decades in the post–World War II period. The monopoly unions were able to provide very high wages. For example, by the 1970s, cement workers were paid as much as U.S. autoworkers (who were the highest-paid manufacturing workers). The unions also had very stringent work rules (as described later on).

In the 1980s, the monopolies in these industries weakened or were dissolved. The union in the cement industry dissolved. In the iron ore industry, the union did not disappear, but lost much of its clout. For example, work rules became much less stringent, and plant managers had more control over how to structure plant operations.

Why the weakening of the monopolies in the 1980s? Foreign producers were now threatening to enter local markets. Brazil offered to sell iron ore in Chicago and Cleveland, the heart of the U.S. market, at half the local price. Firms around the world offered to sell cement on the West Coast and Gulf of Mexico at half the U.S. prices.

How could foreign firms offer to sell at such discounts? There are two proximate reasons. First, transportation costs greatly decreased (relative to production costs) in the postwar period. This, by itself, would have meant a weakened incentive for continued investment in keeping monopoly. But, second, the production cost advantage of U.S. producers decreased. This development was, of course, to be expected, as the whole purpose of creating strong unions was to increase wages (and hence costs). The monopolies also led to lower productivity, increasing costs further. But what was striking is that U.S. producers were at a production cost disadvantage.

An obvious question is: Why did the unions (and other groups discussed later on) push wages so high and lower productivity to the point where foreign producers could offer such steep discounts? At least three possibilities come to mind. First, the groups realized that wage demands and work rules would lead to the demise of monopoly, but that this strategy was the best. Second, the groups realized that wage demands and work rules would lead to the possibility of foreign entry, but they expected more government protection than they were able to receive. Many calls for protection were made, and some protection was given, but it was not enough. Third, perhaps the outcome (foreign entry) was not expected. Although I do not know which story best describes the events, the story itself is not important for the issue at hand. The main point is that there are significant costs of monopoly and tariffs.

Monopolies can arise on their own, of course, without the help of government policy. Consider U.S. freight transportation by water in the 19th century. Well before any collective bargaining laws were enacted in the United States, strong unions developed in the port of New Orleans. Groups in transportation had a strong incentive to form monopolies in the port, since much freight went through that port. Many groups, such as warehouse owners, riverboat pilots and longshoremen, were thought to have formed strong monopolies. Evidence of strong monopolies is particularly clear for longshoremen.

The weakening of these monopolies in New Orleans resulted from the development of alternative transportation technologies. New technology—railroads—meant that the returns from these port monopolies were greatly diminished. Investments in sustaining the monopolies waned, and the monopolies were considerably weakened.

Monopoly: Splitting the Spoils (and Destroying Them as Well)

During the period when monopolies in these industries were strong, groups set up mechanisms to split rents. Here I discuss some of the mechanisms used and how they led to the destruction of rents—in particular, to low productivity and misallocation. When monopoly was weakened in these industries, the mechanisms were abandoned, leading to large productivity gains in establishments and to resources being reallocated from low- to high-productivity producers.

Mechanisms Limiting Competition
One mechanism used to split rents was competition-reducing rules. Here I discuss two types that were used: quotas and work rules.

Quotas
In the U.S. sugar industry, the New Deal cartel included factory owners, factory workers, farmers, farm workers and others. As the cartel was established, each of these groups sought to secure (for themselves) as large a share of rents as possible. A major mechanism to split rents was quotas. In the cartel, firms were given quotas—the right to sell a certain amount of sugar each year.

Incumbent farmers also sought, and were successful in acquiring, quotas—the right to grow sugar beet crops on a given number of acres each year. Without these quotas for incumbent farmers, nothing stopped firms from moving the locations of their factories or even using different farmers in the same location. Just as firms in the cartel used firm quotas to limit competition, incumbent farmers wanted quotas to limit competition among themselves (and from other farmers). Without these quotas, there was no way to ensure that incumbent farmers would receive a share of the monopoly profits.

As is often the case, these quota rights (both those of the firms and those of the farmers) could not be sold.⁴ Although the allocation of quotas for acres in 1934 was “efficient,” over time there was a change in the comparative advantage of locations in manufacturing sugar. Hence, there emerged a significant misallocation of resources between factories, with low-productivity factories producing too much sugar and high-productivity factories too little.

As the cartel started in 1934, some of the most profitable or productive (measured as revenue per acre divided by costs per acre) areas to make sugar were in California and Colorado. But after a few decades, these areas were no longer high-productivity areas. The opportunity cost of land in California and Colorado, and of the water used in making sugar in these areas, grew much faster than in other parts of the country, in particular, in Minnesota and North Dakota. By the 1960s, these latter states became the most productive areas in which to make sugar. However, given the mechanisms to split rents (i.e., the quotas), the industry could not increase production in these areas. Once the cartel ended in 1974 and the mechanisms to split rents were abandoned, the share of industry production in Minnesota and North Dakota grew rapidly (and declined rapidly in California and Colorado).

I can estimate the magnitude of the welfare loss due to these mechanisms to split rents (the quotas), that is, from the misallocation of resources within the industry. Recall the introduction to this paper, which mentioned that one way to measure welfare loss is to calculate the value of wasted inputs in producing industry output. In the 1960s, the industry was using land in California, which had high value (or opportunity cost), rather than land in North Dakota, which had much lower opportunity cost. The difference in opportunity costs is a measure of the wasted land input. Not only were the opportunity costs of land much lower in North Dakota, but the opportunity costs of many other inputs—for example, labor and the water used in growing sugar beets—were much lower as well. To calculate the welfare loss in, say, 1965, I imagine “moving” some of the quota allocation from California to North Dakota (keeping industry output fixed) and then calculating the value of the inputs that were wasted by producing in California. At this point, it is easier to estimate the value of the wasted inputs relative to industry profits (rather than relative to value added). The estimates indicate that the losses were roughly 20 percent to 30 percent of industry profits.

Work Rules
In the iron ore and cement industries, those who were in a position to gain from the large transportation costs into local markets, and the protection offered by tariffs, were the factory owners, factory workers and even the local governments (e.g., townships) where factories were located. What mechanisms were used to acquire rents? Local townships placed significant taxes on the production of iron ore and cement. Workers formed very strong unions. Although claims of collusion within both industries have been made, these claims are harder to document than the taxes and union contracts that emerged in these industries.

A major mechanism to split rents was the work rules in union contracts. Among other things, work rules were a way to limit competition among workers. They were structured so that managers could not play workers off each other. Let me briefly discuss these features of work rules and their consequences.

Union contracts split the tasks in plants into groups or categories. Workers were then assigned to one of these groups or categories, that is, given the right to complete tasks in that category. Only the workers in this group could complete the tasks assigned to the group. Very often these distinctions among workers were arbitrary in that a worker in a particular category was able, but not allowed, to complete tasks in many other categories.

Consider an important example. Machine operators were given tasks and repair workers other tasks. Machine operators were prohibited from assisting repair workers in their assigned tasks, even mundane tasks that required no repair expertise, such as getting supplies, holding tools and so on.

In addition to a sharp distinction being made between the tasks of machine operators and the tasks of repair workers, there was also a sharp distinction among tasks assigned to repair workers. Repair workers were grouped into many classifications, as many as 30 in a plant.

These types of work rules dividing work among members of the union are most often called job classification systems. They are similar to the quotas discussed earlier. In particular, work rules are a way to limit competition between workers, just as quotas limited competition between farmers. They ensure that groups of workers receive a share of the monopoly profits. But they also destroy profit, as I now discuss.

What are the negative consequences of such rules? These work rules in the iron ore and cement industries lead not only to overstaffing, but also to idle machinery. When, for example, the Finish Grind Department is down, workers from other departments are not allowed to help restore its machines to operation. Hence, machines are down longer than necessary, and capital productivity suffers. But clearly, energy productivity suffers as well. Fuel is being burned in other parts of the plant, and electricity used, even as the disabled machines are idle and output is not produced. As a result, such rules lead to low total factor productivity.

What is the quantitative significance of work rules? In the 1980s, when the work rules in the iron ore and cement industries were made much less stringent, labor productivity doubled in a few years. Other productivities increased as well. If these increases in productivity can be tied in large part to the relaxing of work rules, then obviously these are big welfare gains. The iron ore and cement manufacturing papers cited earlier argue that most of the productivity gains were due to the relaxing of work rules. Proving such a claim is difficult, but the papers, by looking at both direct and indirect evidence, have marshaled much evidence that this was indeed true.

We can estimate the magnitude of this welfare loss due to these mechanisms to split rents (the work rules), that is, from the low productivity in establishments. Again, one way to measure the welfare loss is by the value of the wasted inputs. With these work rules, machines were down longer than necessary. The energy that was being consumed elsewhere in the plant when output was not produced was a wasted input. The value of this wasted energy was its opportunity cost per unit multiplied by its quantity. The opportunity cost was its price per unit.

Next consider labor. With these work rules, labor input was wasted. For example, a machine operator could not hold a tool for a repair person (who would need to bring in another repair person for such tasks). The value of this wasted input was the opportunity cost of the machine worker’s time multiplied by the amount of time involved.

How much time was wasted in these plants? And what was the opportunity cost of this time? When work rules were changed in these industries, labor productivity doubled in a few years. Half of the workers were able to produce the same output. A rough estimate then suggests a dead-weight-loss-to-industry-value-added ratio of 16 percent to 17 percent. (See Schmitz 2012, pp. 14-15, for details of this calculation.)

In addition to energy and labor, capital was also wasted, as work rules meant that disabled machinery took longer to repair than was necessary. In considering estimating the welfare loss due to wasted energy and wasted capital, I note two things. First, energy productivity and capital productivity both increased significantly with the loosening of work rules, but not to the extent that labor productivity increased. So, not as much was wasted. Second, the price paid for these inputs was likely close to its opportunity cost. Using a dead-weight loss for the wasted capital and energy of a few percentage points (possibly more) of value added, together with the wasted labor estimate of 16 percent to 17 percent of value added, gives an estimate of over 20 percent in total.⁵

As just discussed, work rules in iron ore and cement manufacturing led to low productivity in establishments. But work rules can also lead to misallocation within industry. I will go on to briefly discuss some other industries, arguing that work rules likely led to the same type of misallocation as quotas did in sugar—with low-productivity plants producing too much output and high-productivity plants too little.

The historical studies have shown that monopoly and tariffs can lead to welfare losses within industry on the order of 20 percent. But what about the losses for the entire economy? How many other industries, and of what size, have incurred such losses because of monopoly? I briefly discuss these questions below.

Side Payments Between Groups
Rules to reduce competition (such as quotas and work rules) were an indirect means to split rents between groups. Direct means were also used, whereby some factories would send money to other factories. This was done, for example, in the sugar cartel. These side payments in sugar manufacturing were not lump sum, but involved mechanisms that led to distortions. In particular, the side payments exacerbated the misallocation problem discussed earlier (of having production in California and not North Dakota).

Splitting the Spoils: Other Industries, Other Countries and a U.S. Cost Estimate

In this section, I argue that the mechanisms to split rents just discussed are prevalent throughout industry. Moreover, evidence suggests that these mechanisms have had negative consequences in other industries (and other countries) as well. However, I cannot be sure of their quantitative significance because no studies like those described in the preceding section have been completed for these industries.

Many U.S. industries had significant market power after World War II, first by virtue of the devastation that many countries faced as a result of the war and later because of government protection of U.S. manufacturing. Monopolies emerged; in particular, the postwar years saw the emergence of industrywide unions in the auto, steel, paper, tire, airplane and chemical industries, to name a few.

What mechanisms were used to split rents? The job classification systems discussed earlier are prevalent throughout manufacturing (though for the most part are less stringent today than a few decades ago). Some observers of these industries hold the view that work rules led to low productivity in plants.⁶

Just as they did in the cement and iron ore industries, stringent work rules likely led to low productivity in establishments in many manufacturing industries. In some, stringent work rules led to other types of distortions and losses (which were not seen in the cement and iron ore industries). First, as I suggested earlier, work rules in these industries likely led to misallocation—resources being transferred from high- to low-productivity plants. Second, high wages (and stringent work rules) have likely led to another type of misallocation in industries: a change of technology (in order to escape the wages and work rules).

A similar phenomenon—that is, monopolists splitting (and destroying) rents—occurs in other countries. In Britain, job classification systems (referred to as “job demarcation rules”) are widespread. Demarcation rules are also used in France. In both countries, research suggests that these rules lead to reduced productivity.

I finish this section with a back-of-the-envelope estimate for the within-industry costs of monopoly and tariffs for the United States. This will enable a preliminary stab at the question, are these welfare losses similar in magnitude to Harberger’s losses (0.1 percent of value added), or can I conclude that they may well be significantly larger?

Industries that are known to have strong unions and rigid work rules include mining, utilities, construction, transportation (in particular, airlines and railroads) and parts of manufacturing, in particular, durable manufacturing (steel, airplanes, autos). Assume that work rules had similar negative impacts on productivity in those industries as they did on the industries discussed in detail earlier—again, about 20 percent of industry value added.

Adding together the total value added of these industries thus affected (just over 25 percent of total gross domestic product in 1977) enables an estimate of welfare losses from monopolies and tariffs of roughly 5 percent of GDP (=20 percent loss of 25 percent GDP share). (Further calculation details are in Schmitz 2012.) Again, this calculation is obviously extremely crude, but it does suggest that the losses may well be orders of magnitude larger than Harberger’s estimated losses.⁷  

Costs of Monopoly: Summary and Observations

Research on the theoretical and quantitative significance of monopoly costs has evolved considerably since the mid-1950s, when Harberger’s (1954) influential paper suggested—in contrast to the prevailing view among economists—that in the United States, the costs of monopoly resulting from resource misallocation across industries were actually quite insignificant. This view soon became the dominant consensus among economists. Subsequent research in the 1960s and 1970s sought to establish a convincing counterargument, but was unsuccessful in overturning the prevailing concept of negligible costs.

This paper reviews a new stream of research that uses a different approach to analyzing the costs of monopoly. It examines the costs of monopoly and tariffs within industries rather than across them. In particular, it examines the histories of industries in which a monopoly is destroyed (or tariffs greatly reduced) and the industry transitions quickly from monopoly to competition.

Over considerable time spans and a wide range of industries, this research finds that monopoly exacts high costs in two ways: (1) through misallocation of economic resources between high- and low-productivity factories and (2) by decreased productivity at each factory. The historical studies call the Harberger consensus into question. At least in the industries studied thus far, monopoly and tariffs have led to significant welfare losses, on the order of 20 percent of value added.

A common thread runs through these histories, one that suggests a theory. When a monopoly is created, rents are generated. But the distribution of these rents—splitting the spoils—causes conflict among shareholders, managers and employees of the monopoly. These parties devise mechanisms to split the spoils, but the mechanisms often lead, paradoxically, to the destruction of rents.

The implications of this theory of monopoly costs, and of the empirical findings of high costs, deserve serious consideration by policymakers as they design and enforce antitrust measures. As described earlier in this paper, government policies themselves, such as tariffs and other forms of protection, are an important source of monopoly. This review of recent economic research indicates that the costs of such protectionist policies are considerable and should be fully recognized and appreciated. Furthermore, policy reforms to minimize these costs should be carefully considered.

As for future economic research, a key question is to understand why mechanisms (such as work rules) are used to split rents when they also self-destructively wipe out rents. Why can’t members of the monopoly structure contracts that avoid such large wasted resources? Differences in information? The inability of parties to commit to future actions? Such reasons may well be why mechanisms intended to split rents also destroy them.

Endnotes

1 The author thanks Tasso Adamopoulos, John Asker, Doug Clement, Tom Holmes, Pete Klenow, Sam Kortum, David Lagakos, Justin Pierce, Todd Schoellman and Arilton Teixeira. The views expressed herein are those of the author and not necessarily those of the Federal Reserve Bank of Minneapolis or the Federal Reserve System.

2 In this usage, “rent” is the difference between what a factor of production is actually paid and what it would need to be paid to remain in use; as such, it is a measure of that factor’s monopoly power.

3 Transport costs were large (relative to production costs) in both industries. Also, the iron ore industry received government protection because the U.S. steel industry did. In the cement industry, the U.S. government at various times ruled that foreign producers were dumping in local markets.

4 Why the limit on selling quotas? When the cartel was being proposed, there were, obviously, complaints from farmers in sugar beet and sugar cane areas who did not grow these crops before 1934 (and hence were being left out of the cartel). A reasonable conclusion is that the cartel limited the quota rights so as to limit these complaints (thereby increasing the likelihood that the cartel would be acceptable).

5 In the calculation of the welfare loss due to the wasted labor input, I made no imputation for the value of leisure that might have been enjoyed at the plants (under the work rule regime). For example, perhaps the machine operator prefers standing next to the repair person not holding tools rather than helping out and holding tools. But this brings up other issues: What happens in a general equilibrium model if a large fraction of the population is standing around at work? We are looking at these issues elsewhere.

6 See, for example, Hoerr (1988), who discusses the role of work rules in the U.S. steel industry collapse of the 1980s, and Simberg (2008), who discusses the recent auto industry crisis, lamenting the focus on high wages and not, in his view, the real culprit: work rules and low productivity.

7 Although work rules in the industries discussed earlier have weakened considerably since the 1970s, in other areas, such as education, stringent work rules have grown. Moreover, occupational licensing has grown dramatically in importance, and its impacts on welfare may be important (see Kleiner and Krueger, forthcoming).

References

Bridgman, Benjamin, Shi Qi, and James A. Schmitz, Jr. 2009. “The Economic Performance of Cartels: Evidence from the New Deal U.S. Sugar Manufacturing Cartel, 1934–74.” Research Department Staff Report 437. Federal Reserve Bank of Minneapolis.

Bridgman, Benjamin, Shi Qi, and James A. Schmitz, Jr. 2012. “A Cost of Monopoly: Misallocation of Resources from High to Low Productivity Plants.” Manuscript.

Dunne, Timothy, Shawn Klimek, and James A. Schmitz, Jr. 2010. “Does Competition Spur Productivity? Evidence from the Post WWII U.S. Cement Industry.” Working Paper.

Galdón-Sánchez, José E., and James A. Schmitz, Jr. 2002. “Competitive Pressure and Labor Productivity: World Iron-Ore Markets in the 1980s.” American Economic Review 92(4): 1222–35.

Harberger, Arnold C. 1954. “Monopoly and Resource Allocation.” American Economic Review 44(2): 77–87.

Hoerr, John P. 1988. And the Wolf Finally Came: The Decline of the American Steel Industry. Pittsburgh: University of Pittsburgh Press.

Holmes, Thomas J., and James A. Schmitz, Jr. 1995. “Resistance to New Technology and Trade Between Areas.” Federal Reserve Bank of Minneapolis Quarterly Review 19(1): 2–17.

Holmes, Thomas J., and James A. Schmitz, Jr. 2001. “Competition at Work: Railroads vs. Monopoly in the U.S. Shipping Industry.” Federal Reserve Bank of Minneapolis Quarterly Review 25(2): 3–29.

Kleiner, Morris M., and Alan B. Krueger. Forthcoming. “Analyzing the Extent and Influence of Occupational Licensing on the Labor Market.” Journal of Labor Economics.

Meltzer, Bernard D. 1963. “Labor Unions, Collective Bargaining, and the Antitrust Laws.” Journal of Law and Economics 6(Oct.): 152–223.

Schmitz, James A., Jr. 2005. “What Determines Productivity? Lessons from the Dramatic Recovery of the U.S. and Canadian Iron Ore Industries Following Their Early 1980s Crisis.” Journal of Political Economy 113(3): 582–625.

Schmitz, James A., Jr. 2012. “New and Larger Costs of Monopoly and Tariffs.” Research Department Staff Report 468. Federal Reserve Bank of Minneapolis.

Simberg, Rand. 2008. “Detroit’s Downturn: It’s the Productivity, Stupid.” PJ Media.

Winter, Ralph K., Jr. 1963. “Collective Bargaining and Competition: The Application of Antitrust Standards to Union Activities.” Faculty Scholarship Series. Paper 2176.

Beginning in late 2009, the Greek government had difficulties selling its bonds to private investors, who demanded high interest rates. In May 2010, the European Union (EU) and the International Monetary Fund (IMF) approved a 110 billion euro loan package to the Greek government in return for promises of spending cuts to sharply reduce the Greek public deficit. The plan, negotiated by German Chancellor Angela Merkel and Greek Prime Minister George Papandreou, was intended to cover the borrowing needs of the Greek government through 2013. In spite of this rescue package and another, 130 billion euro, package put together between July 2011 and March 2012, the debt crisis in Greece continues into 2012.

Ireland and Portugal have required similar EU-IMF rescue packages. Cyprus, Italy and Spain have had difficulties selling their bonds. Similar difficulties threaten other members of the European Economic and Monetary Union (EMU)—the countries in the EU that use the euro as their currency, also referred to as the eurozone—like Belgium and France.

In fact, as of April 2012, of the 17 members of the eurozone, only four—Finland, Germany, Luxembourg and the Netherlands—have long-term government bonds with the highest Standard & Poor’s rating, AAA, while the bonds of five countries—Cyprus, Ireland, Italy, Portugal and Spain—have junk ratings, BBB+ or lower. Greek bonds were given the lowest possible rating, CCC, in July 2011, and are currently not rated, but are listed as SD, meaning that the Greek government has selectively defaulted on some issues.

The countries that have suffered debt crises, or are threatened by such crises, got into trouble in different ways. The two crucial common characteristics are that each of these countries is currently experiencing a deep and prolonged recession and each needs to frequently sell large quantities of bonds, either to finance large fiscal deficits or to roll over—and make interest payments on—a large public debt.

We sketch out a theory for analyzing the European sovereign debt crises based on the research of Harold Cole and Timothy Kehoe (1996, 2000) and Juan Carlos Conesa and Kehoe (2012). In this theory, the need to frequently sell large quantities of bonds leaves a country vulnerable to a financial crisis. This vulnerability gives the government the incentive to pay down its debt to a level where such a crisis is not possible. In the event of a deep and prolonged recession, however, the government has a conflicting incentive to “gamble for redemption”—to borrow to smooth government spending, to reduce the debt if the economy recovers and, possibly, to default if the recession continues for too long.

Using this theory, we analyze the various rescue packages and policy interventions made by the EU and the IMF. Policies that result in high interest rates on government bonds and high costs of default provide incentives for a government to reduce its debt. Policies that result in low interest rates and low costs of default provide incentives for a government to gamble for redemption. We conclude that, up until now, policy interventions by the EU and the IMF have encouraged eurozone governments to gamble for redemption. In the theory we present, a government that gambles for redemption is following a policy that is optimal for the citizens of its country. The policy goals of the EU and the IMF may be different from those of the government of an individual country, however, and, to the extent that the EU and the IMF want the government to reduce its debt to avoid a crisis to preserve the stability of the EU, they should adopt policies to discourage the government from gambling for redemption.

Timeline and some data

The Treaty on European Union—signed in Maastricht, Netherlands, on Feb. 7, 1992, and commonly referred to as the Maastricht Treaty—converted the European Community, which then had 12 members, into the European Union. The treaty established four “convergence criteria” as prerequisites for membership in the EMU. One criterion required a country to have an annual public deficit no greater than 3 percent of GDP and a public debt no greater than 60 percent of GDP. Another criterion required the country to participate in the European Exchange Rate Mechanism (ERM)—set up as a voluntary program in 1979—to maintain its exchange rate in a very narrow band around the European Currency Unit (ECU), which eventually became the euro. The other two criteria imposed restrictions on inflation rates and interest rates.

In the process of ratifying the Maastricht Treaty, Denmark and the United Kingdom obtained opt-out clauses from joining the monetary union. All 15 countries that have joined the EU since 1992 were required to join the monetary union. The ERM suffered a major crisis 1992, with a number of countries forced to drop out, and—when the crisis threatened more countries in 1993—the exchange rates bands were widened considerably. The mechanism was restarted in 1999 and is now referred to as ERM II.

Sweden, which joined the EU in 1995, has managed to exploit a legal loophole to avoid adopting the euro: Its accession treaty required Sweden to join the monetary union after meeting the convergence criteria and participating in the ERM II for two years, but it did not explicitly require Sweden to join the ERM II, and it has not done so. The other seven countries in the EU that are not yet in the eurozone are required to go through the process of participating in ERM II and eventually joining the eurozone.

The accompanying table provides a timeline for the major events related to the sovereign debt crises that are ongoing in the eurozone.

Timeline of Events Associated with Eurozone Debt Crises, 2010–2012

European leaders had seen the need to coordinate fiscal policy in a monetary union. In 1997, at the insistence of Germany, they adopted the Stability and Growth Pact (SGP), which imposed financial penalties on countries that violated the convergence criterion that the public deficit not exceed 3 percent of GDP. Nonetheless, when the French and German governments announced that they had violated this deficit limit in 2003, they were not penalized, reducing the credibility of the SGP.

The details differ on how various countries became vulnerable to sovereign debt crises: The Greek government borrowed heavily during Greece’s rapid growth early in the 2000s and employed fraudulent accounting practices to mask its large violations of the 3-percent-deficit rule. The Irish government was in very good fiscal shape until it guaranteed the debts of the six major private banks based in Ireland in 2008. When the housing market collapsed in late 2008 and 2009, leading to the collapse of these banks, the Irish public deficit and debt soared. Portugal took advantage of the low interest rates at which it could borrow after joining the eurozone to build up a large public debt. When Spain was hit by recession in 2008, the government reacted with massive Keynesian stimulus policies, running large public deficits. Although the Spanish public debt started from a very low level, it increased rapidly because of these deficits.

The problems faced by the Italian government stem from conditions very different from those in Spain. Although Italy had low public deficits, it had started with a very high level of debt. Indeed, Belgium and Italy had entered the EMU with public deficits that substantially exceeded the 60-percent-debt rule only because they were deemed to be on a “satisfactory pace” to reducing the public debt to 60 percent of GDP. Yet, in the case of Italy, much of this satisfactory pace was illusory, the product of creative accounting of exactly what constituted government expenditures and government revenues. France employed similar sorts of creative accounting.²

In spite of these differences in initial conditions, Greece, Ireland, Italy, Portugal and Spain (GIIPS) share two crucial characteristics: First, as the data in Figure 1 show, the recoveries from the 2008–2009 recessions in these countries have been nonexistent. Notice that, in Figure 1, the German economy has started to recover in 2010 and 2011, if only weakly, while the GIIPS are still mired in recession. Second, as the data in Figure 2 show, the GIIPS have large borrowing requirements because of high deficits or large debts or both.

Large Image

Large Image

Self-fulfilling debt crises

The need to frequently sell large quantities of bonds leaves the countries vulnerable to self-fulfilling debt crises of the sort analyzed by Cole and Kehoe (1996, 2000) and Conesa and Kehoe (2012). In such a crisis, if investors expect a government to have trouble repaying its debt, they pay a low price at auctions of new government bonds. The resulting low value of the new bond sales makes it difficult for the government to repay the old bonds becoming due, thus justifying the expectation of a crisis. If, however, investors do not expect the government to have trouble repaying its debt, they are willing to pay a high price for new bonds. This expectation too is self-fulfilling.

To understand the reasoning in the model, we start by examining two crucial relations: the government budget constraint—which relates sales of new bonds and payments on old bonds to government expenditures and tax receipts—and the relation between the price that investors pay for bonds and the probability of a sovereign default. We then explain how the government determines its optimal policy and how financial crises can occur.

Government budget constraint
In every time period—which can be a year or longer or shorter, depending on the application—the government’s budget constraint requires that the total of resources collected in the form of tax revenues and receipts from new bond sales be equal to the total of resources spent in the form of government expenditures and transfer payments and coupon payments on outstanding bonds and payments of face value on bonds reaching maturity. To keep things simple, we lump transfer payments into expenditures and coupon payments on outstanding bonds into payments of face value on bonds reaching maturity. We can write the government’s budget constraint as

tax revenues + receipts from new bond sales
= expenditures + payments on bonds becoming due.

Notice that this constraint does not require the government to balance its budget in the traditional sense because it does not require tax revenues to equal expenditures. The difference between expenditures and tax revenues is referred to as the government’s primary deficit. We can rewrite the government’s budget constraint as

primary deficit = expenditures − tax revenues
= receipts from new bond sales − payments on bonds becoming due.

To analyze the evolution of the sovereign debt crises in the eurozone, however, it is more useful to write the government’s budget constraint in a different way as

net borrowing = face value of new bond sales − payments on bonds becoming due
= primary deficit + face value of new bond sales − receipts from new bond sales.

The payments on bonds becoming due are equal to their face value unless the government defaults. Consequently, unless the government defaults, the face value of new bond sales minus the payments on bonds becoming due is equal to the net government borrowing depicted in Figure 2. Because investors discount the future, they pay a price for the bond less than the face value of the bond when it matures. If investors pay a very low price for new bonds, then receipts from new bond sales fall far short of the face value of the bonds, and net borrowing is substantially greater than the primary deficit. We focus on net borrowing because it measures the increase in government debt.

Pricing bonds
Consider an investor whose holdings of the sovereign bonds of a particular country constitute a small fraction of his or her total portfolio. At a bond auction, we assume that this investor is willing to bid a price equal to the expected present discounted value of the return on the bond.³

First consider the case where the investor is sure that the government will not default. Suppose that the government promises to pay back one euro in one year and that the consumer discounts future returns at 2 percent per year. Then the price of the bond is the present discounted value of one euro next year, which is a little more than 98 euro cents,

1/1.02 = 0.98039.

This number is also referred to as the discount factor. The investor multiplies any promised return in one year by the discount factor to obtain its present discounted value. For example, the investor is willing to pay 980.39 euros for a bond paying 1000 euros next year. Another way to think about this is that, if 980.39 euros are invested at 2 percent, then after one year they are worth 1000 euros. For a bond with a longer maturity, the investor multiplies the discount factor by itself N times, where N is the number of years of maturity. For a bond paying 1000 euros in 10 years, for example, the investor would be willing to pay 820.35 euros,

820.35 = (0.98039)¹⁰ x 1000.

Now consider the case where there is a positive probability of default. The difference between the return if there is no default and the return if there is default is referred to as the haircut. Suppose that the probability of a default is 50 percent and that the haircut is 50 percent, which implies that the payoff if there is default is 50 percent. Then the expected return on the bond with a face value of 1000 euros is 750 euros,

expected return = probability of no default × payment if no default
+ probability of default × payment if default
750 = 0.5 x 1000 + 0.5 x 500.

The price of the bond is the expected present discounted value of the return, which in this case is 735.29 euros,

price = discount factor × expected return
735.29 = 0.98039 x 750.

The yield—which is the implicit interest rate on the bond—is 36 percent,

 0.36 = (1 / 0.73529) - 1

because, if 735.29 euros are invested at 36 percent, after one year they are worth 1000 euros. The spread—the difference between the yield and that on a safe bond—is 34 percentage points.

Our example illustrates how an increase in the probability that investors assign to a default results in an increase in the yield on the bond. Figure 3 depicts the yields on government bonds in the GIIPS compared with those in Germany. Notice how spreads on GIIPS bonds increased starting in 2008.

Large Image

An attractive feature of our numerical example is that it provides a rationalization for the yields in Figure 3. In the fourth quarter of 2011, for example, German bonds, which can be thought of as a safe investment, had a yield of about 2 percent, while Greek bonds had a yield of about 36 percent, which is the yield if the probability of default is 50 percent and the expected haircut is 50 percent. Of course, there are other combinations of probabilities of default and haircuts that result in this same 36 percent yield.

Optimal government policy and crises
In every time period in the model, the government must decide how much new debt to sell and whether or not to default. We assume that the government is benevolent, in that it values the welfare of consumers, that is, the citizens of the country, who value both private consumption and government expenditures. We also assume that consumers—and consequently the government—value smooth paths of private consumption and government expenditures. Sharp cuts in government expenditures are particularly painful. Defaults are also costly in that they disrupt financial markets, which causes a drop in the GDP of, say, 5 percent—which we refer to as the default penalty—that is available for government expenditures, private consumption and repayment of debt. These assumptions are intuitively appealing and fairly innocuous.

We make a number of other assumptions that are more restrictive to keep the analysis simple. We assume, for example, that tax revenues are a constant fraction of GDP because tax rates are fixed. We also assume that the default penalty is permanent and that, if the government defaults, it is permanently excluded from borrowing. Cole and Kehoe (1996, 2000) model consumers within a country as making private investment decisions, but here—again to keep things simple—we follow Conesa and Kehoe (2012) in having consumers consume all after-tax GDP rather than investing some of it. These assumptions can be relaxed without changing the qualitative results of the model. How much quantitative results change depends on the parameterization, of course, and this is a topic that deserves future research.

A financial crisis is self-fulfilling if the expectation that the government will default causes it to default in a situation where it would otherwise pay for the bonds becoming due. For low levels of debt, self-fulfilling crises are not possible. For higher levels of debt—those above a threshold that we call the upper safe debt limit—self-fulfilling crises are possible. For even higher levels of debt—those above a threshold that we call the upper sustainable debt limit—the government prefers to default rather than pay for the bonds becoming due.

The timing within a period is such that investors decide what price to bid in the auction for new government bonds before the government decides whether or not to default on the old bonds becoming due. Suppose that, before the auction, investors receive some sort of bad news that makes them expect the government to default this period. Under what conditions will this expectation be self-fulfilling? The investors expect that the government will be in default the subsequent period because it is excluded from financial markets. The price that the investors offer for new bonds is the present discounted expected payment in the case of default, which is low or zero. The government can either default or pay for the bonds becoming due. For levels of debt equal to or below the upper safe debt limit, the government prefers to pay for the bonds becoming due and suffer the drop in government expenditures but avoid paying the cost of defaulting. For these low levels of debt, investors will pay a high price, equal to the present discounted face value for new bonds, no matter what the news is. If, however, debt is above the upper safe debt limit, a self-fulfilling crisis occurs if there is bad news. For high levels of debt, those above the upper sustainable debt limit, the government chooses to default even if investors buy the new bonds offered.

The probability that investors assign to receiving bad news in a period is arbitrary.⁴ At the beginning of a period, the bad news arrives or it does not. Notice that, in the bond auction in a period, if bad news had not arrived early in the period, then the bond price depends on the probability of receiving bad news in the next period.

Cole and Kehoe (1996, 2000) call the interval of debt levels above the upper safe limit but equal to or below the upper sustainable limit the crisis zone. If debt is in this zone, a self-fulfilling crisis can randomly occur. Since interest rates are high when the debt being sold is in the crisis zone and the probability of a costly default is positive, a government will optimally choose to run surpluses to run its debt down to the upper safe limit. Once debt reaches the upper safe limit, interest rates drop and the probability of default disappears. Since sharp cuts to government expenditures are painful, however, the government may choose to pay down the debt over a number of periods.

In a quantitative model calibrated to match features of European data, Conesa and Kehoe (2012) show that the upper safe limit is about 120 percent of GDP while the upper sustainable limit is about 210 percent of GDP. These numbers make sense in terms of the numbers currently used by policymakers in Europe, in particular, the need to reduce Greek debt below 120 percent of GDP to eliminate the possibilities of future crisis.⁵

Gambling for redemption

As we have just argued, financial crises and defaults on sovereign debt are costly for a country, and the government of a country that finds itself vulnerable to a self-fulfilling crisis has the incentive to pay down its public debt so that it does not need to frequently sell large quantities of bonds. As Conesa and Kehoe (2012) point out, however, countries that are in deep recessions have an opposite incentive: to cut government spending very slowly and increase the public debt, gambling that a recovery in the economy will lead to a recovery in tax revenues, at which point it can stop increasing the debt. If the country is unlucky and the recession is prolonged, however, the country can find itself more vulnerable to a self-fulfilling debt crisis and ultimately may be forced to default.

Conesa and Kehoe (2012) modify the Cole-Kehoe model so that the country finds itself in an unexpected recession, where GDP is, say, 10 percent lower than its otherwise constant level.⁶ This is meant to correspond to the situation in Europe in 2008. In every period there is a constant probability—say 0.2, that is, one in five—that the economy will recover. With this stochastic process, which is like flipping a biased coin with the probability of heads being the probability of recovery, the expected waiting time for a recovery is a number of periods equal to the reciprocal of the probability of recovery. If, for example, the probability of an economic recovery is 0.2 per year, then, at any time where a recovery has still not occurred, the expected waiting time for a recovery is 1 / 0.2 = 5 years.

To understand gambling for redemption, consider first the case where self-fulfilling debt crises are not possible because, for some reason, the probability of bad news is zero. Then, because it wants to smooth expenditures as much as possible, a government would optimally choose to borrow when it is in recession at a high bond price equal to the present discounted face value, planning to pay back when the economy recovers. Like a gambler at a roulette wheel who keeps doubling his bet, the government is gambling that the recession will not continue for too long. Unlike the gambler, the government is doing something beneficial while it is gambling. It is smoothing government expenditures, something that the citizens of its country value.

If the recession does go on, there are two possibilities for the equilibrium outcome, depending on the costs of default: If the costs of default are high, the government will borrow less and less each period until its debt converges to an upper limit above which investors know that the government would default. If the costs of default are lower, the government will optimally choose to default after a finite number of periods, borrowing in the period before default at a price equal to the present discounted expected value of the face value if there is a recovery in the next period and the payoff in default if there is no recovery. This is not a self-fulfilling crisis: Investors and the government correctly anticipate default if there is no recovery. The only uncertainty is whether the economy will recover or not.

Consider now the general case where self-fulfilling crises are possible but where the economy is also in a recession from which it might recover. The government faces conflicting incentives. Various outcomes are possible and reasonable, depending on the values of parameters. The government could optimally choose either to pay down its debt to the upper safe limit or to borrow still more, running up its debt, gambling for redemption. The optimal choice depends on the costs of default, the probability of a crisis and the probability of recovery from recession.

Cristina Arellano (2008) argues that defaults can also occur when GDP is low enough. In her model, countries borrow large amounts in booms because interest rates are low because debt is below the upper safe limit. When a recession hits, however, the same amount of debt may be above the new upper safe limit, and interest rates rise, making it costly to roll over the debt. For a sufficiently large drop in GDP, a level of debt that is safe if GDP is high can be above the upper sustainable limit if GDP is low, in which case the government now prefers to default.

Analyzing EU and IMF policy and extending the model

We can use our theory to evaluate the impact of policies followed by the EU and the IMF. Any policy that decreases the price that a country receives for its bonds (that is, increases the yields that it pays), or increases the costs of default, provides the government with incentives to reduce its debt to exit the crisis zone. In contrast, any policy that increases bond prices (lowers the yields), or lowers the costs of default, provides the government with incentives to gamble for redemption.

The rescue packages listed in the timeline stopped self-fulfilling crises in Greece, Ireland and Portugal. They also provided credit to countries at lower interest rates than the yields presented in Figure 3. These policies can be interpreted as encouraging gambling for redemption. It is worth pointing out, however, that the rescue packages also explicitly required austerity measures, even if these requirements were later violated, especially in the case of Greece.

One policy that very clearly encourages gambling for redemption is the European Central Bank’s Securities Market Program (SMP). The SMP buys bonds of countries whose bond prices fall too low. By propping up their bond prices and keeping yields low, the SMP reduces incentives to pay down the debt and escape the crisis zone. Similarly, the ECB’s policy of reducing its repo rate and relaxing collateral constraints to encourage banks to buy government bonds with high yields drives up the price of bonds and encourages gambling for redemption.

Another policy that may have encouraged gambling for redemption was the 50 percent haircut on Greek bonds planned at the European Summit in July 2011 to be imposed on private investors, principally private banks in the EU. By labeling the haircut voluntary, the EU intended to eliminate some costs of default, such as triggering credit default swaps (CDSs), securities that pay the buyer in the event of a default. EU leaders thought that triggering CDSs would be very disruptive to the financial system, both inside and outside Greece. Greece had already reached a debt level that it could not hope to repay, but planning “voluntary” haircuts on Greek bonds signaled other troubled governments that such a reduction in the costs of default might be available for them.

By March 2012, it was clear, however, that this sort of “voluntary” haircut was not feasible, mostly because courts would not rule out claims on CDSs. Greece ended up imposing a much larger haircut, negotiating with the majority of bond holders and enforcing the settlement on the rest of bond holders by appealing to CACs (collective action clauses). There are currently doubts about the legality of this move, however, because the CACs were inserted into the bond contracts retroactively. A challenge for Europe is how to best design restructuring procedures for countries that might follow Greece into default while minimizing adverse incentives for other countries.⁷

While our theory provides an appealing explanation of why the threat of sovereign debt crises in Europe has been going on for so long, it leaves open a couple of major questions. We can use our theory to understand the behavior of leaders of countries threatened by debt crisis, like George Papandreou in Greece, but it does not help us understand the behavior of EU leaders like Angela Merkel of Germany and Nicolas Sarkozy of France, who have struggled to provide rescue packages. It may that they too have been gambling for the redemption of the eurozone itself, rather than their national economies. Merkel and Sarkozy may have believed that the only thing that will pull the eurozone out of the danger of debt crises is a vigorous economic recovery from the recession, and they are just trying to hold the EMU together until that happens. It would be useful to develop a model of this.⁸

It is also clear that the institutional design of the EMU—in particular, the mechanisms to enforce fiscal discipline, like the Stability and Growth Pact—is inadequate. European leaders are currently struggling to come up with a better institutional design, and it would be worth developing a theory of the optimal design of the EMU.

A related question is why sovereign debt crises like those in Europe do not currently threaten countries like Japan, the United Kingdom and the United States. These countries, like those in the eurozone, have large public debts and have suffered from the recent recession. Thomas Sargent (2012) presents a provocative narrative arguing that a key difference in the United States is that the central government has the power to raise substantial resources through taxation, a power the EU lacks. Another crucial difference is that each of these countries, unlike the eurozone countries, has its own currency whose value can fluctuate freely in response to changing economic conditions. This too is worthy of further research.

Endnotes

1 We thank Tito Cordella, Isabel Correia, Patrick Kehoe, Narayana Kocherlakota, David Levine, Thomas Lubik, Fabrizio Perri, and Pedro Teles for helpful discussions. We also thank Jose Asturias, Wyatt Brooks, and Laura Sunder-Plasssmann for excellent research assistance. The data presented in the figures are available here. The views expressed herein are those of the authors and not necessarily those of the Federal Reserve Bank of Minneapolis or the Federal Reserve System.

2 See Paul De Grauwe (2009) for a discussion of the politics of violations of the Maastricht Accords.

3 We assume that the bond holdings are a small fraction of the portfolio because this implies that the investor behaves as if he or she were risk-neutral, justifying expected-present-discounted-value pricing. If the bond holdings are a large fraction of the portfolio and the investor is risk-averse, then he or she would pay less, to compensate for the riskiness of the bond. We ignore any liquidity services provided by bonds and alternative assets, which can complicate the pricing relation in minor ways.

4 Cole and Kehoe (1996, 2000) model this news shock as what economic theorists call a sunspot, a random variable that affects the equilibrium only through investors’ expectations. The value of bad news is arbitrary and can vary over time, which would account for fluctuations in the spreads in Figure 3. The arbitrary nature of exactly what constitutes bad news is how the model captures what finance ministers refer to when they complain about their country’s bonds being at the mercy of the financial markets.

5 Whether this gives us more confidence in the quantitative properties of the model or more confidence in European policymakers is an open question.

6 To keep things simple, we assume that GDP does not have a growth trend. If GDP is 100 before the recession, it falls to 90 during the recession. A recovery is a return to 100. If there is a default during the recession, GDP falls another 5 percent, to 85.5. A recovery now only increases GDP to 95. It is easy to convert the model to one in which the economy is growing at a constant rate and in which neither the qualitative results nor the quantitative results change. In a more complicated model, the shock could affect the growth trend. Mark Aguiar and Gita Gopinath (2006) argue that shocks to growth rates have stronger effects on default incentives than do changes in levels.

7 David Benjamin and Mark Wright (2009) and Pablo D’Erasmo (2011) provide a theory for renegotiation between a government and a representative of the bond holders. They argue that it is worth delaying restructuring until countries have low default risk and high output because those are times when mutually beneficial outcomes can be obtained more easily. Their results imply that renegotiation is particularly difficult now when many Eurozone countries are still deep in recession and where there is substantial uncertainty about the future.

8 Arellano and Yan Bai (2012) argue that a reason for a lender—and the EU itself has become a major lender to troubled countries though the European Financial Stability Facility (EFSF) and the ECB’s SMP and repurchase agreements—to be lenient with a subset of borrowers in default is to avoid other defaults from other borrowers.

References

Mark Aguiar and Gita Gopinath (2006), “Defaultable Debt, Interest Rates and the Current Account,” Journal of International Economics, 69, 64–83.

Cristina Arellano (2008), “Default Risk and Income Fluctuations in Emerging Economies,” American Economic Review, 98, 690–712.

Cristina Arellano and Yan Bai (2012), “Linkages across Sovereign Debt Markets,” Federal Reserve Bank of Minneapolis.

David Benjamin and Mark Wright (2009), “Recovery before Redemption: A Theory of Delays in Sovereign Debt Renegotiations,” Centre for Applied Macroeconomic Analysis, Working Paper 2009-15, Australian National University.

Harold L. Cole and Timothy J. Kehoe (1996), “A Self-Fulfilling Model of Mexico's 1994–95 Debt Crisis,” Journal of International Economics, 41, 309–330.

Harold L. Cole and Timothy J. Kehoe (2000), “Self-Fulfilling Debt Crises,” Review of Economic Studies, 67, 91–116.

Juan Carlos Conesa and Timothy J. Kehoe (2012), “Gambling for Redemption and Self-Fulfilling Debt Crises,” Federal Reserve Bank of Minneapolis Staff Report 465.

Pablo D’Erasmo (2011), “Government Reputation and Debt Repayment,” Working Paper, University of Maryland.

Paul De Grauwe (2009), “The Politics of the Maastricht Convergence Criteria,” VoxEU, 15 April 2009.

Thomas J. Sargent (2012), “United States then, Europe now,” Nobel Prize Lecture, New York University.

Banks are vulnerable to self-fulfilling panics because their liabilities (such as demand deposits and certificates of deposit) are short term and unconditional, and their assets (such as mortgages and business loans) are long term and illiquid. To prevent wider financial fallout from such panics, governments have strong incentive to bail out bank debtholders. Paradoxically, expectations of such bailouts can lead financial systems to rely excessively—from a societal perspective—on short-term debt to fund long-term assets. Fragile banking systems thus impose external costs, and regulation may therefore be socially desirable.

In light of this fragility and cost, we examine two of the major theoretical benefits from the reliance of the banking system on short-term debt: (1) maturity transformation and (2) efficient monitoring of bank managers. We argue that while both justifications may be compelling, they point us to financial regulations very different from the ones currently in place. These theoretical justifications suggest that the assets funded by banks should not have close substitutes in publicly traded markets, as is currently the case.

Introduction¹

The enormous direct and indirect costs of rescuing banks and related financial institutions during the financial crisis of 2007-08 generated widespread policy debate on future banking regulation, resulting in part in the Dodd-Frank Act of 2010. Largely absent from these discussions was a careful reexamination of the services that banks provide and whether they are sufficient justification for the inherent financial fragilities of banks and the societal costs of this fragility.  

The inherent fragility of banks

In what sense are banks and similar financial institutions fragile? Answering this question requires understanding the balance sheets of financial institutions more generally. The assets of financial institutions are, by and large, financial assets, and claims on them are primarily financial liabilities. They hold few tangible assets such as land, buildings or machinery. Their financial assets consist mainly of promises to deliver dollars at future dates (and perhaps then only under certain circumstances). Among such assets are stocks, bonds and the notes on mortgages and business and consumer loans, as well as an array of more exotic financial instruments such as derivatives. Likewise, their financial liabilities consist mostly of a variety of obligations to deliver dollars at particular dates, under certain circumstances.

Some types of financial institutions, particularly banks, have liabilities that are mostly short term and unconditional. For example, banks issue demand deposits, which promise to pay a fixed amount of money whenever a depositor demands a withdrawal. Likewise, banks issue certificates of deposit (CDs), which promise to pay a fixed amount of money at a particular (usually very near-term) date. Typically, banks rely on the rolling over of CDs, when they mature, into new CDs. In this sense, a failure to roll over a short-term CD can be thought of as equivalent to a withdrawal from a deposit account.

In Figure 1, we display the distribution of firms—financial and nonfinancial—by their ratio of short-term debt to total assets, for all publicly traded companies. This figure shows that financial firms typically have much higher levels of short-term debt relative to total assets than nonfinancial firms. For example, 80 percent of financial firms have a ratio of short-term debt to total assets greater than 40 percent, while only 10 percent of nonfinancial firms do (see vertical dashed line).

Large Image

In Figure 2, we display the distribution of firms by their ratio of short-term debt to short-term assets, again for all publicly traded financial and nonfinancial firms. Firms that have ratios greater than 1 are using short-term debt to finance holdings of long-term assets. The figure shows that nonfinancial firms are far less likely than financial firms to use short-term debt to finance their holdings of long-term assets—less than 20 percent of nonfinancial firms depend on short-term debt, while roughly 60 percent of financial firms do so.

Large Image

The bulk of these financial firms with high levels of short-term debt are legally structured as banks. From an economic standpoint, the financial firms we focus on in this policy paper are those with high levels of short-term debt, and for convenience we’ll refer to them all as banks.

This reliance on short-term debt makes banks fragile in the sense that they are particularly vulnerable to the risks of insolvency and the possibility of confidence crises. Since bank assets are usually much longer term than their liabilities and since the value of these assets fluctuates, a bank’s net worth (its assets less its liabilities) also fluctuates a great deal.

Furthermore, a bank’s assets are typically illiquid. A bank cannot easily and quickly sell its portfolio of small business loans, home equity loans or jumbo mortgages, for instance, to satisfy an unexpected surge of withdrawals from demand deposit accounts. The illiquidity of banks’ assets and the demandable structure of their liabilities thus expose banks to crises of confidence. Since a bank typically will not be able to meet the demands of all depositors within a short period of time should they all choose to withdraw, banks are vulnerable to self-fulfilling panics in which depositors withdraw their funds simply because they believe other depositors will do so.

This panic is an entirely rational response even if the bank is solvent (though illiquid). If a depositor suspects that other depositors will withdraw their funds in the near future, it is rational for that depositor to rush to the bank and withdraw his or her deposits before other depositors can do so. But since this logic holds for all depositors, banks are subject to self-fulfilling panics in which the belief in a run causes the run.

In the midst of such self-fulfilling panics, governments have a strong incentive to intervene to bail out debt holders of banks in order to prevent the entire financial system from failing. Paradoxically, expectations of such bailouts can increase the incidence and depth of financial crises because once depositors believe that their deposits will be protected by general tax revenues in the event of potential systemic failure, they have less incentive to monitor the risk-taking proclivities of bank managers.

Bank managers, in turn, have increased incentive to take on risk, knowing their failures are implicitly insured—taxpayers (not banks themselves) bear the full consequences of this risk-taking. Given that the banking system rationally chooses to take on risk, the possibility of insolvency of the banking system increases, and the need for bailouts rises.

In this way, expectations of bailouts can lead financial systems to rely excessively—from a societal perspective—on short-term debt to fund long-term assets. Fragile banking systems thus impose external costs, and regulation may therefore be socially desirable.

The benefits of banks

Why then do we allow banks—financial institutions that fund long-term assets with short-term debt—to operate? One answer is historical: The very ubiquity of banks throughout history suggests that they serve a valuable function. But the world has much changed since the invention of banks. In particular, the range of economic activities that can be funded through publicly traded assets has expanded enormously. For instance, in the not-too-distant past, publicly traded mortgage-backed securities did not exist. In light of these changing circumstances, it is worth examining how long-term assets should be funded and the role banks should play in funding such assets.

Given that we have already discussed the weaknesses or costs of such financial institutions, we’ll now use economic theory to examine the possible social benefits of having a financial system in which illiquid assets with volatile values are funded by demand deposits and short-term debt. This cost-benefit analysis allows us to ask how modern economic theory can be used to design better regulatory systems for banks. This issue is clearly of pressing importance given the central role that banks are argued to have played during the recent financial crisis. This issue is also of central importance given the importance of banks to financial systems throughout the world.

We examine the role of banks and the role of regulation by considering, in turn, two of the major theoretical justifications for the reliance of the banking system on short-term debt:

• Demand deposits allow banks to engage in socially useful maturity transformation.
• Demand deposits allow for efficient monitoring of bank managers.

We argue that while both justifications may be compelling, they point us to financial regulations very different from the ones currently in place. Specifically, they suggest that while it may be important to have institutions that finance long-term assets with short-term debt, the assets that are so funded should not have close substitutes in publicly traded markets.

Our logic is as follows: The economic case for regulating such institutions at all is due to the external costs they impose. But any such regulation should also, to the extent possible, preserve the benefits that economic theory suggests they convey. We argue that these benefits exist only when these institutions hold assets which do not have close substitutes that are traded in public markets. Indeed, our analysis suggests that bank assets having close publicly traded substitutes destroys the possible social benefit of banks under the maturity transformation view and reveals the social benefit to be zero (or small) under the efficient monitoring view. Thus, institutions that issue short-term debt should be allowed to hold only a limited amount of publicly traded assets.

Theoretical justifications of banking

We now provide closer consideration of the two primary theoretical justifications for a banking system that relies on short-term debt. We first describe and then assess the maturity transformation rationale, and then we turn to the notion that demand deposits permit efficient monitoring of bank managers, again starting with a description followed by our assessment.

Maturity transformation
Banks are often said to perform a miracle of financial alchemy referred to as maturity transformation. They are thought to hold long-term assets that yield a high rate of return and finance these assets with short-term claims (such as deposits) that pay a higher rate of return than what those depositors could earn if they invested directly in short-term assets (such as very short-term bonds).

Such a form of financial alchemy implies that those who roll over their short-term claims until the maturity of the long-term asset will necessarily receive a lower rate of return than if they had directly held the long-term asset. In effect, maturity transformation is a redistribution of resources from those who roll over their short-term claims to those who redeem such short-term claims early. Why would this financial alchemy be socially desirable?

Diamond and Dybvig (1983) developed a model in which such maturity transformation is desirable and possible. Furthermore, they showed that the arrangement that delivers such maturity transformation resembles banks in the sense that (at least conceptually) households get access to their funds whenever they want to, so that the liabilities of banks resemble demand deposits, while the assets of banks are long term and illiquid. (See Appendix A-1 for a simplified technical description of the model.) The basic idea in their model is that the redistribution of resources associated with maturity transformation is desirable because such redistribution allows for insurance. If people are not sure whether they will need resources only at the maturity of long-term assets or at a date prior to maturity, they may wish to insure themselves by accepting a lower return at maturity in exchange for a higher return should they instead desire their funds sooner.

An assessment of the maturity transformation view
While Diamond and Dybvig’s model provides important insights into the possibility of maturity transformation, the mechanism rests on an increasingly implausible feature of the model: nonbank financial institutions do not also hold assets such as the mortgages and mortgage-backed securities typically held by banks as part of their transformation role [a point originally made by Jacklin (1987)]. Put simply, if it is possible for households to hold the long-term assets of banks outside the banking system, the ability of banks to perform maturity transformation is destroyed, along with the social benefit from doing so. (For technical discussion of this point, see Appendix A-2.)

As we show in Appendix A-2, if people can hold long-maturity assets outside the banking system, they have strong incentives to do so. If they happen to not need funds prior to maturity, they can obtain higher returns than under maturity transformation since they are not subsidizing those who need their funds sooner. If they happen to need their funds sooner, they can simply sell their holdings of these long-term assets to customers of the banking system who do not need their funds until maturity. Further, these customers will be willing to sell to them. Recall that such bank customers were promised relatively low returns for funds held long term inside the banking system and relatively high returns if they needed their funds earlier. But for these customers, the event they were insuring against—that they would need their funds early—did not occur. After they know this, they have strong incentives to trade with those outside the banking system in order to achieve higher returns. Analysis of the Diamond-Dybvig model under this alternative assumption—that the assets typically held by banks are also traded in public markets—demonstrates that banks cannot provide the social benefit associated with maturity transformation.

Recent data on holdings of U.S. financial institutions show that, indeed, market trading of such assets is increasingly common.

Figure 3 shows mortgages and mortgage-backed securities held by bank and bank-like entities in the United States relative to their stock of financial assets (typically known as bank credit). Note that such assets constitute almost 50 percent of bank financial assets over the past decade.

Large Image

Figure 4 shows the total outstanding stock of mortgages and mortgage-backed securities held by banks and bank-like entities relative to the total stock of mortgages. Note that these banks and bank-like entities hold only about 40 percent of the total outstanding stock of mortgages over the past decade, and that share has declined steadily since the late 1970s. That is, institutions that are very different from banks now hold a very significant and rising share of the total outstanding stock of mortgages. Such institutions do not fund their assets with short-term debt.

Large Image

In this sense, the ability of banks to achieve maturity transormation is severely limited by the holdings of close substitutes by other institutions. Figure 5 shows bank loans for commercial and industrial purposes relative to the stock of financial assets (again, typically known as bank credit) held by banks. Such bank loans may well not have close publicly traded substitutes. But they also make up only one-quarter of bank assets. This perspective suggests that if we view maturity transformation as the principal reason for the existence of banks, the United States could do very well with a banking sector just a quarter the size it now is.

Large Image

Efficient monitoring
Over the past 25 years, a theoretical literature has emerged arguing that short-term debt and the associated possible runs on financial institutions may in fact be an efficient way of allowing markets to incorporate information that investors have about returns on financial assets. [See Chari and Jagannathan (1988), Calomiris and Kahn (1991), Diamond and Rajan (2001) and Zetlin-Jones (2011).]

The basic idea is that some investors often have access to information about the returns on financial assets and that this information is valuable to other investors and financial markets as a whole. Efficient arrangements will then have to provide such investors with incentives to obtain and act on this information, and those incentives may require market mechanisms that provide higher returns to those investors compared with others. These differential returns can be efficiently provided through “bank runs” in which early withdrawers (“sophisticated” investors who have relevant information through their close monitoring of managers) get higher returns than late (“uninformed”) withdrawers. [See Appendix A-3 for a simplified technical discussion, using the Calomiris-Kahn model (1991).]

One way of implementing the efficient arrangement outlined above is to think of assets being provided and being funded by demand deposits. The demand deposits allow all investors to withdraw their assets at will. In normal times (when sophisticated investors receive favorable signals about the bank’s returns), all investors wait to withdraw their assets at maturity and receive the same return. In crisis times (when sophisticated investors get unfavorable signals), large withdrawals occur. Those who choose to withdraw early receive a higher rate of return than those who wait until maturity. The observation that investors who withdraw early get a higher return resembles a bank run and therefore creates a crisis for the bank.

An assessment of the efficient monitoring view
The efficient monitoring view of the social usefulness of banks rests crucially on the idea that it is difficult to set up alternative methods of compensating sophisticated investors and/or efficient managers. But, in fact, this assumption may be unrealistic. For example, one could easily imagine various kinds of equity claims as well as combinations of equity and debt that could provide sophisticated investors with appropriate incentives to monitor the activities of managers. One could also imagine compensation contracts for managers that are tied to the market valuation of their assets. Such mechanisms are widely used in firms whose claims are traded in public markets.

On the one hand, to the extent that close substitutes to the assets held by banks are traded in public markets, the marginal value of having banks fund these projects as well is likely to be small since these assets would likely be funded by public markets even if banks did not exist. On the other hand, the costs of banks—in the form of crises and their associated bailouts (as well as the changes in bank risk-taking behavior induced by the expectation of bailouts)—are clearly quite large.

In this sense, under both the efficient monitoring and the maturity transformation views, banks provide significant social value only when the assets they hold do not have close substitutes that are traded in public markets. Bank assets having close publicly traded substitutes destroys the possible social benefit of banks under the maturity transformation view and reveals the social benefit to be zero (or small) under the efficient monitoring view.

Finally, the efficient monitoring view also poses a severe challenge to those who view bailouts as necessary for the proper functioning of financial markets. Note that under the efficient monitoring view, crises do occasionally occur, and, at the time of the crisis, if governments were particularly concerned about the well-being of unsophisticated investors, bailouts of all investors would be considered desirable. But if such bailouts are anticipated, sophisticated investors have no incentive to monitor banks when such monitoring is costly. Absent such monitoring, either valuable projects will not be undertaken or, given the expectations of large transfers from the government, inefficient projects will be undertaken.

Implications for policy

To answer the question “Do we need banks?” we first need to answer the question “What are banks?” Guided by the data, we have chosen to think of banks as institutions that hold long-term, illiquid, volatile assets and issue large amounts of short-term debt to finance these assets. Given the possibility of bank crises, this institutional arrangement seems puzzling on the face of it. Nevertheless, we need to take seriously the observation that such institutions have been a durable part of the economic landscape for many centuries and, arguably, have played a significant role in the industrial revolutions the world has been fortunate enough to experience.

Both the maturity transformation and the efficient monitoring views suggest that it may desirable to fund long-term assets that have no close substitutes in publicly traded markets with short-term debt. Both views also suggest that the social value of funding long-term assets that have close publicly traded substitutes with short-term debt is small. Given the costs imposed by crises and attendant bailouts, both views suggest that it may be desirable to allow these institutions to issue short-term debt only if their assets do not have close publicly traded substitutes. Otherwise their benefits are either zero or negligible.

These implications suggest policies that are dramatically different from those undertaken by most bank supervisors or various forms of international regulation under the Basel agreements. Current supervisory systems encourage banks to hold assets with close substitutes for publicly traded assets by giving such assets lower risk ratings and by requiring less equity capital to back up these assets. These supervisory systems encourage the banking system to hold publicly traded assets that are risky in the full knowledge that if the returns on these assets are poor, governments will step in to bail out short-term debt holders.

The framework for regulatory policy implied by our analysis would lead to a banking system that is radically different from the one we currently have. Institutions that issue large amounts of short-term debt relative to their assets would be regulated and required to hold relatively little of their assets in publicly traded securities.

This new system would not eliminate all crises. Indeed, we have argued that bank runs may well be an essential ingredient of an efficient economic system. In all likelihood, however, the crises of this new system would be much smaller and less costly than those we have experienced in the recent past. One reason: The volume of assets backed by short-term debt would be dramatically smaller under our proposed system than under the current system.

The economic theories explored in this paper suggest we do need banks. These theories also point us to constructive ways in which we can reform the financial system to make it more efficient and to minimize the spillover costs imposed on the broader economy by crises that affect particular financial institutions.

Endnote

1 The authors thank Narayana Kocherlakota, Dick Todd and Kei-Mu Yi for useful comments and Doug Clement for editorial assistance. V. V. Chari thanks the National Science Foundation for supporting the research that led to this paper.

References

Calomiris, Charles W., and Charles M. Kahn. 1991. The role of demandable debt in structuring optimal banking arrangements. American Economic Review 81 (3): 497–513.

Chari, V. V., and Ravi Jagannathan. 1988. Banking panics, information, and rational expectations equilibrium. Journal of Finance 43 (3): 749–61.

Diamond, Douglas W., and Philip H. Dybvig. 1983. Bank runs, deposit insurance, and liquidity. Journal of Political Economy 91 (3): 401–19.

Diamond, Douglas W., and Raghuram G. Rajan. 2001. Liquidity risk, liquidity creation, and financial fragility: A theory of banking. Journal of Political Economy 109 (2): 287–327.

Farhi, Emmanuel, Mikhail Golosov and Aleh Tsyvinski. 2009. A theory of liquidity and regulation of financial intermediation. Review of Economic Studies 76 (3): 973–92.

Jacklin, Charles J. 1987. Demand deposits, trading restrictions, and risk sharing. In Contractual Arrangements for Intertemporal Trade, ed. Edward C. Prescott and Neil Wallace. Vol. 1 of Minnesota Studies in Macroeconomics. Minneapolis: University of Minnesota Press.

Wallace, Neil. 1988. Another attempt to explain an illiquid banking system: The Diamond and Dybvig model with sequential service taken seriously. Federal Reserve Bank of Minneapolis Quarterly Review 12 (Fall): 3–16.

Zetlin-Jones, Ariel. 2011. Efficient financial crises. Working paper. University of Minnesota.

In this note, we demonstrate and analyze the inability of standard neoclassical models to generate accurate estimates of the fiscal multiplier (that is, the macroeconomic response to increased government spending). We then examine whether estimates can be improved by incorporating recently developed theory on demand-induced productivity increases into neoclassical models. We find that neoclassical models modified in this fashion produce considerably better estimates, but they remain unable to generate anything close to an accurate value of the fiscal multiplier.

Introduction¹

There is a vast empirical literature studying the “fiscal multiplier,” the response of macroeconomic variables to an increase in government spending. A summary of this literature is that the multiplier—defined as the response of gross domestic product to an exogenous change in government expenditures—is somewhere in the range of 0.7 to 1.0 (see Hall 2009 for a discussion of the empirical findings) and perhaps even higher in extraordinary times like ours with a very low nominal interest rate and a very high unemployment rate. Put more concretely, if the federal government were to increase spending by $1 billion, GDP would increase by between $700 million and $1 billion.

Government expenditure provides a variety of goods and services (public roads, national representation and the like), many of which benefit a wide range of people regardless of who provided the revenue to pay for them and thus are unlikely to be privately provided. Policymakers also view public expenditures as a fiscal tool to affect macroeconomic variables such as output and employment. This function has particular relevance during recessions, when policymakers seek mechanisms that can restore employment and economic growth to healthy levels.

To assess the usefulness of government expenditure as a fiscal tool, we need to weigh the costs of fiscal expansion that arise because of the higher taxes that are eventually required against the benefits generated from higher government spending in terms of higher output. This requires explicit models, suitable for policy analysis, that faithfully capture the relevant features of the economy, and this realism can be judged by seeing whether a given model is capable of generating the fiscal multiplier seen in the data.

At least until recently, the workhorse of both macroeconomics and public finance has been the real business cycle model, also referred to as the standard neoclassical model, or SNM. However, this model of the economy is not able to deliver realistic results for the fiscal multiplier—that is to say, the multiplier values the SNM predicts are close to 0, nowhere near the estimates provided by actual data (again, between 0.7 and 1). In other words, according to the SNM, an increase in government spending would have no discernible effect on economic output, as it will be accompanied by sharp reductions in consumption and investment. Clearly, something is missing from the SNM.

Another type of macroeconomic model, referred to as New Keynesian, is able to yield a multiplier in the range of the empirical literature.² But at the same time, these New Keynesian models generate other unrealistic results, such as a decline in the markup ratio of price over cost when output rises, and a dramatically procyclical labor share. The data show the opposite: Markups tend to rise during economic booms and decline in recessions (again, Hall 2009), while labor share tends to shrink in expansions. Unfortunately, then, these models also fail to faithfully reproduce the macroeconomy and so are of limited use to economists or policymakers in designing fiscal policies.

Other models that may be helpful in generating a realistic multiplier use frictions in labor markets to generate involuntary unemployment, but these have yet to be fully developed.

Seeking better results

In this note, we briefly explore the implications for the fiscal multiplier of a family of models developed in some of our recent research (Bai, Ríos-Rull and Storesletten 2011, Dyrda, Kaplan and Ríos-Rull 2011 and Dyrda and Ríos-Rull 2012) that, while squarely in the neoclassical tradition, imply changes in some of the main properties of neoclassical models.

The impact of an increase in government expenditure on output depends on several factors, and key among these is the extent to which government spending replaces or “crowds out” private consumption and investment. If government spending simply takes the place of private spending that would have occurred in its absence, then there will be no net effect on total output, and the multiplier is 0.

However, a multiplier of 1 results when increased government spending does not induce a reduction in private consumption and investment, but rather adds to it. But, of course, increases in government expenditure have to be paid for eventually, and the effects of this burden depend on the duration of the increase in expenditures and on the form of taxation. Throughout this note, we will explore the effects of short-lived additional public expenditures that are paid for in a very efficient manner, with lump-sum taxes.³

Our research suggests that models in the neoclassical tradition can produce higher estimates of the fiscal multiplier via two channels. Dyrda, Kaplan and Ríos-Rull (2011) demonstrate the importance of adjusting neoclassical models to increase the response of hours worked to temporary changes in wages. Model parameters of this responsiveness—that is, the change in labor supplied by workers when wage rates rise or drop temporarily—should be consistent with new evidence about how household size changes in response to macroeconomic change. In contrast to the traditional notion that people can be identified with households, our data analysis reveals that people’s living arrangements change often in ways that are partly synchronized with the business cycle: During recessions, households take in more members as young adults move into (or delay departure from) their parents’ home. Seniors may also move in with middle-aged relatives, and unrelated singles may form group households.

More importantly, in terms of the fiscal multiplier, Bai, Ríos-Rull and Storesletten (2011) articulate a neoclassical model in which movements in productivity are not the result of technological shocks, but of greater willingness to spend or higher demand. As we will see, this explanation gives solid theoretical rationale for macroeconomic expansion when government spending increases.

In the next section, we describe reasons behind the inability of the SNM to generate realistic values of the fiscal multiplier. We then move to a brief discussion of our findings about fluctuations in household size that affect the responsiveness of hours worked to changes in wages (the “Frisch elasticity”) and how that affects the fiscal multiplier. Finally, we discuss neoclassical models with “demand-enhanced” productivity based on our earlier and ongoing work (Bai, Ríos-Rull and Storesletten 2011 and Dyrda and Ríos-Rull 2012) and the extent to which these modifications to the SNM can deliver values of the fiscal multiplier closer to the empirically documented range. Along the way, we make some observations about what could be capable of generating fiscal multiplier values in the actual empirical range without drastic departures from the neoclassical tradition.

The failure of the standard neoclassical model

Two features of the SNM are central to its estimates of the fiscal multiplier; both concern household decisions about how many hours they’ll work. And both ultimately lead to the model’s inability to realistically predict the fiscal multiplier.

The first is that the number of hours worked is the outcome of choices that households make in response to wages and interest rates. That hours worked respond to wage rates is rather intuitive, but households also adjust labor in response to interest rates because, among other things, interest rates affect the value of the present relative to the future. Through so-called substitution effects, hours worked are then affected.

The model’s second key feature: The two inputs essential to producing economic output—capital and labor—are subject to diminishing returns, and each is paid its marginal product (the market value of the last unit of output it contributes). So, the more units of labor or capital that are added to the economy, the less additional output will be obtained. Therefore, as the supply of capital increases, its marginal product (the interest rate) decreases. The same applies to hours worked and the wage.

These features, and their complex interaction with other economic factors, mean that increasing the number of labor hours in order to increase economic output—the mechanism through which a fiscal multiplier works in the neoclassical world—ultimately requires that both consumption and investment decrease in the short term.⁴

But to estimate the net effect of a fiscal multiplier, the question is, How large are these drops? That is, to what degree will the positive impact of government expenditures on economic output—through increased labor supply—be negated by the short-term decreases in consumption and investment?

To answer this question quantitatively, economists feed specific parameter values into the model’s equations. The values are selected to match key statistics in the U.S. economy. So, government expenditure is usually set at 16 percent of GDP, a typical figure for the United States. Investment and private consumption are about 19 percent and 65 percent of GDP, respectively. Labor income is roughly two-thirds of total GDP. The nominal interest rate is 4 percent. Another important statistic is the “intertemporal rate of substitution,” a term economists use for people’s willingness to forgo consumption now in order to consume in the future; this is usually set at 0.5. And people are assumed to work 30 percent of their nonsleeping time. All of these parameter values are widely agreed upon by macroeconomists.

The only controversial number in this model is the one that describes the willingness of people to work longer than usual if the wage is temporarily high—again, the Frisch elasticity. (See Chetty, Guren, Manoli and Weber 2011 for a discussion of the debate among economists as to its magnitude.) Although disputable, a very conventional value used in such models is 0.7. The value is calculated using a married couple as the notion of household (see Heathcote, Storesletten and Violante 2010).

With these parameter values, the model predicts that a temporary increase in government expenditures of 1 percent of GDP will have very little net impact. It results in just a 0.023 percent increase in GDP. Why so little effect? The primary cause is a dramatic drop in investment. A 1 percent increase in government expenditures leads (according to this model) to investment levels falling from 19 percent of GDP to 18.04 percent, wiping out 96 percent of the 1 percent boost in government spending.

Consumption also drops, but just slightly, falling from 65 percent to 64.98 percent of GDP. And, indeed, the increase in government spending barely increases the number of labor hours supplied by households, from 30 percent of total available time to 30.01 percent. Put in more practical terms, the neoclassical model predicts that if government boosts public spending by 1 percent of GDP for three months, the average adult would work no more than 10 minutes longer.

First candidate for a bigger multiplier:
A higher Frisch elasticity

How can the model achieve better results, closer to actual estimates of the impact of a fiscal multiplier (again, in the range of 0.7 to 1, rather than the 0.023 just obtained)? The lynchpin appears to be labor’s response to an increase in wages, the Frisch elasticity. A higher figure would have dramatic influence over the model’s estimate of the fiscal multiplier.

As mentioned above, a conventional if debatable value used for Frisch elasticity is 0.7. But our recent work, Dyrda, Kaplan and Ríos-Rull (2011), makes the case for a higher estimate. The gist of our argument is based on the fact that standard measurements of the Frisch elasticity of labor assume that households are stable in that they keep the same characteristics over time (for instance, that married people stay married). But, of course, many people are not “stable” in this sense; young people move in and out of their parents’ home, some people divorce and become two households and the like.

In our work, we show that those “unstable people” display a higher Frisch elasticity than the 0.7 figure based on a stable population. In addition, we argue that the concept of the household itself is not set in stone, and we document that the size of households is larger in recessions as many people move in with family or friends to bear the harder times. When they do so, they work even fewer hours than they would normally. According to our calculations, properly accounting for unstable people and their movements in and out of households changes the Frisch elasticity of the economy as a whole from 0.7 to 1.1.

When we apply this higher value of the Frisch elasticity of labor to the standard neoclassical model, we obtain a higher multiplier.⁵ Indeed, the multiplier goes up by 31 percent. Unfortunately, this improvement is less impressive than it sounds: Given the low initial value of 0.023, a 31 percent increase yields very little: a prediction of just 0.029—still nowhere close to empirical measurements in the range of 0.7 to 1.

Why standard neoclassical models fail

Neoclassical models contain two “first-order conditions” (or mathematical requirements) that together determine the response of workers to increases in government expenditures. Understanding these conditions helps explain why neoclassical models generate such low estimates of the fiscal multiplier.

One of these conditions—the “intratemporal”—links variables such as working hours, consumption levels and wage rates in the here-and-now: “today.” So, the “intratemporal first-order condition” says what all hourly workers know: Consumption by a household is partially determined by the prevailing wage rate times the number of hours that household members work at that wage.

But because labor in neoclassical models is paid its marginal product and is also subject to diminishing returns, prevailing wages in the economy as a whole will decline as the labor force grows, discouraging households from offering more labor hours. To increase working hours (in order to generate more output) as wages decline, a fall in consumption is necessary.

But to increase output in a neoclassical model, a fall in consumption isn’t enough. The other first-order condition—the “intertemporal”—must also be satisfied. This condition refers to the way variables interact over time—that is, between the here-and-now and the future: “today” and “tomorrow.” So in this context, this condition mathematically links the ratio of hours worked “today” and “tomorrow,” on one hand, with the ratio of wages today and wages tomorrow, multiplied by tomorrow’s interest rate, on the other hand.

Because the intratemporal condition has just indicated that wage rates are declining, the only way the intertemporal condition can generate an increase in the first ratio (working hours today over working hours tomorrow) is to significantly increase interest rates tomorrow. Because interest rates are the marginal productivity of capital (just as labor is paid its marginal product) and capital’s marginal productivity is higher when there’s less of it, then the only way to get a high interest rate is to decrease the amount of capital. This means that investment “today” has to be low.

So, the bottom line (of neoclassical models with these two conditions) is that to obtain the increase in working hours necessary for an increase in GDP, both consumption and investment must fall. But as we’ve seen in our model simulations, only investment fell significantly; consumption barely budged. Therefore, the models fail to generate a realistic increase in working hours or—thereby—a rise in GDP.

Other economic models exist in which the intratemporal first-order condition is not necessary. New Keynesian models with rigid wages, for example, operate without it. In this class of models, firms are unwilling to hire more workers at the prevailing wage. Increased government expenditures increase profits for firms, and that induces them to hire more workers at the current fixed wage. The number of hours worked then rises.

Another type of model holds that firms and workers expend considerable and costly effort in seeking good fits. These search models of unemployment hold that there are always willing workers, but they are costly to find. In this class of models, the intratemporal first-order condition is absent, but a zero-profit condition on firms applies. Increases in government expenditures may provide firms with incentives to look for more workers, and this might provide a satisfactory mechanism for generating a realistic multiplier.

Can government expenditures increase wages?

As we’ve seen, the neoclassical model is unable to generate realistic estimates for the fiscal multiplier, in large part because decreasing returns to labor—the diminishing additional output from each additional hour of work—mean that workers face diminished wages as government expenditures increase. Is there a different way to design a model so that government expenditures could increase wages instead of reducing them? If so, that would certainly help to increase hours worked.

In fact, this is what New Keynesian models with rigid prices do. As government expenditures increase, firms in these models are required to deliver the goods or increase prices or both. Since some firms cannot adjust prices, they need to hire more labor to deliver the goods. To induce households to provide more labor hours, wages have to go up. However, fixed prices and higher wages imply that profits and markups fall as a result of the increase in government expenditures.

Again, Hall (2009) provides a very nice description of the issue and of the lack of evidence for a drop in markups when government expenditures rise. Let us add that the behavior of labor share, which is slightly countercyclical in the data, is also at odds with the prediction of New Keynesian models with fixed prices.

In fact, greater potential lies in a model in the neoclassical tradition, with a twist.

A second candidate: The shopping model

In a recent paper, Bai, Ríos-Rull and Storesletten (2011) built a modified neoclassical model that goes a significant way toward generating a realistic multiplier. This model allows any increase in expenditures (which we refer to as demand) to increase productivity and, with it, to increase wages. The innovation in this model is that generating output requires not only inputs of production, but also the active participation of the purchasers of goods and services.

A few examples can help to illustrate this mechanism, and perhaps the best illustrations are from service industries. The tourism industry needs tourists, restaurants need diners, hotels need guests, hospitals need patients, movies have to be seen, advice has to be heard, pedicures need toes, and so on. In our model, consumption and investment require the active participation of consumers and firms. Without spenders, there is no output. Productivity goes up when the contribution of buyers (consumers, firms and the government) increases. It is the buyers—not producers—who are ultimately responsible for increased productivity by exerting more—but unmeasured—effort to use the economy’s productive capacity more intensely.

In the model, these ideas are implemented by requiring consumers not only to pay for the goods, but also to find them, a disagreeable task that is costly in terms of utility (just as is work in the standard model for those who prefer leisure). An increase in consumption can be achieved through both an increase in labor that raises the potential output of the economy, and an increase in search effort that allows households to find more output, thus making the economy operate at higher capacity.

Firms stand ready to produce, with capital and labor, but output occurs only when consumers find the firms and generate demand for that output. The search efforts of consumers are not measured in GDP, and the higher output is imputed to higher productivity.

While the spirit of this “shopping economy” is neoclassical—prices are flexible, people and firms are restricted only by technology, markets clear—the aggregate production function with constant productivity (a traditional workhorse of the SNM) does not hold.

In the shopping economy, as in all models, there is a budget constraint that requires households to pay for what they (and the government) purchase. An additional constraint, unique to this economy, requires households to search in order to find consumption goods. The more search effort consumers expend, the more goods will be found and, thus, the higher output will be. Consequently, output can increase even if there are no changes in measured inputs (since search effort itself is unmeasured).

In our paper, we use the modern concept of competitive search that achieves an optimal allocation and thus guarantees that the model has a unique prediction. However, we include all forms of hassle associated with searching for consumption goods, such as receiving worse service in restaurants at capacity, a lengthy wait in emergency rooms on Saturday nights, not getting the right options or color when buying a car and so on.

All of these hassles are greater when the economy is in an expansion, generating higher productivity as a result of higher demand. During recessions, hassles diminish—parking spaces, shopping lines, waiting times all decline; clerks and salespeople stand idle as they wait for customers. This dynamic applies to firms, as well, since they have to search for investment goods. Purchasing departments and shopping professionals need to find the right type of capital goods, a task that is clearly less costly during recessions.

The shopping economy model holds potential for generating a fiscal multiplier more in line with the empirical estimates. This is because increased government spending in this model generates higher productivity, and that may generate higher wages, in contrast to the SNM. To gauge this potential, we added a government sector financed with lump-sum taxation to the model in Bai, Ríos-Rull and Storesletten (2011).⁶

As in the SNM, an increase in government expenditures induces an increase in hours worked.⁷ But unlike in the SNM, productivity goes up, which can potentially increase wages, or at least slow down their reduction. Productivity increases because now people and firms have to look harder to find goods since higher government spending has raised effective demand.

When the relevant parameters are applied to the shopping model, it generates much larger effects than did the SNM. An increase in government expenditures of 1 percent of GDP (from 16 percent of the average value of GDP to 17 percent) now yields a fiscal multiplier of 0.172, over seven times that generated by the SNM. This results from both a 0.07 percent increase in productivity and a 0.09 percent increase in hours worked. By comparison, the SNM generates no increase in productivity and a 0.034 percent increase in hours worked.

The main difference between the shopping economy and the SNM is in investment, with the reduction being much smaller in the shopping economy. Interestingly, consumption in the shopping economy falls by more than in the SNM—from 65 percent to 64.94 percent—and investment falls from 19 percent to 18.22 percent. As in the SNM, an increase in the Frisch elasticity to 1.1 also increases the multiplier, albeit by a small amount (from 0.172 to 0.187).

These fiscal multiplier estimates are still a far cry from the values between 0.7 and 1 found in the empirical literature. One way for models of the neoclassical tradition to produce a higher value would be to have “time to build” features to avoid wild oscillations in investment.⁸ For example, we could pose a requirement that investment projects take a long time to both plan and implement. Therefore, in any period, only a fraction of the current investment is chosen, the rest being the outcome of previous decisions. Similarly, in the current period, decisions are made about future investment. With this formulation, we would obtain some rigidity in investment, not because markets do not work well, but because of technological—indeed, engineering—reasons. Short-lived increases in government expenditures would not in this case disrupt investment very much, and a swift increase in government expenditures would imply a larger multiplier. The exact calculations would require a specification of how to absorb the losses of firms that are required to secure the investment goods. In this scenario, a much larger drop in consumption is likely. This is a topic that deserves more attention.

The multiplier in a recession

So far we have looked at the multiplier in average times, when the economy is operating normally. But “normal times” are not when policymakers consider using government expenditures as a tool to expand output. Leading researchers (Christiano, Eichenbaum and Rebelo 2010, for instance) postulate that in certain circumstances, such as when the nominal interest rate is 0 percent, New Keynesian models with fixed prices give rise to large fiscal multipliers, suggesting that such a policy tool would be particularly potent under such conditions.

To analyze the multiplier in recessions, we need to “engineer” a recession in our models. In neoclassical contexts, recessions occur because of poorer-than-average technology or because of preference changes that induce people to work less. In this exercise, we first consider shocks to preferences in both SNMs and shopping models that make work less agreeable. We set the sizes of the shocks so that GDP shrinks by 1 percent, and we use a high Frisch elasticity of 1.1.⁹

In the SNM, a 1 percent drop in output is generated via a 1.5 percent reduction in hours worked. However, consumption barely changes (it goes down just 0.02 percent), while the bulk of the reduction in output is accommodated by lower investment. In actual recessions, investment typically falls by more than consumption, but in a less extreme manner. In the shopping economy, a similar 1 percent output drop comes from a drop not only in labor, but also in productivity, which falls by 0.4 percent.

Now that we’ve engineered a recession in the model economies, we can consider an expansionary policy that increases government expenditures the same amount in both. We find that the fiscal multipliers are higher than in normal times, as theory has suggested, but barely so. In the SNM, the multiplier is up to 0.0298 relative to 0.0296 in normal times. In the shopping economy, it is 0.1916 in a recession, up from 0.1871 in normal times. The gains in the shopping economy come from productivity (0.07 percent) and from labor (0.12 percent); in the SNM, the gains are all from labor.¹⁰

The shopping economy allows for an additional type of recession: a shock to the cost of search—the willingness of people to put up with the hassle needed to find goods and services. An increase in search costs thus acts as a shock to demand and generates a recession. Such a recession compounded with the expansionary policy generates the largest multiplier found in these exercises, a value of 0.221.

This quick review of multipliers during recessions generated by neoclassical models arrives at the same destination: The fiscal multiplier is almost 0 for the SNM and about 0.2 for the shopping economy. The reason these models predict such similar multipliers, whether the economy is in a recession and not, is clear. Neoclassical models assume that there are no market frictions, such as wages that don’t adjust quickly or prices that don’t change right away. Therefore, in these sorts of models, recessions are generated because people simply choose to work less. Therefore, the models respond in recessions very much the same as they do in normal times: The model economy operates optimally.

Conclusions

This note shows that neoclassical models of the business cycle have serious shortcomings that limit their ability to evaluate the effect of changes in government expenditures. State-of-the-art extensions can increase the predicted values for the fiscal multiplier, but these estimates are still much smaller than those measured empirically.

Many researchers have embraced New Keynesian models in which the central mechanism generating large fluctuations is the assumption that prices are fixed, so firms and workers are unable to adjust (nominally denominated) contracts. While this may be fruitful, we think that some of its counterfactual implications (specifically, on price-to-cost markups and labor share) and the lack of solid theoretical explanation for why prices are fixed justify the exploration of models with neoclassical flavor, but with enough frictions so as to allow for recessions as situations with idle resources. We think that such models should include frictions in labor markets that make many households work fewer hours than they would like and consume less than desired because of difficulties in borrowing.

Models with these features in addition to those provided by our shopping economy, in which increases in government expenditures boost productivity, may be capable of yielding multiplier values consistent with those in the empirical literature. These models are the subject of our ongoing work.

Endnotes

1 The authors thank Doug Clement and Carolyn Wilkins for their editorial comments and Fabrizio Perri for discussions of theory. The views expressed herein are those of the authors and not necessarily those of the Federal Reserve Bank of Minneapolis or the Federal Reserve System.

2 For example, Drautzburg and Uhlig (2011) quantify the effect of the American Recovery and Reinvestment Act of 2009 using the benchmark New Keynesian model developed by Smets and Wouters (2003) with additional frictions. They yield short-run multipliers around 0.52. Leeper, Traum and Walker (2011), using Bayesian techniques, show that New Keynesian models with sticky prices and wages can generate multipliers around 1.

3 Unlike income taxes, for example, that not only make people poorer but also provide an incentive to substitute away from consumption goods and into leisure, which is not taxed, lump-sum taxes are not distortionary and do not provide such a disincentive to work.

4 See the appendix for a technical explanation of this conclusion.

5 Doing so requires an adjustment in the value of parameter θ (see appendix) to ensure that the amount of hours worked on average is 30 percent.

6 For simplicity, we have assumed that the government has a pile of projects in a drawer and is exempted from having to search for goods (a trivial simplification). Also, it is very easy to model similar search frictions for the government by being explicit about the part of measured government expenditures that are used for search purposes rather than providing strictly useful goods.

7 Both because people are poorer and because the reduction in investment increases the rate of return, propelling a wealth effect and a (intertemporal) substitution effect.

8 “Time to build” is an expression used by Kydland and Prescott (1982), referring to multiperiod construction as a fundamental characteristic of most economies and one that helps explain macroeconomic fluctuations. As they put it: “That wine is not made in a day has long been recognized by economists.”

9 Let’s for now ignore the fact that in this type of recession, a benevolent government should do nothing: The economy is optimal on its own.

10 Another type of recession, one due to a short-lived shock that reduces the household’s relative preference for current consumption. It turns out that a recession of 1 percent in output via a reduction in the willingness to consume generates wild oscillations in consumption and investment. In the SNM with the high Frisch elasticity of 1.1, consumption drops by half(!) and investment goes up two and a half times. The behavior of the shopping economy is more subdued: Consumption falls from 65 percent of GDP to 59.9 percent, and investment goes from 19 percent to 23.1 percent. These recessions are clearly uninteresting, and for what it is worth, the multipliers are not very different from those in normal times: 0.030 in the SNM and 0.195 in the shopping economy.

References

Bai, Y., J.-V. Ríos-Rull and K. Storesletten. 2011. Demand Shocks as Productivity Shocks. Working paper, University of Minnesota.

Chetty, R., A. Guren, D. Manoli and A. Weber. 2011. Are Micro and Macro Labor Supply Elasticities Consistent? A Review of Evidence on the Intensive and Extensive Margins. American Economic Review Papers & Proceedings 101 (3): 471-75.

Christiano, L., M. Eichenbaum and S. Rebelo. 2010. When Is the Government Spending Multiplier Large? Mimeo, Northwestern University.

Drautzburg, T., and H. Uhlig. 2011. Fiscal Stimulus and Distortionary Taxation. Working Paper 17111, National Bureau of Economic Research.

Dyrda, S., G. Kaplan and J.-V. Ríos-Rull. 2011. Business Cycles and Household Formation: The Micro versus the Macro Labor Elasticity. Mimeo, University of Minnesota.

Dyrda, S., and J.-V. Ríos-Rull. 2012. Expansionary Policies when Demand Increases Productivity. Mimeo, University of Minnesota.

Hall, R. 2009. By How Much Does GDP Rise if the Government Buys More Output? Brookings Papers on Economic Activity 2: 183-231.

Heathcote, J., K. Storesletten and G. Violante. 2010. The Macroeconomic Implications of Rising Wage Inequality in the United States. Journal of Political Economy 118 (4): 681-722.

Kydland, F. E., and E. C. Prescott. 1982. Time to Build and Aggregate Fluctuations. Econometrica 50 (6): 1345-70.

Leeper, E. M., N. Traum and T. B. Walker. 2011. Clearing Up the Fiscal Multiplier Morass. Working Paper 17444, National Bureau of Economic Research.

Smets, F., and R. Wouters. 2003. An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area. Journal of the European Economic Association 1(5): 1123-75.

Abstract

In this paper, we explore the impact of the Great Recession on economic inequality and redistribution in the United States. We analyze many sorts of inequality (in earnings, disposable income, consumption expenditures and wealth) for different sections of the economic distribution.

Here we highlight three central findings.

• In 2010, the bottom 20 percent of the U.S. earnings distribution was doing much worse, relative to the median, than in the entire postwar period. This is because their earnings (including wages, salaries, and business and farm income) fell by about 30 percent relative to the median over the course of the recession. This lowest quintile also did poorly in terms of wealth, which declined about 40 percent.


• Redistribution through taxes and transfer programs reached historically high levels in 2010. As a result, spending power, captured by disposable income and consumption expenditures on nondurables, of this same lowest 20 percent did not significantly change relative to other economic groups during the recession.


• Although government redistribution protected households from fully bearing the impact of an earnings decline, households that experienced such a decrease nonetheless endured sizable drops in disposable income and drops in consumption expenditures.

Introduction¹

Although there is little doubt that the Great Recession constituted a watershed for overall business cycle dynamics in the United States, the jury is still out on its distributional consequences. Did economic inequality change significantly during the recession? If so, which dimensions—income earnings, wealth and consumption—saw the largest changes? And what impact did government policies, such as taxes and transfer programs, have over this time period on both inequality and economic well-being?

Analyses focused on the first two years of the downturn seem to find no increase in economic inequality; indeed, some report a decline. For example, a recent comprehensive volume (Jenkins et al. 2011) that analyzes income distribution in 21 Organisation for Economic Co-operation and Development (OECD) countries (including the United States) across the Great Recession sees “little change in household income distributions in the two years following the downturn.” Heathcote et al. (2010b) and Petev et al. (2011) study inequality in consumption expenditures in the United States up until 2009 and also find little change (if anything, they find a decline).

A longer-term view, however, suggests that high levels of unemployment and the large drop in housing prices, both of which started during the Great Recession but persisted well after, might have had longer-term adverse distributional consequences. In particular, the recession may have left a significant fraction of the U.S. population with very little wealth (due to the fall in asset prices) and poor labor market prospects (due to high unemployment).

The goal of this paper is to paint a more complete picture of the distributional impact of the Great Recession, including more recent data from 2010 and part of 2011. Most importantly, this paper considers inequality in a wide array of variables, such as earnings, disposable income, consumption expenditures and wealth, and looks at inequality for all of these variables at different sections of the economic distribution.

Our first finding is that during and after the Great Recession, the bottom of the U.S. earnings distribution has fallen dramatically. This is the result of historically high unemployment and nonparticipation. In terms of earnings, the bottom 20 percent of the U.S. population has never done so poorly, relative to the median, during the whole postwar period. We also show that this group experienced rapidly declining wealth.

Despite this, we find that inequality in disposable income and consumption did not increase at either the top or bottom of the distribution, confirming the findings of other studies. In other words, the same bottom 20 percent of the earnings distribution that fared so poorly during the Great Recession in terms of earnings and wealth is in pretty much the same relative position in terms of disposable income and consumption in 2010, after the recession officially ended, as it was in 2006, before the start of the recession.

Such a divergence of trends in earnings and disposable income at the bottom of the distribution is unprecedented in U.S. history, and we show that it is mainly due to government transfers and taxes, as opposed to private components of unearned income.

We conclude our study using panel analysis (i.e., following a specific set of households through time) to better assess the role of government taxes and transfers. This allows us to distinguish between the experience of a given section of the income distribution (e.g., the bottom 20 percent of the distribution, whose members change each period) and the experience of a fixed group of households (e.g., those households that were at the bottom 20 percent of the distribution in 2006 but whose position may have changed by 2010. If the “Smiths,” say, were in the bottom fifth in 2006, we use panel analysis to understand where the Smiths ended up later on).

Our main finding is that although the bottom 20 percent of the earnings distribution experienced constant disposable income or consumption expenditures despite earnings losses, individual households that face earnings losses and enter the bottom 20 percent group do suffer significant losses in disposable income and small losses in consumption.

Our main substantive conclusion is that government redistribution in the Great Recession was at historical highs and partially shielded households from experiencing large declines in disposable income and consumption expenditures. The same households, though, have experienced losses in net wealth, and this might make them more vulnerable to further or more persistent earnings declines in the future.

We believe our analysis provides useful data to inform the policy debate about whether or not, looking forward, the government should take a more aggressive role in providing assistance for households that experience earnings losses.

Income inequality in U.S. recessions: Some historical perspective

We start our analysis by putting the Great Recession in historical perspective, in particular by comparing the patterns of income inequality in the Great Recession with patterns of inequality in previous recessions.

For this analysis, our data source is the March supplement of the Current Population Survey (CPS), an annual survey of about 60,000 households selected to represent the U.S. civilian noninstitutional population. The longest series that is comparable, for the purpose of our analysis, starts with the March 1968 sample (which refers to 1967 calendar year) and ends with the survey collected in March 2011, which covers incomes of the calendar year 2010. Because of our interest in the recession, which mostly affected households in labor markets, we select only those households with at least one member between the ages of 22 and 60 years.

Throughout this paper, we focus mostly on two simple measures of inequality: the 50/20 ratio and the 95/50 ratio. These are ratios of percentiles in the economic distribution. For example, the 50/20 ratio for income is the ratio of median income (the “50”) to the income of the richest household in the bottom fifth of the income distribution (the “20”). The 95/20 ratio for earnings is the ratio of the lowest-earnings household in the top 5 percent of the earnings distribution (the “95”) to the median earnings figure.

These ratios have two advantages over other inequality indicators. First, as ratios of variables, they are easy to translate directly into inequality magnitudes and inequality changes. The second advantage is that they concisely capture inequality at the bottom and at the top of the distribution, respectively.

In terms of income measures, we first focus on three measures of household resources. The first is earnings, which includes wages, salaries, and business and farm income from all household members. The second is total income, which includes all sources of household income, including not only earnings, but also interest, dividends, rents, private transfers (such as alimony and child support) and government transfers (such as Social Security, unemployment insurance and welfare). The last measure is disposable income, which subtracts tax liabilities from total income.² To account for different household sizes, we divide all three measures of household income by the number of “adult equivalents” in the household.³

Figures 1 and 2 report the evolution, from 1967 to 2010, of the 95/50 ratio and of the 50/20 ratio for these three measures of household resources.

Large Image

Large Image

Figure 1 confirms the finding, highlighted by several authors (e.g., Piketty and Saez 2003), that inequality at the top of the distribution has increased substantially during the 1980s and the 1990s. It also shows that there was an increase in earnings inequality during the Great Recession. This increase was due to the fact that median earnings per adult equivalent fell quite substantially from 2008 to 2010 (from around $26,300 to roughly $24,700), while earnings of the 95th percentile have been more stable (about $89,170 to $88,640).

Notice also that the increase in earnings inequality resulted in a rise in inequality in total income but not disposable income, suggesting that taxes reduced the differential impact of the recession on the top and on the median. Overall, though, the changes in inequality at the top of the distribution were small compared with changes at the bottom, as seen in Figure 2.

The first feature of Figure 2 that we want to highlight is an extraordinary fall of the bottom of the earnings distribution. This is captured by the 50/20 ratio, which increased sharply during the recession from roughly 2.7 to nearly 3.5. Note that the 50/20 ratio rises in all recessions, which are, by definition, periods of increasing unemployment. Higher unemployment raises the fraction of households with no or very low earnings, and this causes the 20th percentile of the earnings distribution to fall relative to the median, thereby raising the 50/20 ratio.⁴

But note that while unemployment in 2010 was slightly below its postwar historical high,⁵ the 50/20 ratio in the same year was well above its previous historical high, reaching almost 3.5, while in previous recessions it never exceeded 3. This suggests that the cause for the high inequality at the bottom is not just unemployment but also nonparticipation in the labor market.

A second feature seen in Figure 2 is that the 50/20 ratios in both total income and disposable income have much lower levels and, during the Great Recession, experienced much smaller increases than the 50/20 ratio in earnings. Indeed, despite the substantial increase in earnings inequality, inequality in disposable income was about the same in 2010 as in 2003.

This lack of change is quite remarkable; in all previous U.S. recessions, with the exception of that in 1973, disposable income inequality at the bottom increased. Constant inequality in disposable income during recessions has been experienced in some European countries (Sweden, for example; see Domeji and Flodén 2010), but it is unusual in the United States. This suggests that mechanisms like private or government transfers played an important role in mitigating the effect of the Great Recession on inequality in disposable income. The next section investigates the impact of such mechanisms in greater detail. 

Income inequality in the Great Recession: Getting to the bottom of it

First we look more closely into the large increase in earnings inequality at the bottom of the distribution, and then we identify more precisely the causes of the divergence between inequality in earnings and in disposable income.

Earnings are the product of hours worked and wages per hour, and the CPS provides data on hours worked per household. Following a similar analysis to that done by Heathcote et al. (2010a) for previous recessions, Figure 3 plots average real earnings, average hours worked and average disposable income for the bottom 20 percent and the middle 10 percent of the earnings distribution. In both panels, all statistics are “normalized” (or mathematically set) to 1 in 2008.

Large Image

Three features of this figure are quite striking.

The first is that the increase in earnings inequality was the result of a large absolute (and not relative) fall of earnings at the bottom of the distribution. The top panel shows that earnings at the bottom fell more than 30 percent (in real terms) from 2008 to 2010, while the bottom panel shows only a moderate 5 percent earnings fall for the middle.

The second is that the sharp fall in earnings at the bottom can be attributed largely to the decrease in total hours worked, which fell by 25 percent, and not to a possible change in hourly wage rates.

The third striking feature of this figure—evident in the top panel—is that the sharp fall in hours worked and earnings for the bottom 20 percent of households did not result in a parallel decline in disposable income.

Taken together, these three facts suggest that government and private support for unemployed individuals played a major role in muting the impact on disposable income of lower earnings and employment during the Great Recession.

In Table 1, we analyze more precisely which components of disposable income were most important in mitigating the fall in earnings of the bottom 20 percent. To do so, we look at the increase in the 50/20 ratio when we add to earnings each of the individual components that constitute “disposable income.”

Large Image

The table’s first column simply identifies each component, one by one. The second column reports the change (from 2006 to 2010) in the 50/20 ratio for the income measure that includes earnings plus the respective component in the first column. The third column is simply the second column minus the increase in the 50/20 ratio for earnings alone: The smaller the number, the more that particular disposable income component prevents a rise in inequality at the bottom. (For instance, Social Security payments, with a -0.08 impact, plays a greater role in preventing a rise in 50/20 disposable income inequality than does rental income, with a -0.04 impact.)

The rows are ranked according to the third column; the first row reduces the inequality increase the most. Not surprisingly, unemployment benefits contribute greatly to the lack of increase in disposable income inequality as do taxes (most likely through the Earned Income Tax Credit program). Overall, the table suggests that government programs, as opposed to other non-earned-income categories like interest or dividends, are an important factor in explaining why increased earnings inequality did not translate into an increase of disposable income inequality during the Great Recession.

Consumption inequality during the Great Recession

Previous research on inequality (e.g., Blundell and Preston (1998) and Krueger and Perri (2006)) has suggested that the distribution of consumption expenditures, not of income, gives greater insight into the distribution of household well-being. Financial markets permit consumption expenditures that are more closely related to a household’s lifetime resources (sometimes referred to as the “permanent income” of the household). Consumption, therefore, is a better indicator of the well-being of the household.

This logic might also be relevant in evaluating the distributional impact of the Great Recession, for two reasons.

One reason is that current consumption better reflects expectations about future income prospects than do current earnings—an individual who expects to lose his/her job may well reduce expenditures even when she/he is still employed.

We have just established that during the Great Recession, inequality in disposable income did not increase because government transfers like unemployment insurance supported disposable income of low-earnings households. But if shocks to earnings are persistent and transfers to low-income households have limited duration, then the permanent income of some low-earnings households will fall, and so we would expect to see a drop in consumption expenditure, despite stable disposable income.

The other reason consumption expenditures might be a better indicator of distributional changes is related to the fact that the defining event of the Great Recession was the large fall of asset prices, particularly of housing.

Consider two households with the same income but very different shocks to the value of their wealth. Looking only at income would not inform us about distributional changes between them, but looking at consumption would, as the households would adjust their consumption in response to changes in their net wealth. More concretely, when housing prices fall, households feel less wealthy and spend less—even when their salaries and other income streams do not change.

For these reasons, understanding the evolution of consumption distribution during the Great Recession may shed light on the impact of the recession. Here we present household-level consumption data from the Consumer Expenditure (CE) Interview Survey. The CE Survey is a rotating panel of households that are selected to be representative of the U.S. population. Each quarter the survey reports, for the cross section of households interviewed (about 6,000), detailed demographic characteristics for all household members, detailed information on consumption expenditures for the three-month period preceding the interview and information on income, hours worked and taxes paid over a yearly period. The most recent data available are from the first quarter of 2011.

The statistics we present track closely those analyzed earlier. We start with inequality at the top, captured by the 95/50 ratio, together with inequality in earnings and disposable income (from the CE sample). Figure 4 reports measures of inequality in expenditures on nondurable goods (labeled “Nondurable Consumption”) and inequality in expenditures on nondurables plus a few durables such as cars and furniture (labeled “Nondurable Consumption+”).⁶

Large Image

Overall, the figure suggests that despite some swings in the inequality measures, the Great Recession did not significantly change inequality in consumption for households at the top of the distribution.

Notice how, in the initial phase of the recession, inequality in both consumption measures seemed to fall; this might be simply due to the large fall in purchases of durables that took place in the middle of the recession. If many consumers stop purchasing durables, fewer large consumption expenditures are recorded and, hence, inequality at the top falls. Consistent with this hypothesis is the fact that by the end of 2010, inequality in expenditures at the top returned to the same level as in 2006.

Figure 5 reports inequality at the bottom, the 50/20 ratio. The plot lines for earnings and disposable income mimic closely those observed for CPS and, perhaps not surprisingly, given the steady path of disposable income over these years, inequality in consumption barely moves during the Great Recession. Overall, Figures 4 and 5 suggest an overall stability of consumption inequality over the course of the Great Recession.

Large Image

In the last part of this section, we investigate the issue further by focusing, as we did in Figure 3, on the bottom 20 percent of the earnings distribution. Figure 6 plots, for the bottom 20 percent of the earnings distribution in each quarter, average earnings and disposable income, average nondurable and nondurable+ consumption and average total net wealth, all normalized to 1 in 2007q4.

Large Image

First, notice that earnings and disposable income behave very similarly to the corresponding CPS series in Figure 3. Nondurable consumption tracks disposable income closely and does not seem affected by the fall in earnings. In contrast, both the average wealth and the nondurable+ series for this group fall considerably.

It would no doubt be difficult—and certainly it is outside the scope of this paper—to establish a causal link between the fall in earnings and in wealth of this group. Nevertheless, the figure suggests that the bottom 20 percent of the earnings distribution in 2010 was a very different group than it was in 2007. The bottom 20 percent groups in 2010 and 2007 had the same disposable income, but both earnings and wealth of the 2010 group were 40 percent lower. In absolute terms, this means that the average wealth of the bottom 20 percent fell from around $80,000 in 2007 to a little over $50,000 in 2010. The lower wealth is particularly important, as it makes this group more vulnerable if government support for low-earnings households were to cease.

A panel analysis

The cross-sectional data analyzed in the previous sections show that while earnings for the bottom 20 percent of households fell dramatically over the Great Recession, disposable income in the same group was virtually constant. However, these cross-sectional data do not necessarily tell us how individual households are faring over time, since the group of households in the bottom 20 percent changes each year; some previously higher-earnings households move into the bottom 20 percent, and some households that were previously in the bottom 20 percent move out of it.

In this section, we use panel data from the Panel Study of Income Dynamics (PSID) to study the importance of two components of the bottom 20 percent group: (1) changes in income and expenditures of households that stay in the group, and (2) changes in the composition of the group.⁷

The PSID is the longest-running representative household panel study in the United States. The PSID data sets provide a wide variety of information on geographic location, income, employment, wealth and expenditures for many households that are followed, after 1996, at a biannual frequency. We concentrate our analysis on the 2007 and 2009 surveys (which provide data on 2006 and 2008) to study distributional dynamics since the start of the Great Recession.

Panel (a) of Table 2 (labeled “Bottom 20 percent of earnings”) shows the results, for the PSID, of the same cross-sectional analysis we did above on the CPS and CE. In particular, we look at a series of statistics for a particular group: the bottom 20 percent of the earnings distribution.

Large Image

In Panel (a), we see that earnings for the bottom 20 percent rose slightly between 2006 and 2008, by about the same amount as in the cross-sectional data in Figure 3. Disposable income rose as well. Panels (b), (c) and (d) of the table decompose, or separate, these changes into changes of the households that stay in the group (“Stayers,” Panel d), which make up about 75 percent of the group, and the difference in income between the group that entered the bottom 20 percent in 2008 (“In-switchers,” Panel b) and the group that left the bottom 20 percent in 2008 (“Out-switchers,” Panel c).

What we want to highlight is that while disposable income for the bottom 20 percent as a whole rose slightly between 2006 and 2008, this increase is due only to a change in the group’s composition. The panel analysis allows us to control for these compositional effects and shows instead that, on average, households in the bottom 20 percent of the earnings distribution in 2008 actually suffered a significant decline in disposable income.

Putting this in actual dollar figures might make the point still clearer: While the bottom 20 percent of households experienced an average disposable income increase of $228, this was simply because the 2008 income of those shifting into the group was higher than the 2006 income of those who left the bottom earnings quintile. Households in the bottom 20 percent in both 2006 and 2008 had a $159 disposable income decline, on average, and those who experienced such a severe drop in earnings that they moved to the bottom quintile had, on average, $12,236 less income to spend in 2008 than they had two years earlier.

Consumption of in-switchers also fell slightly, although interestingly, the drop in consumption has not been as large as the drop in disposable income. The relatively high average wealth of in-switchers ($70,146 in 2006) may have provided sufficient resources for them to smooth their consumption.

We find this panel analysis very instructive, as it reveals that looking at simple cross-sectional measures of inequality is not enough to assess the full distributional impact of the Great Recession. The panel analysis suggests that although government redistribution policies—taxes, unemployment insurance and others—have provided an important cushion against the effect of earnings declines on disposable income and consumption, they have not fully shielded households’ disposable income from these earnings fluctuations.

This further suggests that the Great Recession could have indeed had major redistributive effects at the bottom of the distribution. As panel data become available on the 2009-10 period, in which earnings of the bottom 20 percent fell dramatically (seen in Figure 3), it will be especially important to monitor the disposable income and consumption of households that moved into the bottom 20 percent in 2008 and remained there for the remainder of the Great Recession. For those that remain in the bottom 20 percent, their depleted wealth may not have been enough to prevent persistently low earnings from impacting consumption and welfare.

Conclusions

This paper provides an empirical analysis of inequality and redistribution during the Great Recession.

On one hand, we find that redistribution (through taxes and transfers) from high-earnings to low-earnings households in the United States was at its historical high, which possibly explains the calls by some for cutbacks in government programs that provide such assistance. On the other hand, we provide evidence that households that experience a severe earnings loss also face a large loss in disposable income and a loss in consumption, and that low-earnings households have become, during the course of the Great Recession, more vulnerable due to large losses in wealth.

This analysis should help inform future policy action regarding the extent of social insurance. For example, it could assist in assessing the consequences of extending, or curtailing, the duration of unemployment insurance benefits.

Endnotes

1 The authors thank Doug Clement and Kei-Mu Yi for valuable comments.

2 The CPS does not provide data for disposable income for all years in the sample. Therefore, we compute disposable income figures with TAXSIM, a widely used tax simulation program provided by the National Bureau of Economic Research. In years for which we have disposable income from the CPS, summary measures of disposable income in the CPS are very similar to our measures.

3 Following the commonly used OECD scale, the number of “adult equivalents” in a household is a weighted sum of household members in which the first adult is given a weight of 1, each additional adult has a weight of 0.7 and each member under the age of 17 has a weight of 0.5.

4 For more on how unemployment affects the dynamics of inequality over the business cycles, see Castañeda et al. (1998).

5 The U.S. unemployment rate in 2010 was 9.6 percent, just under the postwar high of 9.7 percent in 1982.

6 Specifically, the nondurable expenditures category includes expenditures on food and beverages, utilities and fuels, education, medical supplies, clothing and personal care, reading and transportation services. The nondurable+ category adds to this purchases of cars, furniture, jewelry and durable entertainment goods.

7 To see this more precisely, let’s define the following relation:

$$\overline{Y}_{B20}(t) - \overline{Y}_{B20}(t-1) = \alpha\left(\overline{Y}_{B20}^{Stay}(t) - \overline{Y}_{B20}^{Stay}(t-1)\right)$$
$$+ (1-\alpha)\left(\overline{Y}_{B20}^{In}(t) - \overline{Y}_{B20}^{Out}(t-1)\right)$$
where ~\overline{Y}_{B20}(t)~ is the average income measure of the bottom 20 percent of the earnings distribution in period t, ~\alpha~ represents the share of households that stay in the bottom 20 percent, ~\overline{Y}_{B20}^{In}(t)~ is the average income of the households that enter the bottom 20 percent at time t (and were not in the bottom 20 percent in period t-1) and ~\overline{Y}_{B20}^{Out}(t-1)~ is the average income of the households that were in the bottom 20 percent at t-1 and exited the group at time t. The equation highlights that observed changes in the cross-sectional data ~\overline{Y}_{B20}(t) - \overline{Y}_{B20}(t-1)~ are driven both by changes in income/expenditures of households that stay in the group (the term ~((\overline{Y}_{B20}^{Stay}(t) - \overline{Y}_{B20}^{Stay}(t-1)))~) and by changes in composition of the group (the term ~(\overline{Y}_{B20}^{In}(t) - \overline{Y}_{B20}^{Out}(t-1))~).

8 The 14 percent figure is derived by adding (a) the disposable income loss of the stayers (2 percent) times their numerical share of the group (75 percent) and (b) the disposable income loss of the in-switchers (50 percent) times their weight in the group (25 percent).

References

Blundell, R., and I. Preston. 1998. Consumption Inequality and Income Uncertainty. Quarterly Journal of Economics 113 (2): 603-40.

Castañeda, A., J. Díaz-Giménez and J. V. Ríos-Rull. 1998. Exploring the Income Distribution Business Cycle Dynamics. Journal of Monetary Economics 42 (August): 93-130.

Domeij, D., and M. Flodén. 2010. Inequality Trends in Sweden 1978-2004. Review of Economic Dynamics 13 (1):179-208.

Heathcote, J., F. Perri and G. Violante. 2010a. Unequal We Stand: An Empirical Analysis of Economic Inequality in the United States, 1967–2006. Review of Economic Dynamics 13 (1): 15-51.

Heathcote, J., F. Perri and G. Violante. 2010b. Inequality in Times of Crisis: Lessons from the Past and a First look at the Current Recession. VoxEU.

Jenkins, S., A. Brandolini, J. Micklewright and B. Nolan. 2011. The Great Recession and the Distribution of Household Income. Working paper. Fondazione Rodolfo de Benedetti.

Krueger, D., and F. Perri. 2006. Does Income Inequality Lead to Consumption Inequality? Evidence and Theory. Review of Economic Studies 73 (March): 163-93.

Petev, I., L. Pistaferri and I. Saporta Eksten. 2011. Consumption and the Great Recession: An Analysis of Trends, Perceptions and Distributional Effects. In Analyses of the Great Recession, D. Grusky, B. Western and C. Wimer (eds.) forthcoming.

Piketty, T., and E. Saez. 2003. Income Inequality in the United States, 1913-1998. Quarterly Journal of Economics 118 (1): 1-39.

National policymakers have long been interested in technological innovation and its potential contribution to economic growth and improved well-being. The Obama administration has embraced innovation as “the foundation of American economic growth and national competitiveness.” In launching the “Strategy for American Innovation” in November 2010, the president remarked, “[T]he key to our prosperity … as it has always been—will be to compete by developing new products, by generating new industries, by maintaining our role as the world’s engine of scientific discovery and technological innovation.”²

Policies to encourage innovation by firms include government funding for research and development (R&D), direct and indirect subsidies, tax credits and other tax benefits, such as deductibility of research expenses. Other policies not typically thought of as aimed at stimulating innovation, such as the corporate profits tax, also impact firms’ decisions to innovate. But to channel support effectively, policymakers need to know which policies are most successful in spurring innovation at companies, given their fiscal cost to taxpayers, and to what extent the firm-level innovation induced by these policies truly generates broader economic growth. Also important to policy: What factors influence the effectiveness of innovation subsidies in promoting economic well-being over the long term?

In particular, the idea that innovative activity by firms has “spillovers” that promote the wider diffusion (intentional or inadvertent) of new innovations and knowledge created at just one location, firm or industry is central in justifying government subsidies for innovation. As a result, we want to know how important these spillovers are for the economy as a whole.

Economists who study these issues have generally followed two paths. The first is an effort to understand the impact of policy on innovation decisions taken by individual firms—the companies that develop new products and services or improve methods of production or service delivery. More relevant for overall economic well-being, the second looks at the macroeconomy, seeking to measure policy impact on a national level: To what extent do policies to encourage innovation generate broad economic growth? For the most part, these approaches have rarely intersected, leaving a significant gap in our knowledge of the mechanisms through which policy initiatives may or may not improve the economic well-being (or welfare) of Americans.

This policy paper reports on our effort to fill that gap by combining these two perspectives, the micro and the macro, thereby providing greater insight into several key considerations in innovation policy. (See “Aggregate Implications of Innovation Policy,” Minneapolis Fed Staff Report 459, revised September 30, 2011.) The overarching issue is: How do policies that affect firms’ innovation costs and benefits impact aggregate output, productivity and welfare?

To answer this question, we have designed an economic model that is sufficiently detailed to capture the dynamic decisions of individual firms in response to innovation policy changes, yet is still mathematically manageable, allowing us to aggregate these many firm-level decisions and thereby gauge overall policy impact at the economywide (or “aggregate”) level on output, productivity and economic well-being.

We break our analysis into two parts. In the first, we study which policies are most efficient in the long run in balancing their fiscal cost to taxpayers against their benefits in stimulating overall innovative activity by firms, including both those firms that are already operating when the policy is put in place and those that will enter under the new policy regime.

In the second part, we study how a policy-induced increase in innovative activity by firms impacts aggregate output, productivity and welfare (taking into account the fiscal cost to taxpayers of the policies used to stimulate that innovation) over both the long run and a medium-term horizon of 15 years.

Our research gives new answers to both of these questions.

Consider first the balance between the fiscal cost of various innovation policies and their effectiveness in stimulating innovative activities by firms. The standard analyses of fiscal efficiency of innovation policies attempt to fathom the many intricate details of the effects of the new policies on individual firms’ decisions about changing investments.³ By contrast, in our research, we embed a model of firms’ innovation decisions in an overall model of the macroeconomy and show that with such a model, we can estimate the policy effectiveness of stimulating innovative activity simply by calculating the approximate impact of that policy on the profitability of new firms that might enter under this new policy.

Moreover, our results imply that, under some conditions, a broad set of innovation policies are all equally efficient: Two policies have the same impact on aggregate innovation, output and productivity in the long run if they have the same fiscal impact on taxpayers. These results provide a simpler procedure for evaluating the effectiveness of innovation policy and other government efforts to stimulate innovative activity and suggest that policies currently in place to stimulate innovation might also be dramatically simplified.

To gauge policy strength in encouraging innovative investments by firms, we focus on the size of policy impact on the potential profitability of new firms. By doing so, our research implies that innovation subsidies and tax preferences are only part of a much broader set of government policies with both positive and negative effects on firms’ incentives to innovate. In particular, the negative impact of the corporate profits tax on incentives to innovate through its impact on potential profitability of new firms⁴ may very well undo, at the aggregate level, much of the benefit of current direct federal support for R&D. In this respect, our research indicates that in terms of their effect on innovation investment, the current mix of federal subsidy and tax policies may negate with one policy the impact of others.

On the second question of how a policy-induced increase in innovative activity by firms affects long-run aggregate output, productivity and welfare (taking into account the transition from the status quo to the long run), our research indicates that it may be very difficult to reach definitive conclusions, given available data. Our model predicts on the one hand that a policy-induced change in innovative activity by firms may have a very large impact on output and productivity in the long run and on welfare, particularly if the spillovers from innovative activity are large. On the other hand, if spillovers are small, this may not be the case (and policies to stimulate innovative activity may not raise economic well-being).

Our research indicates that analysts may not be able to distinguish between those divergent long-run outcomes because our model’s predictions for the macroeconomic response to an innovation policy over a reasonable time horizon, such as 15 years, look quite similar whether spillovers are very large or very small. Our model’s simulations of the economy’s medium-term response to a significant increase in innovation subsidies suggest that analysts working with real-world data would have difficulty obtaining reliable estimates of the magnitude of innovation spillovers for the economy as a whole and hence the implications of actual policy changes for welfare.⁵

Our research approach

To analyze the micro- and macroeconomic dynamics of innovation policy, we’ve built an economic model that is rich, yet tractable. By this, we mean that it combines the fundamental and detailed elements of innovation processes at a company level, but nonetheless allows us to generate estimates of the overall national economic impact of these firm-by-firm decisions as influenced by changes in government innovation policy. This policy paper describes our model and research in broad terms, shares the analytical and quantitative insights we’ve gained and then discusses implications of these findings for both research and policy.

Our research approach and this paper consist of several steps.

• First, we build a model that enables us, with a two-stage procedure, to assess the impact of changes in firm-directed innovation policies on macroeconomic output, productivity and economic well-being.
• We then use this procedure to establish several analytical results about the long-run response of a macroeconomy, through its microeconomic units, to innovation policy change.
• These results allow us, in a third step, to assess the relative and absolute size of the medium-run and long-run macroeconomic impact generated by several distinct real-world innovation policy options.
• These findings imply several directions for future policy, discussed briefly in a final section.

Our model

We use a dynamic general equilibrium model common to macroeconomic research. It includes households that work and consume according to their preferences and budget constraints, firms that invest and produce with specified technologies and objectives, and a government that has objectives, revenues and expenditures.

To this standard framework, we introduce a number of special features that allow us to analyze macroeconomic (also referred to as “aggregate”) implications of innovation policy.

We build a model of monopolistically competitive⁶ firms that engage in either process innovation, which will increase their productivity (a more efficient assembly line, as a mundane example), or product innovation, which enables them to create a new type of product (an iPad, if they’re very lucky and skilled). More simply put, firms can invest in R&D to become more productive or expand the range of goods available to consumers.

To consider the impact of policies on the cost of innovative activity, our model includes a research good that firms use as an input for innovation. Research goods aren’t unlimited. Producing them requires a combination of goods and labor (lab equipment and scientists) along with scientific knowledge that is freely available. Also, most crucially for purposes of our analysis, the production of the research good benefits from innovation spillovers—the knowledge and experience that researchers gain through innovation activities that neither they nor their company directly capture (at least financially). Through these spillovers, current innovative investments by firms have an added benefit to society of increasing the productivity of future R&D workers and thus reducing the cost to firms of future innovation.

In our model, we consider the impact of a range of subsidies financed by taxes collected from households and equaling aggregate fiscal expenditures by government. These subsidies—fairly abstract when we derive analytical results and later made concrete in our quantitative estimates—include a subsidy to variable profits from production, a subsidy to process innovation and a subsidy to product innovation. In addition, firms are taxed on their use of physical capital, essentially a negative subsidy.

While this brief description hardly does justice to a relatively complex model, it provides a sense of the key features that allow us to analyze the impact of innovation policy on both individual firms and the macroeconomy.

A two-step procedure

Detailed examination of the interaction of these features and the more standard variables in our model yields insight into what is (and isn’t) fundamental to analysis of the macroeconomic impact of changes in innovation policy. We discovered that a relatively straightforward procedure—a two-step algorithm, or sequence of formulas in which results from the first are inputs to the second—would provide approximate estimates of the long-run impact on macroeconomic outcomes of changes in innovation policy and thereby enable us to compare the relative and absolute magnitudes of the impact of various policy alternatives.

The first step in this procedure is using a basic formula to measure the impact of policy changes on the profits an entrepreneur might expect from starting a new firm.

The second step is to then use the model’s macroeconomic structure to infer long-run changes in aggregate output and wages that must result, in general equilibrium, to restore the incentives of entrepreneurs to create new firms or products in the face of the estimated change in expected firm profitability calculated in the first step.

In other words, the procedure gives us estimates of the new long-term level of macroeconomic outcomes that corresponds to whatever change in firm profits is generated by a new government innovation policy. And it does so without having to fathom the many intricate details of the new policy’s effect on firms’ decisions about changing investments, hiring, corporate structure and the like. We need only compute how the policy changes firm profitability—a far easier task.

This straightforward procedure (and the reasoning behind it) allows us to analyze more fully the implications of innovation policy changes. We do so in the next section, followed by an examination of the quantitative application of the procedure.

Analytical results

A central insight offered by our model and the algorithm just described is that a subsidy to all types of innovative activity has the same impact on macroeconomic outcomes as a direct subsidy to firm profitability. The reasoning is quite intuitive. Subsidizing a firm’s innovative activities—in this case, by changing the price of the research good with a uniform subsidy to process and product innovation—lowers its costs, or equivalently, raises profits. Since profits here are the returns to innovation, supporting firm innovation through a subsidy has an identical impact on firm behavior and aggregates as a direct subsidy to firm profits.

We also find that, under some conditions, whether the subsidy is directed toward process or product innovation makes little difference in computing the effect on the macroeconomy as long as the impact on firm profitability is the same; this is because of dynamics that ensure that in macroeconomic equilibrium, with free entry of firms, companies will start up in an industry until doing so would no longer offer profits to entrepreneurs. (A policy directed specifically at either process or product innovation may have a dramatic impact on firm-level behavior, however, particularly on the innovative investments of existing firms.)

The zero-profit condition for entrepreneurs considering starting firms in a given industry limits the aggregate response of innovative investments by both existing firms and entrepreneurs contemplating a startup venture. This analytic insight is what is distinctive about our method for measuring the response of firms’ innovative investments to a change in policy.

Previous research has often focused on the innovative response of existing firms only and neglected to consider that—in the long run, in general equilibrium—the zero-profit condition for entrepreneurs creating new products is key to assessing the overall response of the economy to the policy change. With this analytical insight, we argue that regardless of how existing firm investments react to specific subsidies, the response of the macroeconomy will be the same.⁷

In terms of policy, this implies further that, as we alluded to earlier, the details of firms’ responses to changes in innovation policy are not of great importance for aggregate outcomes; beyond pure subsidization of profits, there is no special role for innovation policies. An example clarifies the implications of this argument. Consider the current design of the Research & Experimentation Tax Credit. This innovation policy sets out a complex set of rules by which a firm can gain a corporate tax credit for “qualifying research and experimentation expenditures” over and above a defined “baseline amount.” The underlying idea is to reward existing firms only for new or incremental investments in innovation and to avoid subsidizing firms for innovation they would have done anyway.

Our research indicates that this policy focus on incremental expenditures at the firm level is misguided, since the impact on existing firms’ investment is not the factor that determines the impact of the policy in the long run. Instead, it is the impact of the tax credit on the incentives of entrepreneurs to start new firms or introduce new products. Our results imply that the Research & Experimentation Tax Credit is, therefore, an administratively expensive way of offering a small reward to entrepreneurs who consider starting a new firm and spending money on R&D that qualifies for the credit somewhere down the line as their new firm grows. It would be more straightforward (and more efficient in terms of administrative costs) to subsidize firms in the relevant industry directly.

Quantitative results

These analytical results lead to the question of magnitude. How can we measure the effectiveness of various innovation policies in stimulating innovative investments given their fiscal cost to taxpayers? And how can we measure the impact of this induced innovation on aggregate productivity, output and welfare? We conducted two sorts of quantitative analyses. The first measured the relative impact of several innovation policy options. The second calculated the absolute size of the economic effect of parallel policy options.

Comparison of relative policy impact

To understand the effect of innovation policy on broad economic growth and welfare and to evaluate the relative efficacy of different policy options, a means of quantifying and comparing financial cause and effect—that is, cost and benefit—is essential. With our model, we show that, to a first-order (or ballpark) approximation, the relative impact of a policy change on firm profitability and on macroeconomic aggregates in the long run is proportional to the impact of the policy change on government fiscal expenditure.

In other words, to compare, roughly, how large an impact alternative innovation policy options will have both at the level of firm profits and on broad economic outcomes in the long run—GDP and productivity—we need only calculate how much that policy costs. The two figures aren’t equal, just proportional, and the calculation is only a rough estimate, not a precise figure. But it means that to evaluate the relative merits of alternative policy options, we need only know their fiscal impact; the difficult task of gauging how millions of firms will respond to the policy isn’t necessary.

To apply our results to actual policies in the United States, we looked at (1) the Research and Experimentation Tax Credit program, (2) federal spending on research and development and (3) the corporate profits tax. (Beyond the well-understood effects of the corporate profits tax on investments in physical capital, the tax influences innovation decisions in two ways: It affects variable after-tax profits generated from improved products or process, and firms may expense a portion of the cost of innovative activity and thus deduct these expenses from taxable profits. To the extent that firms are not able to fully deduct all of their expenses for innovation or are not able to carry forward all of the loss when attempts at innovation are unsuccessful, the net effect of the corporate profits tax is to reduce the profitability of starting a new firm or introducing a new product.)

Data from 2007 indicate that fiscal expenditure on the Research and Experimentation Tax Credit was $10 billion. In the same year, federal spending on the five categories grouped into R&D by the Office of Management and Budget—basic research, applied research, development, R&D equipment and R&D facilities—totaled $139 billion. (In contrast, business R&D spending in 2007 was far higher, about $260 billion.) Comparing these two figures (and applying the appropriate discount factor since subsidies to product innovation are paid upfront while variable profits are received in the future), we can clearly see that the long-term impact on aggregate output of federal R&D spending is far larger than the impact of the Research & Experimentation Tax Credit.

Calculating the impact of the corporate profits tax—which raised $445 billion in federal revenue in 2007—is more complicated because it depends on parameter values in a quantitative model that affect the physical capital-to-output ratio. But once parameters are chosen, we find that the long-run impact of the corporate profits tax (per dollar of revenue raised) exceeds that of innovation policies (per dollar spent) unless innovation spillovers are very large.

Hence, in our calibrated model, described below, reducing the corporate profits tax to collect $100 billion less in revenue would have a comparable or even larger impact on innovation spending and aggregate output in the long run than increasing either the Research & Experimentation Tax Credit or federal R&D spending by $100 billion, unless spillovers are very high. We thus conclude that the corporate profits tax may very well be a relatively potent, counterproductive policy in terms of discouraging the long-run accumulation by firms of both physical, tangible capital and intangible capital (that is, patents, trademarks, intellectual property and the like).

Comparison of absolute magnitude of policy impact

In a second quantitative exercise, we evaluated the absolute magnitude of both the long-run and medium-term impact on the macroeconomy of innovation policies after putting some concrete figures into our model, giving it further realism by providing reasonable values for parameters such as the GDP growth rate, interest rate and capital depreciation. With this calibrated model, we measured the absolute magnitude of impact on GDP, welfare, productivity, research intensity and other economywide outcomes of two policies:

1. A uniform subsidy to innovative activities (meaning that both process and product innovation would receive support).
2. A subsidy to process innovation only.

In each case, the subsidy represented a fiscal expenditure of 3 percent of GDP, or about $420 billion in 2007 (similar to the revenue raised from corporate profits taxes that year). These are two typical policies aimed at stimulating innovation.⁸

Long-run response

In the long run, we find, innovation policies have an impact on the scale of firms’ investments in innovation similar in magnitude to their fiscal impact, both relative to the level of GDP. Specifically, the research intensity of the economy (defined as the ratio of firms’ spending on innovative activities to GDP) increases by roughly 3 percentage points of GDP in response to a subsidy of 3 percent of GDP. Moreover, this response of firms’ innovative activity to innovation policy change is the same in the long run and roughly the same in the medium term regardless of the level of spillovers from innovative activity.⁹

Do these policies aimed at stimulating innovation increase consumers’ welfare? The answer to this question is not obvious. At first glance, it appears that such a policy might not increase economic well-being—the taxes that consumers must pay to finance these innovation subsidies are roughly the same as the increase in firms’ investments in innovation that result. In the absence of spillovers from firms’ innovative activity, a policy of taxing households to pay for firms’ investments in intangible capital is not likely to improve households’ well-being. In the presence of spillovers, however, such a policy might bring substantial welfare benefits.

Our model confirms this logic. We found that none of the subsidies has significant impact on economic welfare if innovation spillovers are small. Output and productivity rise in the long run (and perhaps by a lot), but this increase comes at the cost of inefficiently high investments in innovation and low consumption by households in the transition from the present to the long run.

If spillovers are large, however, the subsidies have far greater impact on economic well-being. In fact, in this case, innovation subsidies of 3 percent of GDP can bring huge gains for households. The numbers from our model simulations below illustrate this point. We measure improvements in household economic welfare from policy changes by the amount that household consumption would have to be increased each and every year under the old policy to make households as happy as they would be with the consumption they attain under the new policy.

When we set our parameter for innovation spillovers at zero, the impact of the innovation subsidies on welfare is very close to zero—consumers would be just as happy with or without the innovation policy. In contrast, when we set our parameter for innovation spillovers close to its maximum possible value consistent with balanced growth, the impact of innovation policies on welfare is very large. Consumption under the old policy would need to rise by roughly 50 percent every year to attain the same level of household welfare as achieved in the equilibrium with innovation subsidies. Welfare gains like these are why Nobel Prize-winning economist Robert E. Lucas Jr. wrote that once one starts thinking about long-run growth and economic development, “it is hard to think about anything else” (p. 5).¹⁰

Our results on the long-run impact of innovation policies on aggregate output and productivity are also highly sensitive to our assumption for the parameter governing spillovers from innovative activity. When we set our parameter for innovation spillovers to zero, GDP is estimated to increase by a factor of only 1.03 (that is, by 3 percent) in the long run. In other words, in this case, the subsidies have little impact on either output or welfare.

But when we set the spillover parameter close to the maximum value consistent with balanced growth, the impact on GDP is much larger: It increases by a factor of 9.88 for policy 1 and 8.25 for policy 2. These nearly tenfold changes in GDP are comparable to the growth that the United States experienced from the beginning to the end of the 20th century and are brought about by a substantial, but perfectly feasible, level of innovation subsidies.

Clearly, our model’s implications for the long-run impact of a given change in policies vary tremendously depending on the assumed spillover parameter. If spillovers are large, there is a lot at stake for consumers in getting innovation policy right.

Medium-term response

Our results on the impact of innovation policies on welfare and on output and productivity in the long run prompt the question: Can we use data on the response of the macroeconomy to changes in innovation policy over the medium term (say, 15 years) to figure out if spillovers from firms’ innovative activities are small or large?

There is a large literature that attempts to answer this question, but, as Griliches (1988) and CBO (2005) discuss, the changes in the innovation intensity of the U.S. economy seen in the historical data are relatively small. It is therefore difficult to distinguish the effects on the macroeconomy of such small changes in R&D spending from the effects of all the other major factors at play—education, population growth and international trade, to name a few.

To shed light on the question of whether we might be able to measure economywide spillovers from innovative activity using available data even if we were to observe a large change in the innovation intensity of the economy arising from a change in innovation policy, we examined how the model performed over a shorter time frame, a 15-year medium-term period. The idea here is to understand transition dynamics—between now and the long-term equilibrium, how does the economy evolve, and what factors are important in that evolution? Again, and for all policies, we use the same subsidy size: a fiscal expenditure of about 3 percent of GDP.

Surprisingly, perhaps, we found that over this time frame, the two innovation policy options have a similar impact on economic growth regardless of innovation spillover size. In all cases, the cumulative factor increase in GDP in the 15th year is between 1.01 (or 1 percent) with no spillovers and 1.05 (5 percent) with high spillovers. Such small differences in GDP over a 15-year horizon would likely be difficult to discern in real-world data. Therefore, our results indicate that data on the response of GDP to innovation policy changes over the medium term will not shed much light on the size of such spillovers, suggesting that estimating policy outcomes over the long term will remain difficult, since an accurate measure of spillovers can’t be obtained from shorter-term data.¹¹

What explains the significance of spillovers for welfare?

The contrast in findings between long- and medium-term significance of innovation spillovers raises the question of why spillovers would have importance on innovation’s macroeconomic impact only in the long run.

The intuition for this result is simply the idea of compound interest. Over the medium term, innovation policies have a similar impact on GDP growth regardless of the level of innovation spillovers. The real impact of spillovers comes only at longer time horizons. In the absence of spillovers, the boost to growth from innovation subsidies peters out relatively quickly and households are left paying roughly the same amount in taxes as the gain to innovation spending and the increase in GDP achieved. In contrast, if spillovers are large, the boost to the growth of GDP from increased investments in innovation lasts for a long time, well beyond the medium-term horizon, and innovation spillovers compound over time, bringing large benefits associated with a moderate boost to growth that lasts over 100 years.

Summary and implications for policy

We’ve established a benchmark model of innovation that provides a straightforward procedure for estimating relative magnitudes of long-run macroeconomic impact of a range of innovation policy options. The procedure gauges approximate policy impact on macroeconomic outcomes quite simply, through computing the government’s fiscal expenditure on innovation policies.

The response of economic welfare and GDP over the long run to changes in innovation policy is highly sensitive to the size of innovation spillovers; welfare gains could vary between virtually no change and a 50 percent increase in equivalent consumption, depending on spillover size.
Unfortunately, we cannot accurately measure these long-run effects without accurate estimates as to the magnitude of innovation spillovers. Results from our model indicate, however, that even under ideal conditions, it should be very difficult to measure spillovers using data on medium-term response of the macroeconomy from changes in innovation policy. That is, evidence from the medium term is not likely to help differentiate long-run effectiveness because all policies have similar medium-term outcomes regardless of the size of spillovers.

What does this imply for policy?

The clearest implication of our research is that to the extent that policymakers choose to subsidize innovative activity by firms, they should consider the full set of tax and regulatory policies that impact aggregate innovation through firm profitability. Taxing corporate profits or enacting regulations that make it more costly for firms to start up or operate has a significantly negative influence on innovation, undercutting the stimulative impact of R&D subsidization. The net effect may be to depress, rather than encourage, innovation by firms. 

Endnotes

1 This paper is based on: “Aggregate Implications of Innovation Policy,” Minneapolis Fed Staff Report 459, June 2011. The authors thank Doug Clement for assistance in preparing this text.

2 See “Strategy for American Innovation: Introduction.” Also see Chairman Ben Bernanke’s May 16, 2011, speech, “Promoting Research and Development: The Government’s Role,” for a discussion of the importance of innovation by firms to long-run growth and a summary of the questions regarding the rationale for, the effectiveness of and the impact of federal support for research and development that we address in this policy paper.

3 Specifically, following the methodology developed by Hall and Jorgenson (1976) for physical capital, a standard approach is to first estimate the impact of a policy change on the “user cost of R&D” and then estimate the elasticity of firms’ demand for R&D in response to such a policy-induced change in the user cost of R&D. See Hall and Van Reenen (2000) and CBO (2007) for examples of such analysis.

4 See Gentry and Hubbard (2000) and Cullen and Gordon (2007) for a discussion of the mechanisms through which the U.S. tax structure reduces the incentives of entrepreneurs to start new firms.

5 In this sense, our research casts doubt on the methods economists have previously used to measure the relationship between innovative activity by firms and aggregate productivity in the long run. See, for example, CBO (2005) and Hall, Mairesse and Mohnen (2009) for summaries of this research.

6 A monopolistically competitive market combines characteristics of competition and monopoly. There are many buyers and many firms, with free exit and entry into industries, as under perfect competition. But consumers perceive sufficiently great nonprice differences (branding, for example) among similar products that producers can exercise a degree of control over pricing, as in a monopoly. Brand-name cereals and restaurants are textbook cases; laptop computers might be another example.

7 Innovation policies in our model do impact the user cost of R&D and do have an impact on the innovative investments by incumbent firms that does depend on the responsiveness of these incumbent firms’ innovative investments to changes in the user cost of R&D. This responsiveness, or elasticity, of R&D investments is not of first-order importance, however, in the calculation of how a change in innovation policy affects the expected profitability of a new firm. For example, in calculating the impact on the expected profitability of a new firm from a change in a tax credit for R&D, what is of first-order importance is the change in taxes that a new firm can expect to pay given the investments in R&D that it had planned to undertake before the policy change was proposed. For small changes in policy, the additional accuracy gained by considering the impact on the expected profitability of new firms that arises from considering changes in policy and firms’ investments simultaneously is necessarily very small.

8 In the full paper, we also consider a third policy, a subsidy to physical capital to compare the impact of policies aimed at promoting firms’ investment in intangible capital and those promoting investment in physical or tangible capital. We make this comparison to analyze the impact of the corporate profits tax, which is a combination of taxes on firms’ profits from intangible and tangible capital.

9 Our findings here are consistent with those summarized by Hall and Van Reenen (2000) on the effectiveness of fiscal incentives for R&D.

10 See Lucas (1988).

11 We note that substantial research has sought to establish a link between research intensity and output or productivity. This research has generally used regression analysis of disaggregated data at the firm or industry level. Unfortunately, this evidence is less than conclusive for answering the questions addressed by this paper.

First, many of these results are driven by long-term differences across firms or industries: Firms and industries that invest more in R&D also appear to have higher levels of productivity. It is not clear, however, how to interpret this observation. Klette and Kortum (2004), for example, argue that it should be accounted for by models with intrinsic factors that vary across firms and industries and that it does not necessarily indicate that a policy of stimulating further R&D would have a substantial impact on the aggregate economy.

Second, even in our model, a policy aimed at stimulating innovative investments by a select group of firms or industries can have a large impact in the short term on output and productivity that suggests spillovers are high even if aggregate spillovers are absent. It is a simple matter for a subset of firms or industries to invest in innovation and grow at the expense of the other firms or industries in the economy. Therefore, evidence of specific firm or industry responses to policy changes does not necessarily shed light on the central question of the macroeconomic response.

References

Bernanke, Ben S. 2011. “Promoting Research and Development: The Government’s Role.” Speech at the Conference on “New Building Blocks for Jobs and Economic Growth.” Washington, D.C., May 16.

Congressional Budget Office (CBO). 2005. “R&D and Productivity Growth.” Background Paper, June.

Congressional Budget Office (CBO). 2007. “Federal Support for Research and Development.” A CBO Study, June.

Cullen, Julie Berry, and Roger H. Gordon. 2007. “Taxes and Entrepreneurial Risk-Taking: Theory and Evidence for the U.S.” Journal of Public Economics 91(7/8): 1479-1505.

Gentry, William M., and R. Glenn Hubbard. 2000. “Tax Policy and Entrepreneurial Entry.” American Economic Review 90 (2): 283-87.

Griliches, Zvi. 1988. “Productivity Puzzles and R&D: Another Nonexplanation.” Journal of Economic Perspectives 2 (4): 9-21.

Hall, Bronwyn H., Jacques Mairesse and Pierre Mohnen. 2009. “Measuring the Returns to R&D.” National Bureau of Economic Research Working Paper 15622.

Hall, Bronwyn H., and John Van Reenen. 2000. “How Effective Are Fiscal Incentives for R&D? A Review of the Evidence.” Research Policy 29 (4/5): 449-69.

Hall, Robert E., and Dale W. Jorgenson. 1976. “Tax Policy and Investment Behavior.” American Economic Review 57: 391-414.

Klette, Tor Jakob, and Samuel Kortum. 2004. “Innovating Firms and Aggregate Innovation.” Journal of Political Economy 112 (5): 986-1018.

Lucas, Robert E. Jr. 1988. “On the Mechanics of Economic Development.” Journal of Monetary Economics 22 (1): 3-42.

In 1950, U.S. Steel employed 30,000 workers at its Gary, Ind., plant, and Bethlehem Steel had a factory of similar size in Sparrows Point, Md. Ford’s massive Rouge River plant near Detroit employed even more workers—over 100,000 in the 1930s.

Things are far different today. Gigantic employer plants like these are virtually extinct in the United States. Indeed, as of 2007, only 47 plants with more than 5,000 workers exist, half as many as just 10 years earlier. To find massive-employer manufacturing plants, look to China. The Foxconn complex in Shenzhen where iPhones are assembled, for example, is credited in news reports with employing an astonishing 300,000 workers.

The decline of manufacturing in the United States has generated widespread concern and intense discussion about what government should do, if anything, to prevent (or even reverse) the painful downward trend. “The answer is to build things better, make things better, right here in the United States,” declared President Obama in 2010, as he signed the Manufacturing Enhancement Act.¹

Many Americans believe there is a close connection between the international competitiveness of the U.S. manufacturing sector and the nation’s ability to remain a prosperous country. A world where China sends container ships filled with manufactured goods to the United States effectively in exchange for U.S. Treasury notes is unsustainable in the long run. Manufacturing also relates to income distribution and inequality trends because it has long provided stable, well-paying jobs for blue-collar workers not skilled in high tech or high finance and ill-suited to design the next iPad or Wall Street innovation.

Unfortunately, most discussions of manufacturing employment trends lump together plants of all sizes, big and small. Obama noted in his 2010 speech, “Over the last decade, the manufacturing workforce shrank 33 percent.” While not inaccurate, such statistics can be misleading, because they obscure diverse trends within the manufacturing sector.

In this paper, I hope to illustrate this diversity by focusing specifically on what is happening at the top, to the large-employer plants: those with 1,000 or more employees. To do so, I use published government statistics in a rather novel way to track large employers over time, and since the number of these plants is declining rapidly, there is much activity in the data. Because the largest plants are more likely to be exporters and tend to pay higher wages, this focus on the biggest employers is particularly relevant for issues related to the trade deficit in manufacturing and trends in inequality.²

Before I go into details of the analysis, a broad overview that begins with a specific example might be helpful. Go back to the steel plant in Gary with 30,000 workers in 1950. The plant is still in operation, but according to Dun and Bradstreet, its current employment is down to 5,000. Remarkably, with one-sixth as many workers, the plant produces even more steel now than in 1950, as capacity has increased from 6 million to 7.5 million tons a year.³

This example of growth in labor productivity illustrates a general long-term trend of technological change and mechanization. One reason some large-employer plants have disappeared is that they have “downsized” into relatively smaller-employer plants, but remained steady or even “upsized” in output. At such plants, tasks once done by American workers are still being performed in the United States, but by machines instead of people. Of course, it’s also true that other plants are no longer on the large-employer list because they have closed outright and the work has shifted overseas.

In this paper, I take on the case of the disappearing large-employer manufacturing plants. In the end, I don’t find much that is mysterious. Many of the plants that disappear from “large-employer” status are simply dropping down to the next-lower size category. Yet there are also plenty of instances of dramatic employment decline or actual closure. To better understand these trends, I focus on specific industries hit hard by imports from China, including the apparel and furniture industries. And I focus especially on the Piedmont region in southeastern United States.

For most of the last century, the Piedmont played the same role relative to the industrialized Northeast and Midwest of the United States as China is playing today vis-à-vis the United States as whole. In the earlier period, labor-intensive factories in places like Pennsylvania and Michigan closed down and moved operations to North Carolina to take advantage of low wages. The Piedmont region ended up with huge factories employing large numbers of unskilled laborers in routine tasks.

Today, these large employers in the Piedmont are being closed at a disproportionately high rate compared with the rest of the country. Given their industry specializations, this turns out not to be a mystery. There is tremendous cost pressure to eliminate routine, labor-intensive tasks from manufacturing in the United States, where labor is relatively expensive, and everything I find is consistent with the power of this force.

This paper starts at a broad level—all of manufacturing—and successively narrows down. By the end, the discussion focuses on what is happening in just two furniture plants in the Midwest, including “nano-level” details about job postings. These are not simply two random plants pulled out of a hat for the sake of an anecdote. Rather, they are the two largest plants that have managed to survive in an industry otherwise decimated by Chinese imports. These two plants alone account for about 10 percent of all that is left of employment in their industry.

Large employers are interesting not only for all the “action” noted above, but also because they are disproportionately important as a source of jobs. Understanding the nitty-gritty about just a few large plants can therefore provide information that is quantitatively important for the industry as a whole. Readers will see that these two Midwest plants are full of white-collar workers and so, ultimately, it will be no mystery why these plants have survived, while the Piedmont plants, once filled with thousands of blue-collar workers, are gone.

Matching plants over time

To track large employers, I use public data from the Census of Manufactures taken by the U.S. Census Bureau every five years. The Census publishes a tabulation of the number of plants at each location and industry in various employment size ranges, such as “2,500 and more employees,” “1,000-2,499 employees” and so forth.⁴ From these data, I determine the list of all plants in the 1997 Census of Manufactures with 1,000 or more employees and define these as “large employers.” I then go 10 years forward to 2007 and look for a match in the same location and industry. The appendix describes the matching algorithm in detail.

For smaller employers it would be difficult or impossible to match specific plants over time, because business starts and closures (entry and exit) are so common. A restaurant reported in the 1997 Census in a particular location with 1-19 employees might be the same restaurant observed in the 2007 Census, or—just as plausibly—the 1997 restaurant might have closed down, and the 2007 report is a new, similar-sized restaurant in the same location.

Large-employer plants, by contrast, are extremely rare, so when they are linked over time, I can be highly confident the link is true. For example, in the 1997 publication for the industry “Iron and Steel Mills” in the place “Gary, Indiana,” there is exactly one “2,500 plus” plant and no other plant with more than 250 employees. In the 2007 publication, there again is exactly one “2,500 plus” plant. My matching algorithm links these as being the same plant, which of course is a correct match.

While the algorithm isn’t always perfect, it seems to work very well overall. It greatly helps matters that in the more recent censuses, the location information has been published in greater geographic detail than the county level used in earlier censuses. For example, in the 1997 Census, not only is there a “2,500 plus” steel plant in Gary, but there is another “2,500 plus” steel plant in “East Chicago, Indiana.” These two places are in the same county, so these two plants would be grouped together if the place-level detail in the 1997 Census were not available. Having data at narrow geographic detail makes it possible to reliably match plants over time. The analogous tabulation with detailed geography for the 2007 Census of Manufactures was only just released in January 2011. Combining this freshly available, detailed public data from 1997 to 2007 provides a wealth of information about American manufacturing over a decade of dramatic transformation—invaluable evidence for untangling the “mystery” of disappearing large employers.

A broad overview

Table 1 and Figure 1 show the long-term decline of large-employer plants (defined in this paper as 1,000 or more employees). Employment in such plants fell from 5.1 million in 1977 to only 2.1 million in 2007. The number of such plants decreased by about half, from 2,061 to 1,014 (Figure 2). The decline is even more remarkable in plants with 5,000 employees, where the numbers fell from 192 plants in 1977 to only 49 by 2007.

Table 1: Long-Term Trends in U.S. Manufacturing Employment

Source: U.S. Census Bureau, Census of Manufactures. The source for plants with 5,000 or more employees is County Business Patterns (1977, 1987, 1997, 2007).

Large Image

What has happened to these large employers? It is well known that the U.S. manufacturing sector is in decline generally, that is, across plants of all sizes. Over the 30-year period, overall manufacturing employment fell from 18.5 million to 13.4 million. Since nonmanufacturing employment grew during these decades, manufacturing’s share of employment fell from 22.4 percent to 9.7 percent. While the overall decline of manufacturing is indeed significant, what is happening at the top, to large-employer plants, is even more dramatic. Table 1 and Figure 1 make this point very clear.

Large Image

By looking more closely at these large plants and the enormous changes they’ve undergone, I can get a better sense of the forces behind the overall transformation of the manufacturing sector. And I can do this by tracking plants over time, using the algorithm described above to match large employers in 1997 to the same (if changed) plants in 2007.

To illustrate the matching algorithm at work, first look at huge employers (2,500 plus) that have newly appeared as of 2007, in the sense of not matching to a plant in 1997 with 500 or more employees. There are only 15 of these, making it possible to put all of the plants in a table (Table 2) to get a sense of the data. The plants listed include both brand-new entrants that started from scratch over the 1997-2007 period and existing plants from 1997 with fewer than 500 employees that grew to huge status (2,500 plus) by 2007. Both kinds of expansion are extremely interesting, and it simplifies the algorithm when I don’t have to separate them out.

Table 2: List of “2,500 or More Employee” Plants from 2007 that Are New Entry*

* “New entry” is defined as no match in 1997 with 500 or more employees in the same industry and location.

Source: This table was constructed by the author using published tabulations of the Location of Manufacturing plants from 1997 and 2007 Census of Manufactures.

Five of the new huge plants in Table 2—one-third of the total—are auto plants. These are all highly publicized new plants, for example, the new Nissan facility in Canton, Miss., the Hyundai plant in Montgomery, Ala., and so on. Auto plants are highly capital-intensive facilities, where robots do much of the assembly work; it is no surprise that they are still opening in the United States.

The next four on the list are meat-processing plants, which make intensive use of low-skilled labor. A reporter taking a job at a huge meatpacking plant vividly describes the work: Men standing at assembly lines using knives to hack meat off bone by hand.⁵ Given the difficulties inherent to transporting live animals and fresh meat, it makes sense that this work is still done in the United States. The remaining six plants on the list are generally in high-tech industries, where it is understandable why new capacity is being added.⁶

Table 3 reports the main results of the matching algorithm regarding the disappearance and new appearance of large (again, 1,000 plus) employer plants between 1997 and 2007. The top panel answers the question: Where did the large employers from 1997 go? The table shows that of the 1,503 large employers from 1997, just under half of them (708 plants) remained as large employers 10 years later. Fully one-quarter of them (383 plants) downsized employment to the “500-999” category, and 6.5 percent (97 plants) downsized even further to the “250-499” category.

Table 3A: Large-Employer Plants in 1997: Where did they go?

Table 3B: Large-Employer Plants in 2007: Where Did They Come From?

*Closure includes shrinking to a plant size below 250 employees. See the discussion in the text. Analogously, entry includes starting with a plant in 1997 with fewer than 250 employees.

Source: This table was constructed by the author using published tabulations of the Location of Manufacturing plants from 1997 and 2007 Census of Manufactures.

The remaining 21 percent (315 plants) either closed outright or contracted to a plant size of below 250 employees. Both kinds of decline represent an extreme level of contraction, and I simplify the algorithm by grouping these two outcomes together and calling it “closure.”

The bottom panel answers the related question: Where did the large employers from 2007 come from? Here the table shows that the vast majority of such plants were already large employers in 1997. About 10 percent of them either didn’t exist in 1997 or expanded from a very small base of below 250 employees, an outcome I label “entry.” The industry composition of the entrants is very similar to the entry of new huge employers in Table 2. Nearly 70 percent are in four broad industries: food, transportation, computers and chemicals.

Table 3 reveals a broad overview of what is happening to the disappearing large employers. But to get a clearer picture of what is going on, I need to dig deeper.

Narrowing the investigation

To examine further the case of the disappearing large employers, I narrow the investigation to industries that have been heavily impacted by imports from China. I put particular focus on what is happening in the Piedmont region.

For much of the 20th century, the Piedmont region in the southeastern United States, at the foothills of the Appalachian Mountains, has been a center of low-wage labor, attracting industries that use unskilled labor intensively, in much the same way that China does today. Holmes and Stevens (2004) presents a map of manufacturing activity in the region and some early references. For simplicity, here I am going to define the region broadly to include the following seven states: Virginia, North and South Carolina, Tennessee, Georgia, Alabama and Mississippi. (While Tennessee and Mississippi are not geologically part of the Piedmont plateau region, for this economic analysis, it makes sense to include them.) In 1997, these states accounted for 14.1 percent of the U.S. population.

The two right-hand columns of Table 3a and 3b present an analysis of disappearing large employers as before, but just for plants in the Piedmont region. In 1997, the Piedmont was home to 326 large-employer plants. This is 21.7 percent of the nation’s total of 1,503 large-employer plants at the time, much greater than the Piedmont’s 14.1 percent share of the U.S. population. Note that the closure rate for Piedmont’s large employers is 30.4 percent, well above the national rate of 21 percent.

To get a sense of why the closure rate in the Piedmont is particularly high, it is useful to sharpen the focus still further by looking at industries that have been knocked around by imports from China over the 1997-2007 period. Here I’ll call these the “China Surge” industries.⁷ Table 4 lists the 17 industries. Total employment declined dramatically from 1997 to 2007 for all 17, with infant apparel declining at an astonishing rate of 97 percent. In these industries, imports grew from about 20 percent of the U.S. market to 60 percent over the decade, and China’s share of these imports grew from 20 percent to 57 percent.⁸

Table 4: Employment Change in the Piedmont Region’s “China Surge” Industries

Source: The percent employment change is calculated using the 1997 and 2007 Census of Manufactures. The selection of industries is discussed in Holmes and Stevens (2010).

Now I’ll track what happened to large employers in the China Surge industries between 1997 and 2007. Table 5a shows that the Piedmont had 21 of the large employers in 1997, while the rest of the country had 29. These numbers show the high concentration of these industries in the Piedmont—just 14 percent of the nation’s population, but 42 percent of the large employers in China Surge industries. In other words, the Piedmont region specialized in the same labor-intensive industries, like apparel and furniture, that have now shifted over to China.

Table 5A: Large-Employer Plants in the China Surge Industries in 1997: Where Did They Go?

Table 5B: Food Processing Plants in 1997: Where Did They Go?

*Closure includes shrinking to a plant size below 250 employees. See the discussion in the text.

Source: This table was constructed by the author using published tabulations of the Location of Manufacturing plants from 1997 and 2007 Census of Manufactures.

Things have been rougher for these industries than for the manufacturing sector as a whole, and things are particularly rough for the Piedmont plants. Of the 21 large China Surge employers in the Piedmont in 1997, only one was still a large employer 10 years later. Moreover, as I’ll discuss later, this one plant switched to a different industry little threatened by Chinese imports. Therefore, not a single one of the 21 large employers in the Piedmont survived as a large employer competing head to head with the Chinese. And 13 of them ended up in the closure category. While matters are also rough in the rest of the country, where 14 of 29 closed, China Surge industry plants have fared a little better than those in the Piedmont; five plants outside this region somehow managed, as of 2007, to continue on as large employers. I will further investigate some of these later.

The China Surge industries contrast strongly with food processors, which experience little pressure from imports because of transportation issues. Food processing plants in the Piedmont are doing well (see Table 5b). Of 52 large-employer food processors in 1997, only three ended up in the closure category, a rate of only 5.8 percent, compared with the 10.6 percent closure rate in the rest of the country. Note also in Table 2 that three of the four newly entering huge meat processing plants are in the Piedmont. The bottom line is that in food industries not under import threat, the Piedmont plants are doing better than the country as a whole. But in the China Surge industries, the Piedmont is doing far worse.

Manufacturing in the Piedmont has been hit hard, not only because it has specialized in low-skill-intensive industries, like apparel and furniture, that have been heavily impacted by Chinese imports, but also because even within these industries it has specialized in that segment of the business that makes standardized goods with heavy use of low-wage labor—precisely that part of an industry that is most vulnerable to competition from China. Holmes and Stevens (2010) provide a related analysis. Here, I make the case by digging deeper into the furniture industry.

Making the case with the casegoods

In 1997, wood furniture, such as bedroom and dining room furniture—the industry uses the term “casegoods”—sold anywhere in the United States was very likely made in the vicinity of High Point, N.C., in one of the many towns like Thomasville or Lexington that have lent their names to well-known brands of furniture. This area was turned upside down in a remarkably short time by Chinese imports. Over the years, furniture makers have tried to adopt mass production techniques, but making quality wood furniture requires human craftsmanship—expensive in the United States, but not in China. There is an interesting recent video about the last day of work at the Hooker Furniture Factory, a plant near High Point that closed in 2007. It is striking to see the extent of the hands-on nature of the production process, the physical touches of the wood, the spraying of stain by hand and so on. The piece is fittingly called “With These Hands: The Story of an American Furniture Factory.”⁹ With the relative ease of transporting casegoods from overseas, the U.S. industry collapsed in remarkable fashion.

To understand what has happened, it is useful to contrast casegoods with two related, but very different industries: kitchen cabinets and upholstered furniture. Kitchen cabinets are usually built to the specifications of a particular kitchen. There are two great advantages in having this work done locally: quicker turnover and better communication. The high value of proximity in this industry has kept imports to a minimum. Table 6 shows that the import share is quite small and changed little between 1997 and 2007. Custom plants don’t have assembly lines and tend to be small, craft-oriented shops, averaging only 12 employees in each plant. This is in sharp contrast to the average employment size of 87 workers in casegoods plants in the Piedmont region.

Table 6: Comparison of Different Kinds of Furniture Industries

Source: Author’s calculations with published tabulations of the 1997 and 2007 Census of Manufactures. The import shares use import information posted by the U.S. International Trade Commission at its website, as well as revisions for the furniture industry reported at the website of the International Trade Administration..

Upholstered furniture is yet another story. With wide varieties of fabric patterns and colors, there are more variables to deal with than for casegoods with their limited selection of finishes. This makes managing inventory a central issue. The first key advantage then of U.S. production is that it allows for quick inventory turnaround. The second is the shipping expense of bulky sofas. Therefore, the upholstery work shifted to China tends to be the labor-intensive “cut and sew” of fabric into a “kit.” These fabric kits are cheap to ship overseas, and U.S factories finish sofas by stuffing locally built frames of foam and wood into the imported kits.

The upshot is that the import share for upholstery has remained relatively low, unlike what is happening with wood furniture. While upholstery is like cabinetry in that the work is still done in the United States, it differs in that it is done in large plants, not custom craft shops. (See Table 6.) In this respect, upholstery plants are like the casegood plants: Both are large and produce standardized sizes and shapes.

Table 6 presents some sharp contrasts between the Piedmont and the rest of the United States in terms of these three related industries.¹⁰ First note the Piedmont’s extremely large shares of the casegoods and upholstery businesses, with 47 percent and 69 percent of U.S. production in 1997, vastly exceeding the region’s 14 percent population share. Again, both industries tend to produce standardized products in large factories with low-wage employees.

In contrast, the region’s share of the cabinet industry is relatively close to its population share. This industry does not tend to have large plants full of low-wage, unskilled workers, so—unlike casegoods and upholstery—had no incentive to concentrate in the Piedmont.

A second contrast: average plant size within each of the industries. For casegoods and upholstery, average plant size (in number of employees) is four times larger in the Piedmont than elsewhere in the United States—87 employees in the average Piedmont casegoods plant versus 22 in the rest of the country; 111 employees versus 26 in upholstery plants (see Figure 3). In fact, in terms of average size, casegoods plants in the rest of the United States are closer to cabinet plants than furniture plants in the Piedmont. Furniture plants outside the Piedmont are not making low-skill-intensive, assembly-line-style standardized goods. Instead, they are making either craft-oriented, custom furniture (like an Amish furniture shop) or furniture from highly mechanized production.

Large Image

The final thing to see in Table 6 is what has changed over time. The entire U.S. casegoods industry has been battered by imports, with the share of imports more than doubling from 1997 (29.5 percent) to 2007 (61.8 percent). (See Figures 4 and 5.) But the Piedmont has been especially hard hit. Its share of what is left in the United States has plummeted from 47 percent to 28 percent. In contrast, imports have had relatively little impact on the Piedmont’s shares of U.S. employment in upholstery and cabinetry/countertops, dropping just 1 and 2 percentage points, respectively.

Large Image

Large Image

What about the large employer casegoods manufacturers that have survived?

Having established some facts using data based on plants of all sizes, I’ll complete my analysis of what is happening to casegoods by making use of the linked data on large employers that I created for this paper. In 1997, there were 12 large U.S. employers in casegoods, seven of them in the Piedmont. As of 2007, only one of seven Piedmont plants remained a large employer. But interestingly, this plant switched from casegoods (a tough place to be) to upholstery (a relatively safe place). This is the plant mentioned earlier as the only one of 21 China Surge industry plants in the Piedmont to have remained a large employer. Remarkably, as of 2007, there are no longer any large-employer casegood plants in the Piedmont region.

If I look outside the Piedmont region in 2007, however, I can find a few large employers in China Surge industries. In particular, there are two huge (2,500 plus) plants classified in casegoods, one in Archbold, Ohio, and the other in Trempealeau, Wis. By using publicly available information, I can figure out quite simply what these plants now do.¹¹

Sauder Woodworking’s website claims that its facility in Archbold is one of the “most technologically advanced furniture facilities in the world.” The product is “ready-to-assemble” furniture, so buyers, not workers, perform the labor-intensive task of putting the pieces together. “All Sauder furniture has a paper laminate finish,” implying that machines put on the finish and there is no human handwork, unlike the Hooker plant mentioned above where workers spray on finish. The manufacturing facility in Archbold is the company headquarters and includes management and engineers designing new products and other workers not directly engaged in furniture manufacture. For example, one recently posted job at the Archbold facility is for a “social media specialist” in the marketing department.

Ashley Furniture is one of the largest furniture companies in the world. Its website explains that the facility in Arcadia is both its worldwide headquarters and its core manufacturing center in the United States. While the Wisconsin plant is classified in the casegood industry, on a recent trip to an Ashley store, I found that all of the casegoods were made outside the United States, with tags like “Made in China” or “Made in Malaysia.” A salesperson explained that while the wood furniture is imported, the Wisconsin plant did the upholstery. But even the upholstery’s labor-intensive cut-and-sew work has been sent offshore to a 5,000-employee plant in China.¹²

In summary, there are only two casegood plants with more than 2,500 employees in the 2007 Census, one in Ohio, the other in Wisconsin. With more than 5,000 employees between them, they account for approximately 10 percent of the entire 2007 U.S. casegood employment of 63,000. Thus, it is quantitatively important to understand these two stories. These plants do not look anything like the casegood plants in the Piedmont that have been decimated by Chinese imports. In previous decades, the Piedmont plants had been full of low-wage workers doing routine tasks, but the current Ohio and Wisconsin plants are full of white-collar workers running the company and marketing its products. The plants do indeed make things—that’s why they are classified as “manufacturing”—but the routine, labor-intensive work has moved elsewhere: The Ohio plant has shifted this labor to the consumer, and the Wisconsin plant has sent it to China.

Summing up

Debate about the disappearance of manufacturing jobs in the United States sometimes implies that mysterious forces are at work. But a closer look at recent trends, especially at large-employer plants, reveals no such puzzles. To survive competition from overseas—particularly from China—large employers in the U.S. manufacturing sector have been engaged in a relentless drive to cut routine, unskilled production tasks out of processes taking place in facilities in the United States, where labor is relatively expensive. Unless precluded by transportation barriers (as for live animals and fresh meat) or government restrictions (related to national security), these forces have led to a dramatic decline in the number of large-employer manufacturing plants in the United States.

By linking plants over time, I have shown that most plants that have fallen out of large-employer status have done so by shrinking down to the next-lower size category. Yet closure also has been substantial in industries—such as apparel and furniture, and especially in the Piedmont region—where large plants have tended to employ low-wage workers doing routine tasks. I have noted that the shift of this kind of work out of the Piedmont area to China today is a reprise of the previous century’s shift of this kind of work within the United States. In the earlier case, it was a migration to the Piedmont region away from the high-labor-cost Northeast and Midwest. In both industry migrations, the lure of lower wages was a primary attraction.

There was much consternation and painful adjustment in the earlier period, as industries shifted from North to South within the nation. In the end, things seem to have worked out for former manufacturing giants like Chicago and Boston that have become great centers for services and innovation. For Detroit, things have not gone as well.

How the second showing of this story will play out, with China newly replacing the role of the American South, is an issue of great importance for policy discussions. I believe there is much to be learned through particular focus on large-employer plants; this paper is a step in that direction.

Notes

References

American Iron & Steel Institute. 1951. “Directory of Iron and Steel Works of the United States and Canada: 1951.”

Bernard, Andrew B., and J. Bradford Jensen. 1995. “Exporters, Jobs, and Wages in U.S. Manufacturing, 1976-1987.” Brookings Papers on Economic Activity: Microeconomics, 67-112.

Brown, Charles., and James L. Medoff. 1989. “The Employer Size-Wage Effect.” Journal of Political Economy 97 (5), 1027-59.

Holmes, Thomas J., and John J. Stevens. 2004. “Spatial Distribution of Economic Activities in North America,” in Handbook of Urban and Regional Economics 4, Vernon Henderson and Jacques-François Thisse, eds. New York: Elsevier.

Holmes, Thomas J., and John J. Stevens. 2010. “An Alternative Theory of the Plant Size Distribution with an Application to Trade.” NBER Working Paper 15957.

LeDuff, Charlie. 2000. “At a Slaughterhouse, Some Things Never Die: Who Kills, Who Cuts, Who Bosses Can Depend on Race.” New York Times, June 12.

Russell, Thomas. 2006. “China Cut-and-Sew Pays Off for Ashley: Expansion Underway at Shanghai Facility.” Furniture Today, Nov. 17.

United States Steel Corp. 2010. 2010 Annual Report and Form 10-K.

U.S. Bureau of the Census. 1977, 1987, 1997, 2007. County Business Patterns.

U.S. Bureau of the Census. 2001.1997 Economic Census. CD-ROM. Washington, D.C.: U.S. Dept. Commerce. C1-E97-NAFI-17-US1.

U.S. Bureau of the Census. 2007. Sector 31: EC0731SA1: Manufacturing: Subject Series: Location of Manufacturing Plants: Employment Size for Subsectors and Industries by U.S., State, County and Place. (Downloaded at U.S. Bureau of the Census, FTP.)

Liquidity crises that induce or exacerbate deep recessions, as in 1930 or 2008, are situations in which individuals and firms want to build holdings of liquid assets. Heightened risk, or a perception of it, substantially increases demand for these assets. This reduces the supply available for normal transactions, leading to production and employment declines.

What happened in September 2008 was a kind of bank run. Creditors lost confidence in the ability of investment banks to redeem short-term loans, leading to a precipitous decline in lending in the repurchase agreements (repo) market. Massive lending by the Fed resolved the financial crisis, but not before reductions in business and household spending had led to the worst U.S. recession since the 1930s.

In this essay, we first sketch theoretical ideas that bear on the sources of liquidity crises: bank runs, sunspots and contagion effects, and the moral hazard problem created by deposit insurance. We then describe the repo market, and argue that these theoretical concepts are useful for understanding that market as well.

We conclude with several lessons for regulatory reform and for the role of Federal Reserve policy in coping with future liquidity crises:

• Bank regulation can reduce the likelihood of liquidity crises, but cannot eliminate them entirely.
• During a liquidity crisis, the Fed should act as a lender of last resort.
• The Fed should announce its policy for liquidity crises, explaining how and under what circumstances it will come into play.
• Deposit insurance is part of the answer, but has a limited role.
• The Fed’s lending in a crisis should be targeted toward preserving market liquidity, not particular institutions.

Department of Economics, University of Chicago. The authors thank Douglas Diamond, Gary Gorton, Anil Kashyap, Allan Meltzer, Edward C. Prescott, Harald Uhlig, Warren Weber and Motohiro Yogo for helpful comments. The views expressed herein are those of the authors and not necessarily of the Federal Reserve Bank of Minneapolis or the Federal Reserve System.

Introduction

It is hard to imagine better-motivated legislation than the Dodd-Frank Act, to date the one measure directed at preventing future financial crises. Yet it is hard to find an economist who argues that Dodd-Frank represents any appreciable progress toward this goal, nor is there anything like a consensus among its critics on what legislation should supplement or replace it. Economists cannot yet offer a complete, agreed-upon theoretical framework for thinking about liquidity crises, about the forces that precipitate them or exacerbate them or both.¹ Nevertheless, in our view, many of the main elements are in place. In this essay, we first describe these elements and then discuss how they might be combined to guide legislators and regulators.

Our interest here is in liquidity crises that induce or exacerbate deep recessions, as in 1930 or 2008. These crises are situations in which individuals and firms want to build up their holdings of liquid assets, cash and other securities that are close to cash in the sense that they can be exchanged for cash easily and at a predictable price. These assets have a special role because they are used, indeed required, for carrying out transactions. Heightened risk, or a perception of heightened risk, substantially increases the demand for these assets. This increase in demand has the effect of reducing the supply available to carry out the normal flow of transactions, leading to a reduction in production and employment.

What events are excluded by this definition? The stock market crash of 1929 and the dot-com crash of 2000 are two examples. These large, sudden changes in stock prices reflected changes in beliefs about future returns, but they did not have large, immediate effects on the inventories of cash or other liquid assets that individuals and firms wanted to hold, relative to the volume of their spending. Another example is the unexpected fall in house prices in 2007–08, which led to a reduction in construction activity. Housing construction is a large enough industry that this reduction would have shown up in a decline in overall gross domestic product (GDP), but it would have been comparable in size to other recessions of the postwar era. (Of course, mortgage-backed securities, marketed as liquid assets, did play a central role in the financial crisis, and that role will be discussed below.)

The events that followed the failure of Lehman Brothers in September of 2008 were not a modest recession. The spending declines in the fourth quarter of 2008 and the first quarter of 2009 sent U.S.GDP from 3 percent or 4 percent below trend to 8 percent or 9 percent below, where it has remained ever since. Housing was only a tangential factor in this decline.

We will argue here that what happened in September 2008 was a kind of bank run. Creditors of Lehman Brothers and other investment banks lost confidence in the ability of these banks to redeem short-term loans. One aspect of this loss of confidence was a precipitous decline in lending in the market for repurchase agreements, the repo market. Massive lending by the Fed resolved the financial crisis by the end of the year, but not before reductions in business and household spending had led to the worst U.S. recession since the 1930s.

In this essay, we first sketch several theoretical ideas that bear on the sources of liquidity crises: bank runs, sunspots and contagion effects, and the moral hazard problem created by deposit insurance. We then describe the repo market and argue that these concepts are useful for understanding that market as well. We then draw some conclusions for regulatory reform and for the role of Federal Reserve policy in coping with future liquidity crises.

Bank runs

A simple and widely used theoretical model of bank runs was developed by Douglas Diamond and Philip Dybvig (1983). It describes an economy in terms of the production and consumption of a real good, but to apply their model to actual banking practice, it is helpful to give it a monetary interpretation. In this section, we will sketch their framework, so modified.

Think of an economy as a collection of agents—individuals and firms—who are paid in cash and in turn must pay cash for the goods they buy. That is, all creditors demand payment in cash, so all transactions require the use of cash. Assume that bills for purchases arrive with unpredictable lags and that bills must be paid exactly when due, with draconian penalties for late payment. In a currency-only version of this economy, every agent would need to hold enough currency to cover the worst payment possibility. Even if this worst-case scenario rarely occurred, each individual agent would need to be prepared, and a sizable fraction of each agent’s total wealth would be tied up in non-interest-bearing currency.

In this economy, there is an obvious role for deposit-taking banks, institutions that accept deposits and in return promise to pay interest on those deposits and to redeem deposits for cash at any time. If a bank has a large number of depositors and if the (random) demands for cash of those depositors are less than perfectly correlated, then the worst payment case for the bank as a whole is less than the sum of the individual worst cases for its depositors.

In fact, if the number of depositors is large and if the random cash demands of each depositor have a substantial idiosyncratic (uncorrelated) component, then the bank needs to hold in its cash reserve only something like the average payment for each depositor. If one depositor’s demand for payments is unusually large, the bank can apply the unused portion of someone else’s deposit to honor the check, in effect borrowing from one depositor to finance a loan to another.

By so economizing on its cash reserves, the bank can safely invest a fraction of deposits in interest-bearing securities. Suppose for now that these investments are Treasury bills or high-grade commercial paper, so there is little risk and the bank can quickly convert the assets into cash if its own stock runs low. The return on these investments finances the interest paid to depositors, and in a competitive banking system, all of the return (net of the bank’s operating cost) is used in this way. We can think of such a bank as an institution that pools payment risks, making all of its clients better off than they would be acting on their own.

The bank has promised all of its depositors cash on demand, and it can fulfill this promise if depositors make withdrawals only when they have actual payment obligations. Moreover, if depositors are confident that the bank can fulfill its promise, they have no incentive to make larger withdrawals than necessary: Funds in the bank earn interest while cash does not. Thus, in ordinary times, the bank can and does make good on its promise of cash on demand.

This arrangement, which of course is just fractional reserve banking, has a problem, however. The ability to make good on its promise is fragile. If all its depositors—or a sufficiently large number—simultaneously try to exercise the option to withdraw, the bank in fact does not have nearly enough cash in reserve to cover those withdrawals. In this case, the bank runs out of cash and cannot honor its promise to those who arrive late. If enough other depositors choose to withdraw or if such an event is merely anticipated, then any individual depositor has an incentive to withdraw his own funds as well: Each wants to be at the beginning of the line, not at the end. Diamond and Dybvig’s model captures the essential nature of this second outcome, a bank run.

Notice that the run outcome is just as possible as the outcome in which the bank continues to operate normally. Whether or not a run occurs has nothing to do with mismanagement of the bank or excessive risk in the bank’s portfolio. There are simply two possible outcomes, one good and one bad, and which of them is realized depends on what everyone thinks that others will do.

A crucial feature of this example is that withdrawing depositors must get in line and be served in turn. The bank gives each depositor as much of his deposit as he asks for until the money runs out, and those still in line are simply turned away. This assumption of sequential service has been criticized, and indeed the bank has other possibilities. A bank could prorate withdrawals to distribute default over more depositors, it could temporarily cease honoring demands for withdrawals to allow noncash assets to be liquidated and so on. In some financial markets, such practices are standard. For example, hedge funds typically require investors to commit their funds for a long period and require substantial advance notice for withdrawals.

In a discussion of deposit-taking banks, these criticisms simply ignore the fact that the one function unique to these banks is providing liquidity: facilitating cash payment at low cost. A bank that cannot carry out cash transfers for its depositors has reneged on an important promise, and those depositors will take their business elsewhere. The sequential service assumption in Diamond and Dybvig’s theory highlights the fact that this essential function of a bank is jeopardized during a run. It seems to us an essential feature of any theory of banking.

The main conclusion of Diamond and Dybvig’s theory is that in an economy where cash is required for transactions and banks help agents avoid holding excess cash by pooling their risk, there are always two possible outcomes. The run outcome can be avoided if banks are required to hold 100 percent reserves, but this requirement also makes banks superfluous.

Sunspots, contagion

It is worth stressing that the kind of multiple equilibria that Diamond and Dybvig used to account for observed bank runs can appear in a wide range of economic situations that have no particular connection to banking or the monetary system. This conclusion was established in another remarkable 1983 paper, by David Cass and Karl Shell. They showed that accepting the principle that people act rationally—in their own interest—is not, with any generality, sufficient to determine a unique economic outcome. Fractional reserve banking is but one of many examples where if people somehow come to expect a particular outcome, then that outcome will occur, but if they agree on another, the other will occur. Cass and Shell used the term sunspot equilibrium to emphasize that coordination of beliefs need not make any objective sense: If enough people think the occurrence of sunspots signals a run on a particular bank, it will do so. And if so, who are we to say the sunspots are unrelated to the safety of banks?

At the same time, it is hard not to see patterns in the occurrence of bank runs and currency crises, just as patterns appear in the fashionability of nightspots and in other examples where what you want to do depends on what you think others will do. Our common-sense view that the probability of banking crises can be affected by reserve or capital requirements, by regulation of bank assets or by the general state of the economy is based on real historical evidence. The Diamond-Dybvig model does not help us use this history to design better banking policies. Cass and Shell’s work makes it clear that questions about the origin and influence of beliefs have to be faced.

In a series of papers dating from 2000, Stephen Morris and Hyun Song Shin (2001, 2003) offered a possible resolution of this problem. Like Cass and Shell’s, their approach is abstract, focused on generic situations where people’s beliefs about the beliefs of others have a central role. Nor do Morris and Shin offer any specifics about the objective importance of the sources of their beliefs, so sunspot-based beliefs are not ruled out. The new element in their model is diversity of beliefs.

To stay with the bank run example, suppose that depositors agree on some fundamental measure of the financial health of their bank—the quality of its assets, say—but none of them has exact information about it. All a single depositor sees is a signal, information that is imperfectly correlated with asset quality. Moreover, all depositors get different information, different signals. Those who receive favorable signals—signals that the bank is solid—are content to keep their deposits in the bank. But those who receive signals that the bank is shaky will want to withdraw their funds. There is a cutoff signal value, dividing those who withdraw from those who don’t. And since signals are correlated with asset quality, more people are getting unfavorable signals when asset quality is in fact low. Hence, there is a well-defined tipping point built into the bank’s situation.

Asset quality in our example sounds like something solid and objective, something fundamental. The theoretical argument does not require that, however: It allows anything depositors think is important. But either way, Morris and Shin’s model provides a framework for interpreting historical evidence on the situations that have been correlated with bank failures in the past.

One clear feature of these histories is that bank runs—and financial crises more generally—come in bunches, as though they were contagious. One source of contagion, often emphasized in the crisis of 2008, is that banks lend to each other: If one fails, its creditors can be directly injured. But bank runs can, and often have, spread where such direct connections are minor factors. It seems to be enough that depositors in all banks think there is a useful signal, in the sense of Morris and Shin, in the distress of other banks.

Whatever their sources, these contagion effects are exactly what is systemic about bank failures. Any one bank, no matter how large and respected, can go out of business almost without a ripple. Anyone living in an American city can list the downtown banks he grew up with that vanished in the merger movement of the 1990s. Who misses them? Indeed, who misses Lehman Brothers, for generations one of the most respected financial institutions in the world? Its valuable assets, both physical and human capital, were quickly absorbed by surviving banks without notable loss of services. It was the signal effect of the Lehman failure, whether a signal about the situations of private banks or about the Federal Reserve’s willingness to lend to troubled banks, that triggered the rush to liquidity and safety that followed.

Deposit insurance

The Diamond-Dybvig model and its successors also have implications for government policy directed at bank runs and panics. In all of these models, a system of deposit insurance completely eliminates the incentive to run. By insuring depositors that in the event of a run their deposits will be promptly restored and available to them, it eliminates the possibility of the run outcome altogether. Is this a practical possibility? That is an open question. But the institution of deposit insurance in the United States in 1933 was followed by 75 years without a serious bank run, and that fact must surely be taken as encouraging.

Deposit insurance brings its own problem, however. The models we have discussed thus far are theories about the behavior of bank depositors, about what they believe and what they do. The banks in these models are automatons. This description was adequate for making Diamond and Dybvig’s central point, that any fractional reserve bank is vulnerable to runs even if it is conservatively managed. Indeed, it strengthens their point to show that a run is possible even if the bank invests only in very safe assets. But other issues involve the fact the bankers make choices, too.

In the world we have described thus far, banks invested in Treasury bills and high-grade commercial paper, short-run assets with little or no risk of default. In this world, a bank experiences a run because it is short on vault cash: It is illiquid, not insolvent. But the portfolio of securities a bank holds is a matter of choice, with the usual trade-off between risk and return. Deposit insurance alters the bank’s incentives when it makes that choice, introducing the possibility of insolvency.

John Kareken and Neil Wallace analyzed the incentive effects of deposit insurance in a 1978 paper that has not lost its relevance. Deposit insurance commits the government to pay depositors in the event of asset gambles that turn out badly. An insured bank that takes on risky investments can earn a higher return, and this additional return can be passed on to depositors and shareholders without passing on the added risk. The bank need not fear losing customers by holding a risky portfolio. Indeed, it can gain customers, by offering higher interest than its more cautious rivals. In effect, deposit insurance is a contingent cash transfer from the public to the creditors, depositors and owners of banks, encouraging banks to hold riskier asset portfolios.

Parts of the Dodd-Frank Act are motivated by the desire to protect depositors from unscrupulous or foolish bankers. This is surely a legitimate concern, but it is unrelated to the point of the Kareken and Wallace analysis. Their point, and it is fundamental, is that public funds are committed to banks and their depositors together, altering their joint willingness to take on risk. How they divide the surplus is a secondary matter for this point. Regulating the portfolios of insured banks is the only effective way to deal with this problem.

But regulations designed to prevent depositors from choosing banks with risky portfolios take away options that some of them prefer, without offering any new ones. Some depositors will seek alternative routes to restore these options, and financial institutions will have much to gain from providing them in new guises. This is not just a theoretical possibility. Beginning in the 1970s, as market interest rates rose in response to high inflation rates, depositors began to move their funds out of regulated, insured commercial banks that paid little or no interest and into money market mutual funds and other liquid assets. Even after inflation subsided, depositors were motivated by the higher returns these alternative forms of liquidity offered. In 1965, demand deposits at commercial banks were 18 percent of GDP. By 2005, this ratio had fallen to 5 percent of GDP—about the same as hand-to-hand currency.

The repo market

As deposits moved out of commercial banks, investment banks and money market funds increasingly provided close substitutes for the services commercial banks provide. Like the banks they replaced, they accepted cash in return for promises to repay with interest, leaving the option of when and how much to withdraw up to the lender. The exact form of the contracts involved came in enormous variety. In order to support these activities, financial institutions created new securities and new arrangements for trading them, arrangements that enabled them collectively to clear ever larger trading volumes with smaller and smaller holdings of actual cash. In August of 2008, the entire banking system held about $50 billion in actual cash reserves while clearing trades of $2,996 trillion per day.² Yet every one of these trades involved an uncontingent promise to pay someone hard cash whenever he asked for it. If ever a system was “runnable,” this was it. Where did the run occur?

There were no runs on commercial banks during the financial crisis of 2008. Deposit insurance through the Federal Deposit Insurance Corporation (FDIC) was effective in eliminating the incentive for depositors to withdraw funds. Indeed, as we will see below, demand deposits at commercial banks increased significantly during the crisis. There were two runs on investment banks, however. The run on Bear Stearns in March ended with its purchase by JPMorgan Chase, and the run on Lehman Brothers in September ended with its bankruptcy. In addition, there was an incipient run on money market mutual funds following the collapse of Lehman, halted only when the Treasury stepped in to provide deposit insurance for those institutions.

Of course, for the reasons discussed earlier, these events also heightened the fear of contagion for all financial institutions, altering their willingness to engage in various transactions. Gary Gorton refers to these events, aptly, as a “run on repo.” How did it work?

In economic terms a repurchase agreement (repo) is a securitized loan.³ The lender brings cash to the transaction, while the borrower supplies a T-bill or some other security to be used as collateral. The loans are short term, often one day.

Large lenders in the repo market include money market mutual funds and hedge funds. The repo market performs for these large institutions the same function that commercial banks perform for smaller depositors. In effect, it allows them to pool their cash, collectively economizing on their stocks of non-interest-bearing assets. For lenders, the repo market is attractive because the loans are very short term, so it is a way to earn a return—albeit modest—on cash reserves that would otherwise be idle. In normal times, any lender can withdraw cash by declining to roll over earlier loans. Firms that do not want liquidity do not lend in the repo market, since higher returns are available elsewhere.

What does it mean to have a run in the repo market? Consider a shock that heightens uncertainty about the soundness of financial institutions. Potential lenders will choose to hold more of their cash in reserve, anticipating possible withdrawals by their own clients. As a result, potential borrowers will find it difficult to obtain funds. Actual defaults are rare in this market, but borrowers who hoped to roll over old agreements may have to sell securities on short notice, perhaps at fire sale prices, to obtain cash elsewhere.

The role of collateral in the repo market is similar to the role of deposit insurance at commercial banks. In the cost-benefit calculus that Morris and Shin imagine depositors using when they decide whether to make a precautionary withdrawal, good collateral increases the incentive to continue rolling over short-term loans and hence reduces the likelihood of a successful run.

But while collateral reduces the likelihood of a successful run, it does not eliminate it altogether. Like other forms of fractional reserve banking, the repo market is in effect an institution for pooling cash reserves. Participants can choose to withdraw their cash from the collective pool, and in some circumstances, many will simultaneously choose to exercise this option.

To get a sense of the importance of the repo market, we can look at its size relative to other aggregates. At the end of 2007, $774 billion was held as currency outside banks, $511 billion in private, domestic demand deposits and $3.033 trillion in money market mutual funds. As shown in the table below, all of these figures increased over the following year. In 2008, unlike 1930, demand deposits were a safe asset. Money market mutual funds, which are much larger than demand deposits, increased almost as much.

*End of previous year. Flow of Funds Accounts, Board of Governors of the Federal Reserve System.
**First week of January. Federal Reserve Bank of New York.

The repo market behaved quite differently. At the beginning of 2008, primary dealers held total funds of $3.70 trillion in the repo market, of which $2.54 trillion was in overnight or continuing agreements. Those figures grew slightly during the first half of 2008. Total funds then fell to $2.59 trillion at the beginning of 2009, a 30 percent decline, while overnight and continuing agreements fell to $2.01 trillion, a 21 percent decline. Both figures showed further declines over the subsequent year as well.

Lessons from the panic of 2008

We began by asking what theory and evidence tell us about liquidity crises and about policies to avoid them or to mitigate their severity. The arguments above do not provide a complete answer, but they do point to some broad principles.

(a) Bank regulation can reduce the likelihood of liquidity crises, but it cannot eliminate them entirely.

Banks will fail, and these failures will make failure more likely for others. There is language in Dodd-Frank suggesting that the Fed should take responsibility for predicting and precluding crises, but this task seems to us to be an impossible one, at least for the foreseeable future.⁴

(b) During a liquidity crisis, the Fed should act as a lender of last resort.

In the event of a bank run or a run on the repo market, the Fed can always add liquidity to the system, and there will be occasions—as in 1930 and in the fall of 2008—when it would be irresponsible not to do so.

(c) The Fed should announce its policy for liquidity crises, explaining how and under what circumstances it will come into play.

The events of 2008 illustrate the importance of an announced and well-understood policy. Over the years prior to 2008, investors came to understand that the Fed was operating under an implicit too-big-to-fail policy, in the sense that the depositors/creditors of large banks would be protected. No other policy was ever discussed, and the Fed’s assistance in engineering the orderly exit of Bear Stearns in March 2008 was surely interpreted as evidence that this policy was still in place. The abrupt end of Lehman in September was then all the more shocking.
There is no gain from allowing uncertainty about how the Fed will behave. The beliefs of depositors/lenders are critical in determining the contagion effects of runs that do occur. By announcing a credible policy, the Fed can affect those beliefs, and the Fed needs to use this tool.

(d) Deposit insurance is part of the answer.

When introduced in the Banking Act of 1933, deposit insurance was limited to small deposits, and its role was viewed as consumer protection, not run prevention. Deposit insurance performed this function well during the 2008 crisis: There were no runs on FDIC-insured commercial banks, although many failed or were absorbed by stronger institutions.

Deposit insurance should be retained, although for the reasons described by Kareken and Wallace, the assets held against insured deposits should be carefully regulated.

(e) Deposit insurance has a limited role.

Investment banks, money market funds and the repo market are outside the protection of the insured system, and the liquidity crisis of 2008 involved these other institutions. Could they be brought into the fold, with the relevant portion of their investment portfolio regulated in the same way that commercial banks are?

Higher returns in the uninsured sector will always be attractive for large depositors, and new institutions or arrangements would surely arise, offering liquidity provision on the old, risky terms. Clients will want it, markets will have a strong incentive to provide it and regulators will probably not be able to contain their efforts. Providers will be able to innovate around regulations or move offshore to avoid them. This dilemma leads us to our next point.

(f) The Fed’s lending in a crisis should be targeted toward preserving market liquidity, not particular institutions.

There are two goals here: to have a credible policy for how liquidity will be injected in a crisis and to provide proper incentives for banks during ordinary times. Both goals are met by the Bagehot rule: In a crisis, the central bank should lend on good collateral at a penalty rate. To implement this rule, we need to know how much the Fed should lend and what assets will be regarded as good collateral.

Time consistency requires that no upper bound be placed on crisis lending. The guidelines we have for monetary policy, whether stated in terms of monetary aggregates or interest rates, are directed at long-term objectives and are no help in a liquidity crisis. After the Lehman failure in the fall of 2008, the Fed expanded bank reserves from $40 billion to $800 billion in three months, surely exceeding by far any limit that would have been imposed in August. Even with this decisive response, spending declined sharply over next two quarters.

Because crises occur too rarely for the ex ante formulation of useful quantitative rules, the Fed should have considerable discretion in times of crisis. Nevertheless, because policies should be predictable, the Fed should describe the indicators it will use to decide when lending has reached a sufficient level.

Defining good collateral is more complicated. The quality of collateral is in the eye of the lender, and it can change dramatically from week to week. In this application, though, the lender is the Fed, and it is the responsibility of the Fed to define what it will treat as good collateral. To this end, the Fed should announce an ordering of assets by their quality. The list should be long enough to cover all contingencies, and it would need to be revised from time to time.

In such a regime, banks outside the FDIC would be free to choose their portfolios, with clients, bondholders and equity holders bearing the risk that those choices entail. The lower return on lower-risk assets would be offset, at least in part, by their superior status as collateral in the event of a crisis.

Avoiding liquidity crises altogether is probably more than we can hope for. What we can do is put in place mechanisms to make such crises infrequent and to make their effects manageable.

Notes

1 There is a long tradition of careful historical study of financial crises. Friedman and Schwartz (1963) and Kindleberger (1978) are canonical. Recent books by Reinhart and Rogoff (2009) and Gorton (2010) enrich this literature and bring it to bear on the crisis of 2008.

2 Fedwire Funds Service, Board of Governors of the Federal Reserve System.

3 See Copeland, Martin and Walker (2010), Duffie (2010, 2011) and Gorton (2010) for detailed descriptions of this market.

4 See Meltzer (2009) for a further discussion of this point.

References

Cass, David, and Karl Shell. 1983. Do sunspots matter? Journal of Political Economy 91 (2): 193–227.

Copeland, Adam, Antoine Martin and Michael Walker. 2010. The tri-party repo market before the 2010 reforms. Staff Report 477, Federal Reserve Bank of New York, November.

Diamond, Douglas W., and Philip H. Dybvig. 1983. Bank runs, deposit insurance, and liquidity. Journal of Political Economy 91 (3): 401–19.

Duffie, Darrell. 2010. The failure mechanics of dealer banks. Journal of Economic Perspectives 24: 51–72.

Duffie, Darrell. 2011. How Big Banks Fail and What to Do about It. Princeton University Press.

Friedman, Milton, and Anna J. Schwartz. 1963. A Monetary History of the United States, 1867–1960. Princeton University Press.

Gorton, Gary. 2010. Slapped by the Invisible Hand: The Panic of 2007. Oxford University Press.

Kareken, John H., and Neil Wallace. 1978. Deposit insurance and bank regulation: A partial-equilibrium exposition. Journal of Business 51 (3): 413–38.

Kindleberger, Charles P. 1978. Manias, Panics, and Crashes: A History of Financial Crises. Wiley.

Meltzer, Allan H. 2009. Regulatory reform and the Federal Reserve. Testimony, July 23, Senate Banking Committee.

Morris, Stephen, and Hyun Song Shin. 2001. Rethinking multiple equilibria in macroeconomic modelling NBER Macroeconomics Annual 2000. MIT Press, 139–61.

The news reports have been grim:

• “Slump Creates Lack of Mobility for Americans” (New York Times, April 2009)²
• “The recession is claiming yet another victim: Americans’ near-constitutional right to pick up and move to a better job.” (Washington Post, July 2010)³
• “One of the hallmarks of the American worker has been mobility—the speed with which people … have moved to find opportunities. But the recession of the last two years has produced a profound change, creating conditions that have tethered many people where they are.” (Los Angeles Times, December 2009)⁴

Low migration has raised worries not just in the media, but among policymakers. The fundamental fear is that if unemployed workers don’t move to states with stronger job markets, the economy could remain stalled for years. Citing such arguments, the International Monetary Fund recently blamed “slower inter-state migration, likely related to the housing crash,” for the persistent high level of U.S. unemployment.⁵ Leaders of the U.S. Treasury and the Federal Reserve System have expressed similar concerns.

A closer look at the data reveals that such concerns are unfounded. Our research shows that what one demographer called the “Great American Migration Slowdown”⁶ never really happened.

Figure 1 shows the data that got reporters, researchers and policymakers so worried: the annual interstate migration rate for the past decade, as calculated by the U.S. Census Bureau and published on its Web site.⁷ According to this graph, the migration rate apparently plummeted in 2006 from a relatively high plateau earlier in the decade.

But this graph is misleading. The data for it come from the Census Bureau’s Current Population Survey (CPS), but a 2006 change in how the bureau analyzed the data lowered the measured rate of migration. Indeed, it could be argued that the new method is more accurate than the old and that previous bureau reports overestimated interstate migration rates while understating local migration. In any case, our research shows that interstate migration rates have remained on a slow downward trend over the past decade, and there has been no dramatic change in this trend.

How did a change in data analysis cause such misperceptions? About 10 percent of respondents in the CPS don’t answer the bureau’s questions about where they lived a year ago; to calculate the migration rate, the bureau has to guess whether these people migrated and, if so, from where. Statisticians call these guesses “imputations.” In 2006, the bureau changed the way it calculates the imputations. This change in methods—not any actual change in migration patterns—turns out to be responsible for much of the recent decline in reported migration rates. The change explains 90 percent of the reported decrease in interstate migration between 2005 and 2006, and 42 percent of the decrease between 2000 (the recent high-water mark) and 2010.

Figure 2 illustrates the problem. The figure shows the interstate migration rate for all CPS respondents and, separately, the rate for those with original, nonimputed data and the rate for those with imputed data. From 1996 to 1998 and from 2006 to 2010, the rate for respondents with imputed data is only slightly higher than the rate for respondents with original data, and the rate for all respondents is likewise very close to the rate for respondents with original data. But from 1999 to 2005, the interstate migration rate for respondents with imputed data is three to five times the rate for respondents with nonimputed data.

In this paper, we explain why statisticians impute answers for people who don’t answer survey questions, describe how the Census Bureau’s imputation methods changed, document how the change affected estimates of migration rates and discuss some policy implications of our results. This paper is an informal discussion, written in a question-and-answer format. Readers interested in technical details should refer to our Federal Reserve Bank of Minneapolis Working Paper 681, “Interstate Migration Has Fallen Less Than You Think: Consequences of Hot Deck Imputation in the Current Population Survey.” The views expressed here are those of the authors and not necessarily of others in the Federal Reserve System.

Why does the Census Bureau “invent” answers to survey questions?

In 2010, of every 100 people surveyed for the CPS, 87 said they hadn’t moved between states, 1 said they had moved between states, and 12 didn’t answer questions about migration. What we assume about those 12 people out of every hundred matters a lot: If they all migrated, the migration rate would be much higher than if none of them migrated. We could ignore the 12 entirely—and, in fact, we suggest later in this paper that ignoring the nonrespondents turns out to be a good solution in this case.

But scholars, including ourselves, usually think it’s a bad idea to ignore people who don’t answer survey questions. The reason is that nonrespondents are often very different from respondents. For example, young workers move more often than other people, and young workers might also be too busy with their careers to diligently answer every question on the survey. In that case, the people who do answer the migration questions won’t be representative of all Americans—they will be older and less mobile than the average person. To get an accurate picture of migration, it’s important to take account of these differences. The Census Bureau does so by “inventing” (or imputing) what it believes to be accurate responses for those questions that weren’t answered.

If the CPS has so many skipped questions, why not use a different data set?

The CPS migration data are a unique and invaluable resource for research on internal migration in the United States. CPS migration data have been published annually since 1948, the longest-running migration data series the nation has. Unlike other large data sets, such as the decennial census and the American Community Survey, the CPS can precisely measure fluctuations in migration from one year to the next. The CPS also allows researchers to study the relationship between migration and a vast number of other characteristics of workers and households. In addition, the CPS provides a representative sample of the entire U.S. population, unlike data from the Internal Revenue Service (which cover only income-tax filers) or moving companies (which cover only people wealthy enough to hire movers).

Other data sets do support our argument that migration has not fallen drastically in recent years. Migration rates in the nonimputed CPS data—but not in the imputed data—are consistent with rates calculated from other data sets.

How does the Census Bureau impute answers for people who skip questions?

The Census Bureau uses a method called “hot deck allocation.” Participants’ survey forms are fed through a computer one at a time. If someone didn’t answer a particular question, the computer looks back through the forms it processed previously until it finds the most recently processed survey participant who’s similar to the nonrespondent but who did answer the question. The previously processed forms are called the hot deck.

Let’s say Mr. Smith, a 33-year-old black homeowner living in the Midwest, didn’t answer a particular question. The computer might look through the hot deck for the most recently processed 33-year-old black homeowner in the Midwest who did answer—Mr. Jones, say—and then copy Mr. Jones’ answer onto Mr. Smith’s form. The person whose missing answer is filled in is known as a recipient; the person whose answer is used is known as a donor. The process is called imputation through hot deck allocation.

Why does hot deck allocation help?

Basically, hot deck allocation guesses that people who have similar demographic characteristics, such as age, race and region of residence, are also similar in other characteristics, such as migration. It’s not a perfect method, of course, but it’s much better than assuming nonrespondents are similar to respondents, which is rarely true. For example, hot deck allocation ensures fairly accurate results even if young people are more likely to skip the migration questions.

What changed in 2006?

The Census Bureau changed the order in which it feeds the survey forms through the computer. That sounds innocuous, but it wasn’t. If the forms go into the computer in a different order, then the computer picks different forms from the hot deck when it needs to fill in missing answers. The new computer processing order reduced the reported interstate migration rate by imputing fewer interstate moves and more local moves to nonrespondents.

How exactly does the processing order matter?

For migration data, the imputation procedure fills in missing answers to two questions: Did the person live in the same home one year ago? If not, where did the person live one year ago? Once these variables are filled, additional variables are calculated that categorize movers as having moved within a county, between counties in the same state, between states or from abroad, based on the distance between the respondent’s current location and his or her (possibly imputed) location one year ago.

Since 2006, the Census Bureau has processed CPS surveys in geographic order. So if someone skips the migration questions, the computer usually copies an answer from a donor who lives nearby. Since long-distance migration is rare, the donor’s location one year ago is also usually close to the recipient’s current location. Thus, if the computer imputes that the recipient moved, it usually imputes a local move.

But before 2006, the data were not exactly in geographic order.⁸ That meant donors lived farther from recipients, donors’ locations one year ago were also farther from recipients’ current locations and recipients were more likely to have imputed interstate moves.

For example, suppose a person in Minneapolis fails to answer the migration questions and is matched with a donor who moved, so that the nonrespondent is coded as a mover. If imputations are done in geographic order, the donor will probably also come from Minneapolis, and the donor’s location one year ago was also probably near Minneapolis (since long-distance migration is statistically infrequent). The geographic procedure will thus usually impute that the nonrespondent made a local move, not an interstate move.

However, if imputations are not done in exact geographic order, the donor is more likely to live far away from Minneapolis—in Sioux Falls, say. Because interstate migration is quite unusual, the donor’s location a year ago was probably near Sioux Falls. But the pre-2006 procedure will inaccurately guess that the nonrespondent in Minneapolis made an interstate move, from South Dakota to Minnesota, not because the nonrespondent is similar to a respondent who moved a long distance (from Sioux Falls to Minneapolis), but rather because the nonrespondent is similar to a respondent who moved a short distance in South Dakota.

In Figure 2, we show how the pre-2006 procedure imputed many more interstate moves than the new procedure. It’s also interesting to look at local moves. Figure 3 shows that from 1999 to 2005, the within-county migration rate was depressed among respondents with imputed data. Thus, the pre-2006 imputation procedure spuriously imputed long-distance moves when they were more likely local moves. Our research has found that the change in imputation procedures had little effect on the total migration rate, because the decrease in imputed interstate moves in 2006 canceled out an increase in imputed within-county moves.

Which migration rate estimates are more accurate: pre-2006 or post-2006?

We agree with the Census Bureau that the procedure in use since 2006 is likely to produce the most reliable estimates of migration rates. The old procedure likely overstated the rate of interstate migration and understated the rate of local migration.

Do the problems with imputed migration data mean it’s always a bad idea to use imputed data?

Absolutely not! As we discuss above, imputing missing data is better than dropping nonrespondents in almost all circumstances, because nonrespondents might differ from respondents, and those differences should not be ignored. In addition, we believe the current statistics including imputations are the most accurate estimates of interstate migration—more accurate than either the nonimputed data or the estimates using the old imputation procedure. The problem here is not with imputation itself; it’s just that a change in imputation methods created a misleading trend in the migration statistics. In sum, imputation is extremely useful, but researchers must analyze the data carefully when imputation procedures change from one year to the next.

How has interstate migration changed in recent years?

The change in imputation procedures in 2006 means that simple comparisons of pre-2006 and post-2006 Census Bureau data do not accurately measure trends in interstate migration. Fortunately, there is a simple way to make accurate comparisons: By ignoring imputed data we can eliminate the problem introduced when the Census Bureau switched methods mid-way through the 2000s.

Since 2006, the interstate migration rate including imputed data has been virtually identical to the rate using only nonimputed data. Figure 4 illustrates this point by reproducing Figure 2 without the imputed-data migration rate, to show only the contrast between overall and nonimputed rates. Because the interstate migration rates using nonimputed data and using all data have been virtually identical for the past five years, we think that the rate using nonimputed data is a reliable guide before 2006 as well. We can study trends in the overall interstate migration rate by focusing on the rate in nonimputed data, and doing so removes the fluctuations induced by changes in the imputation procedure.

We have also checked whether the nonimputed data are a reliable guide to the trend over time by constructing our own imputations for every year from 1996 to 2010, using a method that does not change over time. When we do this, we find that the estimated migration rate using our imputation method is virtually identical to the rate in nonimputed data.

Figure 4 shows that, once we remove the effect of changes in the imputation procedure, the interstate migration rate has hewed closely to a smooth downward trend for the past 15 years. With imputations included, the rate peaked at 3.12 percent in the 2000 survey, fell to 2.59 percent by the 2005 survey, plummeted to 1.96 percent in the 2006 survey, and is now down to 1.44 percent. Without imputations, the rate was 2.35 percent in the 2000 survey, 1.93 percent in the 2005 survey, 1.87 percent in the 2006 survey, and 1.38 percent in the 2010 survey. Thus, the change in imputation procedures explains nine-tenths of the 0.63 percentage point drop from 2005 to 2006 and four-tenths of the 1.68 percentage point drop from 2000 to 2010—the apparent declines in mobility that have been of such concern to policymakers and others.

Did interstate migration fall during the Great Recession?

As Figure 4 shows, interstate migration has been trending downward for many years. But, relative to that trend, there was no additional decrease in interstate migration during the December 2007-June 2009 recession. To see this, it’s important to know how the CPS data are collected. The survey is taken in February through April each year and asks people whether they moved in the previous 12 months. Thus, the data points corresponding to migration during the recession are the data points for the 2008, 2009 and 2010 surveys. Figure 4 shows that, in the nonimputed data—that is, the data we consider to be most accurate—migration fell faster than trend in the 2007 survey year, well before the recession began in December.

Will the long-run decline in migration cause problems for the labor market?

This is an important focus for ongoing research. Many factors might affect whether people migrate. It can be costly to sell a house and buy a new one. Some people have a strong preference for living near family and friends. Others may want to move to a better climate or a place where there are more jobs. Which of these factors are mainly responsible for the downward trend in migration will determine whether we should expect falling migration rates to hurt the labor market.

On the one hand, if migration is falling because it is becoming more difficult to sell a house, then it may be difficult for unemployed workers to move to where the jobs are. On the other hand, in the past, the United States has usually experienced high migration rates when some regions of the country had much stronger economies than others. Historically, people migrated from the South to manufacturing centers in the North, and from rural farming communities to big cities. Regional economic disparities have been declining in the United States for some time. If that’s why migration is falling, there is little cause for concern: Migration is lower than in the past because it’s unneeded, not because it’s costly.

How should the government respond if low migration is hurting the labor market?

Determining the best policies requires a careful analysis of why migration has fallen—and, as a first step, accurate data on how much migration actually has fallen. If research shows that migration actually has fallen significantly, that low migration is hurting labor markets and that, for example, the high cost of selling a home is an important factor inhibiting migration, it still isn’t necessarily the case that the government can or should intervene. Studying and designing potential policies is another important area for research. Our hope is that the corrected data series on migration that we provide in our working paper will help make such research possible.

The 2007-09 recession in the United States was, by almost any measure, exceptional. It was markedly different from other post-World War II U.S. recessions; it was also quite unlike near-simultaneous recessions in other advanced economies. Indeed, in a qualitative sense, the U.S. Great Recession resembled the Great Depression far more closely than it did any of the postwar recessions. And similarly, economists have yet to reach consensus on what truly caused the Great Recession. Why was it so severe? Why did it last so long? Why, in particular, did it have such a major impact on labor markets? This economic policy paper describes some of the defining characteristics of the recent U.S. recession and examines two potential explanations for its impact and duration.

A close analysis of the 2007-09 recession reveals that the central commonality between the Great Recession—at least as experienced in the United States—and the Great Depression is not the role of financial panic, as many have claimed, but rather severe distortion in labor markets. The fact that labor market dysfunction, not banking panic, was at the heart of both episodes of chronic high unemployment leads to very different conclusions about policy.

There is little doubt that a panic in financial markets, sparked by a collapse in housing prices and the value of mortgage-backed securities, led to the financial crisis that coincided with the worsening of the recent U.S. recession. But strong questions remain as to whether this dysfunction in the financial system, or poorly designed government policies seeking to ameliorate the recession or perhaps a combination thereof, was responsible for the recession’s depth and duration. Similarly, the role of government policy in the onset and development of the Great Depression, particularly as it affected labor markets, deserves greater attention.

The goal of this paper is to diagnose the recent recession with an eye toward clarifying the factors that caused it to last as long as it did, with such harsh impact, especially on labor markets. The paper begins with a description of the significant differences among recessions just mentioned—particularly by pointing out that in the United States, unlike other countries recently and the United States in other recessions, the decline in economic output and income is due exclusively to a drop in labor input. It then proceeds with a diagnosis of the recession through analysis of factors behind these empirical findings—especially that of lower labor input—using a technique known as business cycle accounting.

The next step is a discussion about whether two potential theories for the recession are consistent with this diagnosis. That is, how well do the financial dysfunction and poor policy hypotheses jibe with the finding of dramatically lower labor input? The paper ultimately concludes that serious questions remain regarding the financial explanation—questions relating to corporate cash positions, small-firm dynamics, contraction in financial intermediation and the duration of economic weakness. It further suggests that the policy explanation, while promising, requires further research, much of which is under way. The views expressed here are those of the author, and not necessarily of others in the Federal Reserve System.

How this recession differed

The 2007-09 U.S. recession differed considerably from earlier post-World War II recessions both in the behavior of key variables like output, consumption, investment and labor as well as in the possible factors that might account for fluctuations observed in these variables. This section will discuss the first: the differences seen in major economic variables in this recession compared with others. The next section will diagnose factors behind the fluctuations.

Table 1 shows the percent changes in U.S. economic variables during the recent recession and during the average postwar recession. (These are calculated on a per capita basis for the “peak-to-trough” span of each recession. Peak values for each variable are normalized to 100.) Clearly, the 2007-09 recession was more severe than the average postwar recession, and this is particularly true for labor hours and consumption. Per capita hours worked declined 8.7 percent during the recent recession compared with a postwar average decline of 3.2 percent. The declines in output (real gross domestic product, GDP), consumption, investment and employment were also much larger in the 2007-09 recession than in prior recessions.

Table 1: 2007-09 Recession versus Postwar Recessions, United States

(Percent Change in per Capita Values)

The recent recession was also much different in the United States than in comparable large, high-income nations such as Canada, France, Germany, Italy, Japan and the United Kingdom, and again the most striking difference is the larger U.S. impact on labor markets as measured by employment levels (hours worked were not available for other countries).

Table 2 compares the 2007-09 recession in the United States and these six other nations, with the average for the six nations in the second row. Again, the decline in per capita employment is much larger in the United States (6.7 percent) than in the other countries (2.0 percent, on average). But despite the much smaller employment decline in the six countries, per capita output fell more there than it did in the United States (8.5 percent versus 7.2 percent), indicating that the nations experienced much different productivity changes during the recession. Given the roughly similar nature of the financial crisis globally, these differences bear scrutiny in efforts to understand the Great Recession. Also notable: Investment fell over twice as much in the United States as in the other nations, 33.5 percent versus 16.4 percent.

Table 2: 2007-09 Recession in the United States versus Six Other High-Income Countries

(Percent Change in per Capita Values)

Diagnosing the differences

Understanding the factors and mechanisms behind the recent recession requires economic insight into the differences just described. Most particularly, what explains the behavior of labor markets? Why did labor hours and employment levels drop so precipitously in the U.S. Great Recession compared with earlier U.S. recessions and with the parallel recessions elsewhere?

To better understand these differences, this paper uses the perspective of neoclassical (or general equilibrium) business cycle theory, a concept developed by economists Finn Kydland and Edward Prescott in the early 1980s.² The theory is based on a framework of explicit optimization problems faced by an economy’s decision makers—households and firms. Each household seeks to optimize its well-being by making decisions regarding how much to consume, how much to save and how much time to spend working, while each firm maximizes its profits by making decisions regarding how much labor to hire and how much to invest in the firm’s business.

These decisions are made within the context of a specific production function in which inputs of capital (from whatever is saved and invested) and labor (from households who decide to provide it) are combined as efficiently as technology allows to produce economic output. This framework, expressed in a set of equations, is solved mathematically, and the resulting solution gives clues to how the economy functions—or in the periods analyzed here, malfunctions.

Kydland and Prescott’s original “real business cycle” model has become considerably more elaborate over the past 30 years, and a particular technique derived from it, business cycle accounting, provides a diagnostic method for parsing the many factors behind economic fluctuations. The accounting procedure is mathematically complex, but it boils down to measuring differences in specific economic variables during normal times versus atypically good and bad times—periods of equilibrium compared with booms and recessions.

The procedure looks at the variables in the economic relationships just described—decisions about consumption or investment; decisions about labor or leisure; and the use of production technologies that combine labor and capital to generate output—during the normal and atypical times and calculates discrepancies between the two. These value differences are usually called “frictions,” “wedges” or, in this paper, “deviations.”³

The key point is that these deviations are more than just numbers: They represent significant economic dysfunction. In the labor or leisure decision, for example, households normally decide to go to work if the wage being offered is sufficiently high that it compensates them (in terms of what they might buy with that wage) for the sacrifice they must make in using their time to work. Economists say the opportunity cost of working will equal the marginal benefit of working, or more accurately—if with more jargon—the marginal rate of substitution between consumption and leisure will equal the marginal product of labor. Said otherwise, if there are no frictions/wedges/deviations, a firm will offer a worker a wage sufficiently large to convince the worker to work for an hour if that hour’s work will produce output equal in value to the wage.

But what if there is a labor deviation? That means something is amiss, economically. If there is a numerical deviation from the equilibrium labor-or-leisure value, that means that income from labor is being taxed (or subsidized) so that the standard equation is upset: The marginal rate of substitution for households and the marginal product of labor for firms are no longer being equated, and labor markets won’t operate normally.

Similarly, there can be productivity deviations when numerical estimates from the two sides of the production function aren’t equal (actual output is higher or lower than can be accounted for by the amounts of labor and capital in use, given current technologies). And capital deviations exist when estimates from the allocation decision between consumption and investment aren’t working out (capital is being over- or undersupplied relative to the marginal benefit that could be derived from investing in physical capital).

Applying the diagnostic tool

Calculating the deviations is, again, a mathematically complicated process. But simply put, it involves feeding actual data into the equations that represent the production function, the labor-leisure decision and the consumption-investment decision, and then subtracting 1 from the ratio of the left- and right-hand sides of each of the three equations. The results are the deviations from equilibrium accounted for by disturbances to productivity, labor and/or capital.

Since an economy is ultimately composed of these elements, pinpointing the source of economic fluctuation is essentially a question of where these various deviations occur during any given business cycle and how big they are. A large, negative productivity deviation, for instance, would mean that actual output is below the level that should be generated by the capital and labor that were actually supplied. A distortion in productivity would then be the locus of the problem.

The current analysis applies business cycle accounting to the recessions previously discussed: previous postwar recessions in the United States, and the 2007-09 recession in the United States and six other high-income countries. Table 3 provides the results of the diagnostic analysis, with the labor, capital and productivity deviations for respective countries and recessions. (The deviations are expressed as percent differences from equilibrium where—in the absence of these deviations—both sides of the three equations would be equal.)

Table 3: Recession Deviations in the United States and Other Nations

The most obvious discovery from this comparison of previous U.S. recessions and the most recent one (see Panel A of Table 3) is that the 2007-09 U.S. recession manifested very little disturbance to productivity processes or capital decisions, but an extremely large distortion to labor supply.

In theory, well-functioning labor markets will equalize the marginal product of labor and the rate at which households are willing to offer their labor rather than enjoy their leisure. During the average postwar U.S. recession, however, there was a -2.4 percent deviation in this theoretical equivalence, meaning that the marginal product exceeded the marginal rate of substitution by an average of 2.4 percent. Essentially, it was as if labor income were being taxed at an additional 2.4 percent rate.

But during the Great Recession, the labor deviation was far greater: -12.9 percent. (For comparison, the next-largest postwar U.S. recession deviation was -4.7 percent during the 1973 recession.) This deviation was also markedly higher than any seen in the six high-income countries, which averaged just 0.9 percent (see Panel B of Table 3). Notably, this was a positive deviation, suggesting a net subsidy rather than a tax on labor income, and it was due to sizable positive deviations in France, Germany and Japan (1.7 percent, 4.8 percent and 2.9 percent, respectively), meaning that employment in those countries was in fact higher than the level consistent with the marginal product of labor.

In contrast, there was remarkably little distortion in capital markets during the 2007-09 recession in the United States. The capital deviation was 0.3 percent. By comparison, the distortions in other postwar U.S. recessions were large: averaging to a 1.8 percent capital deviation. These positive deviations suggest that capital income enjoyed what would be equivalent to a small effective tax cut during those periods, rather than a tax increase that would have depressed economic activity. Indeed, not a single recession analyzed here—in the United States or abroad—shows a large, negative capital distortion; later, this paper will discuss the implications this absence of capital distortion has for the extent to which models with financial system imperfections that affect capital markets can account for the 2007-09 recession.

As for productivity, the 2007-09 U.S. recession displayed virtually no distortion: just -0.1 percent. The production function is the relationship between inputs and output, and so the productivity deviation can be thought of as a measure of any disturbances in that relationship. Disruptions (positive and negative) to technology are part of this, but the productivity deviation will pick up any factors that change the connection between measured labor and capital inputs, and measured output.

So, during the recent recession, the United States experienced very little disruption in the relationship between inputs and outputs. This was an anomaly as recessions go. The postwar U.S. recession average productivity deviation was -2.2 percent, and the productivity deviation in other high-income nations was -7.1 percent in the 2007-09 recession.

The fact that there is essentially no productivity decline suggests that the sources and mechanisms of the 2007-09 U.S. recession differ substantially from earlier postwar recessions in the United States, and also from the parallel recessions of 2007-09 in other high-income economies. Instead, the 2007-09 U.S. recession appears to be almost exclusively related to a factor that affected the labor market substantially, and it did so by changing the relationship between the marginal rate of substitution between leisure and consumption, and the marginal product of labor. (Indeed, in a separate simulation exercise, a labor deviation of this size by itself can account for drops in output, employment and investment that roughly match what actually occurred in the 2007-09 U.S. recession.)

It’s notable that while the recent recession in the United States is unique relative to other postwar recessions, both here and in other high-income nations, it is qualitatively very similar to the Great Depression. Throughout the 1930s, per capita hours worked and output remained well below normal levels, indicating a very large labor deviation. Calculated as was done here for recent recessions, the average labor deviation between 1930 and 1939 was about -26 percent, roughly twice as large as the -12.9 percent deviation in the third quarter of 2009.

Hypotheses of the Great Recession

This diagnostic information regarding deviations in fundamental economic relationships will help assess two hypotheses about the 2007-09 recession: the financial explanation and the policy explanation. Given the key finding of the diagnosis—substantial disturbance in labor markets resulting in a very large and protracted drop in hours worked—what is the potential of each hypothesis for explaining the behavior of labor markets in a severe recession?

The financial explanation

The financial explanation for the Great Recession argues that declining values of asset-backed securities and the near-failure of large financial institutions accelerated the recession through reduced financial intermediation services (that is, mechanisms for borrowing and lending) and associated spikes in interest rate spreads. Gary Gorton and other economists document reduced volumes of commercial paper and repo markets and argue that this decrease in financial liquidity led to broader economic dysfunction, including reduced output and employment.⁴

But documenting the severity of the financial crisis does not establish that it was itself the major factor behind the recession. To make this causal connection, the financial explanation emphasizes that severe downturns such as the Great Depression were associated with financial crises. Proponents also point to theoretical models in which quantitative increases in financial imperfections reduce investment, output, consumption and employment.

This explanation seems intuitively powerful, even obvious, but potential weaknesses lie in its omission of several key issues. These include documenting internal cash positions and declines in lending volumes. As suggested earlier, it also appears to be inconsistent with the diagnostic accounting evidence presented in this paper. Examining these questions further raises a number of significant challenges to the idea that financial distress deepened the recession.

In terms of economic theory, the ways in which capital market flaws affect the economy are largely at odds with the diagnostic findings presented earlier. The financial explanation suggests that capital market imperfections lead to broader economic problems; business cycle accounting would measure this effect with the capital deviation, the wedge between the return paid to suppliers of capital and the cost paid by those who use it. But these capital deviations were extremely small in the United States and other high-income countries during the 2007-09 recession, just 0.3 percent and 0.1 percent, respectively.

A theory in which financial distress generates the large labor deviation that was a hallmark of the U.S. recession might reconcile this discrepancy between evidence and explanation. But even with an effective theory linking capital markets and labor deviation, it would remain unclear why the labor distortion of 2007-09 was so much larger (-12.9 percent) in the United States than the capital deviation (0.3) that captures dysfunction in capital markets.

Other challenges to the financial explanation

Other data challenge the idea that financial market imperfections cause severe economic downturns. The idea’s proponents often argue that the Great Depression was deep and protracted because of associated banking crises, and many draw parallels to the Great Recession. But several details suggest that banking crises were not, in fact, the major causal factor in the Depression.

Contrary to general perception, for example, the 40 percent decline in the number of U.S. banks between 1929 and 1933 had little impact on actual banking capacity because most of the Depression-era banks that closed were either very small or merged. The share of deposits in banks that closed or suspended operation between 1930 and 1933 was 1.7 percent, 4.3 percent, 2 percent and 11 percent in each respective year.

There is also the question of timing. The Depression was “Great” before any of the monetary contraction or banking crises occurred. Industrial hours worked dropped by 29 percent in the United States before the first big bank crisis in late 1930 and also before the nation’s money stock fell.

These facts about capacity decline and panic dates indicate that the Depression would have been severe even in the absence of banking and financial crises, and suggest that drawing lessons from Depression financial crises to other economic downturns is premature.

Regardless of potential parallels between the Depression and the Great Recession, more recent facts challenge the explanatory strength of the financial hypothesis for the 2007-09 recession in the United States. These facts relate to corporate cash positions, small-firm dynamics, contraction in financial intermediation and the duration of economic weakness.

Discussions of problems in financial markets often ignore internal cash held by corporations, though it is a very good substitute for external financing in the event of financial market disruption. The accompanying figure shows that the corporate sector typically has substantial cash reserves. The figure shows available funds and gross investment as a fraction of corporate GDP between 1960 and 2009. It indicates that corporations typically have nearly as much internal cash as they invest on plant and equipment. And, notably, cash is high and rising in recent years.

Other evidence suggests that most corporate investment, regardless of economic sector, is financed internally, contrary to the argument that some sectors suffer disproportionately during financial crises.⁵ The fact that firms have sufficient cash to finance capital spending stands in sharp contrast with the assumption of models where financial market imperfections are the source of broader economic downturns.

Another assertion made by proponents of the financial explanation is that small firms have much less access to capital markets, and they’re therefore affected much more than large firms during crises. Again, evidence suggests that this is inaccurate: Recent research shows virtually no change in the relative sales performance of small versus large firms during the recent recession.⁶

Contraction in financial intermediation (borrowing and lending) is another key point in the financial explanation. But some measures of intermediation did not decline substantially during the 2007-09 recession. Bank credit relative to nominal GDP, for example, rose at the end of 2008 to an all-time high. This ratio declined by early 2010, but bank credit remained at a higher level than any time before 2008.⁷ Similarly, data show that household borrowing levels and their composition are virtually unchanged since 2007, again suggesting that the overall volumes of financial intermediation have not declined markedly.

But perhaps the most challenging issue regarding the financial explanation is why economic weakness has continued for so long after the worst of the financial crisis passed in November 2008 or so. Interest rates on relatively risky Baa bonds jumped about 2.5 percentage points, to about 9.5 percent, from mid-September to late October 2008, when financial markets were reacting to news about AIG, Lehman Brothers and related events. But afterward, it dropped by about 3 points to the level that prevailed before the recession. Still, despite these declining interest rates, the number of hours worked in the United States recovered very little, even through mid-2010.

The continuation of the recession long after the worst of the financial crisis raises a difficult puzzle about why employment has not recovered more quickly. Low productivity isn’t the explanation for continued economic weakness in the United States: As documented above, productivity deviation during the recent recession was very small.

None of this evidence should be interpreted as indicating that the financial crisis did not contribute significantly in some way to the 2007-09 recession here or abroad. However, given the mechanisms through which financial market imperfections are argued to impact economic activity in leading theoretical models, the diagnostics and other data presented here reveal a number of difficult questions about the financial explanation. More research is needed on the issues just discussed, and on the productivity and labor deviation differences between the United States and other high-income countries, before the contribution of financial factors to the 2007-09 recession can be accurately evaluated.

The policy explanation

If the financial explanation is not entirely convincing, particularly for the failure of employment to recover after the crisis, is there another story that could account more fully for the macroeconomic fluctuations of 2007-09? Many researchers offer a policy explanation—that poorly designed economic policies enacted in response to early stages of the financial crisis significantly contributed to the Great Recession by distorting incentives and increasing uncertainty. The policy explanation suggests that government initiatives such as the 2008 tax rebate, the Troubled Asset Relief Program (TARP), the American Recovery and Reinvestment Act, Cash for Clunkers and U.S. Treasury mortgage modification programs aggravated early weakness in the economy and led to a full-blown recession.

Casey Mulligan, for example, studies the effect of Treasury mortgage modification programs on the employment rate; he finds that eligibility requirements for these programs raised implicit income tax rates on some households to levels exceeding 100 percent.⁸

John Taylor contends that a broad set of policies substantially contributed to the recession and supports his argument with a number of studies.⁹ In one recent article, for instance, he shows that some interest rate spreads, and both U.S. and foreign stock prices, deteriorated much more rapidly at the times of the TARP announcement and President Bush’s warning of a possible Great Depression than they did around the Lehman bankruptcy or other major financial events. In another study, he shows that daily sales at Target department stores dropped substantially right after the announcement of TARP on Sept. 19, 2008, but not immediately after the Lehman bankruptcy on Sept. 15. Taylor concludes that government policies contributed significantly to the recession, perhaps because policymaker communication regarding underlying economic strength increased public uncertainty.

Uncertainty, in fact, may be a primary reason why the recession deepened and persisted into 2009, well after the worst of the financial crisis. High uncertainty raises the value of delaying decisions in many economic models, which can depress economic activity. Recent and ongoing research on the impact of uncertainty on economic activity suggests that it can indeed induce recessions; in one forthcoming theoretical article, for example, uncertainty about the accuracy of government pronouncements regarding macroeconomic strength can lead households to reduce the labor hours they supply.¹⁰

Conclusion

Whether researchers lean toward the financial explanation, the policy explanation or another hypothesis altogether, it is clear that deeper exploration of labor markets is essential for understanding the Great Recession. The large labor distortion that occurred during the U.S. recession remains unexplained. Why similar distortions didn’t occur during previous postwar U.S. recessions, nor in high-income countries in 2007-09, is not understood.

Other questions are also unresolved. Factors behind large productivity deviations during the recession in other high-income countries must be explored. The relationship between distress in financial markets and the “real” economy—why the recession continued long after financial crisis abated—is unclear.

Fortunately, much promising research is under way:¹¹

• examinations of labor market distortions in earlier economic crises
• efforts to connect hypothetical financial events to labor deviations
• research linking use of corporate debt and labor markets
• analysis of how implicit labor income taxes can suppress employment levels
• study of productivity fluctuation due to resource misallocation from financial imperfections.

Clearly, much work remains to be done. Furthering this research will be essential not only to economists, but also to policymakers and other decision makers who will, inevitably, again confront the challenge of macroeconomic crisis.

The “secondary loan” market was the focus of much attention during the recent financial crisis. This market—where companies that originate loans sell them to other firms, often packaged as asset-backed securities—appeared to freeze up at the start of the crisis. Potential buyers seemed to lose confidence in the quality of the underlying assets being offered by loan originators. In short, the market was broken.

Policymakers launched several initiatives aimed at unfreezing the market, most prominently the Term Asset-Backed Securities Loan Facility (TALF); several other policies were proposed but not implemented. In retrospect, the advocates of these policies have suggested that they were largely successful in restoring health to damaged secondary loan markets.

In this paper, we analyze the ability of such policies to solve the secondary loan market problems and find that, on the contrary, they do not appear to have been responsible for resolving the underlying dysfunction. While these credit markets are unquestionably operating better now than previously, the reasons for their improved function remain unproven. We hint at policies with greater potential for addressing future episodes of such dysfunction, should they occur, but stress that these policies remain untested in both theory and practice.

The paper begins with a brief description of the market situation and policy response. We then lay out some of the economic theory that illuminates dysfunction in credit markets, highlighting two concepts in particular: adverse selection and reputational incentives.

We then proceed with a short description of our economic model based on these concepts, followed by policy exercises that use this model to analyze whether the programs proposed, and in some cases initiated, could actually improve function in secondary loan markets.

We conclude that, unfortunately, these policies were (or would have been) most likely ineffective, and possibly even counterproductive, and we suggest options that may be more successful in addressing future market crises of this sort. Such considerations are not mere academic concerns. This analysis has direct bearing on proposals that the newly enacted Financial Stability Oversight Council may consider in designing regulations for the so-called shadow banking system. It also should help policymakers in addressing future financial crises of a similar sort, if and when they occur.

Failing markets

In the fall of 2007, the total volume of new issuances of asset-backed securities fell abruptly after an almost uninterrupted climb since early 2000; by the fourth quarter of 2008, new issuances of ABS had virtually halted.

Figure 1 depicts this trend displaying the volume of new issuances of securities backed by assets for various categories, from student loans to subprime home equity, between 2000 and 2009. The consistent rise for almost all ABS categories continued from first quarter 2000 to fourth quarter 2006, climbing from about $50 billion to roughly $300 billion over that span. By third quarter 2007, the total fell to $100 billion, and then to near zero by the end of 2008. (Similar patterns, not illustrated here, have been documented for syndicated loans—that is, very large loans arranged jointly by several lenders for a single borrower.)

This collapse in new issuance volume coincided with a reduction in collateral values. The S&P/Case-Shiller U.S. home price index provides one clear example of this, with steady growth until late 2006 and abrupt decline throughout 2007. (See Figure 2 below.)

Other economists have suggested that a similar boom-bust cycle existed in the United States in the 1920s, and this is seen in Figure 3 below, derived from data on annual changes in publicly traded real estate bonds issued against single large commercial mortgages or pools of commercial and real estate mortgages.² Again, the trend is a steady climb from zero in 1919 to about $145 billion each year of the mid-1920s, followed by a collapse to roughly $50 billion issued in 1929. These large changes in stock of real estate bonds were likely associated with similar changes in the volume of new issuances.

The 2007-08 collapse in the market for such asset-backed securities was a cause for great concern among policymakers, who perceived it as an indication that the secondary loan market had become extremely inefficient. “Secondary markets have become highly illiquid, and are trading at prices below where they would be in normally functioning markets,” declared a U.S. Treasury Department fact sheet on March 23, 2009. Also in March 2009, the Federal Reserve Bank of New York asserted that “[n]ontraditional investors such as hedge funds, which may otherwise be willing to invest in [asset-backed] securities, have been unable to obtain funding from banks and dealers because of a general reluctance to lend.”

Policy response

These Treasury and Federal Reserve statements were drawn from documents concerning the proposed and/or adopted policy responses to the perceived market inefficiency. The Treasury Department proposed a Public-Private Partnership for purchasing assets held by distressed financial institutions, but this partnership was never implemented. The New York Fed proposed the TALF, which was quickly enacted.

Under TALF, the New York Fed was authorized to lend up to $200 billion on a nonrecourse basis (meaning that the lender can recover no more than the collateral pledged) to holders of AAA-rated ABS backed by new or recently originated consumer and small business loans. The intention was to increase credit availability and support economic activity by facilitating renewed issuances of consumer and business ABS at normal interest rate spreads. The New York Fed noted that as the ABS market came to a near-complete halt in October 2008, “interest rate spreads on AAA-rated tranches of ABS rose to levels well outside the range of historical experience, reflecting unusually high risk premiums.”³

To the extent that the interest rate charged by the Federal Reserve under TALF was below market interest rates, this program, which terminated in June 2010, was effectively a subsidy for the private purchase of assets in the secondary loan market. To the extent that the Fed charged market interest rates, it is not clear why it could have been effective. These observations will be evaluated later in this paper.

Also, of course, the Federal Reserve System rapidly lowered its target for the federal funds rate from 5.25 percent in the summer of 2007 to 2 percent by April 2008, and it now stands between 0 and 0.25 percent. The Fed also engaged in massive purchases of assets, a policy referred to as “quantitative easing,” which eventually lowered market interest rates in many related credit markets.

The Fed’s assessment

Credit markets, including the secondary loan market, have indeed improved since their nadir in the fall of 2008, but the question of whether this improved function was (or could have been) due to implemented (or proposed) policies has not been closely scrutinized. The Federal Reserve, for its part, does believe that TALF was effective in restoring efficiency and normal levels of liquidity to dysfunctional markets.

“Overall, the TALF performed impressively,” said Brian Sack of the New York Fed in a June 2010 speech.⁴ “The program contributed to a substantial improvement in conditions in the securitized credit market, facilitating an increase in the availability of credit to households and businesses.”

Sack acknowledged that other factors played a role in the increased efficiency of secondary loan markets: “To be sure, improvements in funding markets broadly and in the macroeconomic outlook during the course of the program clearly influenced the recovery of securitized credit markets.” Nonetheless, he asserts that TALF “has been widely credited with helping to jumpstart those markets.”

Theory on credit markets

Relying on anecdotal evidence is insufficient for rigorous policy evaluation. To better assess policy effectiveness, we must know what underlies function and dysfunction in credit markets; to do so, we developed a mathematical model based on economic theory. Economic research on credit markets generally, and asset-backed markets in particular, has developed rapidly in the wake of the recent financial crisis. But this leading-edge research is based on long-understood principles, including those of adverse selection and reputational incentives. Our model builds directly on these concepts, so a brief review of each is in order.

Adverse selection is the idea that in markets where buyers and sellers have different levels of (or asymmetric) information, some sellers—often those with goods of the highest quality—may exit the market. Much economic theory on this concept was sparked by economist George Akerlof’s celebrated 1970 paper, “The Market for Lemons,” which illustrated the idea with a used-car market.⁵ Potential sellers of high-quality used cars are likely to leave the used-car market, he showed, because if buyers are unable to judge quality, they will pay no more than an average market price. In the absence of a mechanism to better inform buyers about quality or guarantee their purchases (through “lemon laws”), bad cars will push out better cars, and markets will collapse.

Adverse selection is highly germane to secondary loan markets because loan originators (those who initiate the mortgage or other loan contract with the borrower) know the quality of the assets underlying the loan (the home’s market value, the borrower’s creditworthiness) better than the potential secondary buyer. Indeed, Akerlof’s article used credit markets as an additional example to illustrate the theory. There is therefore considerable potential for high-quality loans to exit the market, leaving behind only poor credit risks and bad underlying assets.

Since the mid-1980s, economists have studied adverse selection in asset markets and more recently in markets where assets are securitized. In our analysis, we assume that buyers of secondary loans have less information about loan quality, and there is substantial scholarship supporting this assumption. For example, a recent study by Downing et al. (2009) found that loans which banks held on their own balance sheets yielded more on average than those which they securitized and sold, indicating that they kept the high-quality loans and sold the “lemons.”⁶

Reputational incentive is the second central concept behind our analysis of secondary loan markets. Akerlof pointed out that certain practices and institutions have developed to counteract the effects of quality uncertainty: Guarantees, brand names, store chains and licenses certifying proficiency are examples. Each is a means of creating trust or confidence in quality of the good or service being sold, and might therefore defuse concerns about true value and adverse selection. Each, in short, seeks to build reputation.

In loan markets, trust is paramount, so those who seek to borrow or, by extension, to resell loans they’ve originated, have a strong incentive to establish a reputation of trustworthiness. These reputational incentives have been studied in a number of economic settings, from central banks to chain stores to predatory monopolists. In a 1989 analysis of debt markets, Diamond analyzed the reputational incentives that borrowers face in the markets where adverse selection is a problem.⁷ “The value of a good reputation rises over time, as does the cost of a default,” he argued. “If there is sufficient adverse selection, then a typical equilibrium path for a borrower … is to choose risky projects when ‘young’ and, if able to survive long enough without a default, to switch to safe projects from that point forward.”

Generally speaking, the economic literature suggests that the existence of incentives to build a good reputation improves economic welfare—that is, equilibrium outcomes are better in models with reputational incentives than in models without them. But more recent work has suggested that in some settings, reputational incentives can result in worse outcomes. If participation in a market is optional for short-run players and if actions by long-run players that encourage participation by short-run players can be interpreted as a signal that the long-run player is “bad,” then reputational incentives have bad economic consequences.

Like much game theory, that sounds quite abstract. To make it more concrete, the economists who have done this research use an example, as did Akerlof, from the automotive world.⁸ Consider car mechanics who have a choice of whether to replace a car’s engine (and charge the customer accordingly) only if it’s necessary, or to replace it regardless of whether the engine is faulty. If customers can only gauge mechanic quality by whether their car runs well after the visit to the mechanic’s shop, and bad mechanics increase profits by charging for repairs they don’t perform, then even good mechanics have a pecuniary incentive to become bad—that is, to charge for unneeded repairs. So, reputational incentives, interacting with adverse selection, can lead to bad outcomes.

A model of secondary loan markets

The same, we found, is quite true in secondary loan markets: Our analysis demonstrates that reputational incentives can lead to poor outcomes in these markets when adverse selection is present. In particular, our model of the secondary loan market demonstrates how adverse selection and reputation interact to yield abrupt collapses in loan volume, with increased inefficiency. This “freeze” in the secondary loan market is precisely what policymakers perceived during the U.S. financial crisis of 2007-09 and sought to address with a variety of initiatives.

We begin with a very basic model—we call it our benchmark—which is static: There is just one round of transactions in the secondary loan market, rather than a series carried out over time. There are three types of actors or agents in this model: a loan originator (referred to as a “bank” in the following discussion), a set of buyers and a set of lenders. Banks have one loan apiece (a home mortgage, for example, or an asset-backed security). A bank with a high-risk loan is considered a low-quality bank; those with low-risk loans are high-quality banks. Banks are also sorted by their expense levels as either high-cost or low-cost.

Buyers offer to buy the banks’ loans on the secondary market, and the primary decision of each bank is whether to hold onto its loan or to sell it to the buyer who offers the highest price. Lenders provide financing to banks that decide to hold onto their loans, receiving principal and interest at the going rate. In deciding which loan to purchase, buyers consider a bank’s reputation, which is the lender’s belief about the probability that the bank is high-quality.

Exploring the mathematical properties of this static benchmark model, we find that it produces an efficient allocation of loans. That is to say, with a single round of transactions between banks and buyers, loans will be allocated with complete economic efficiency to those parties with the highest comparative advantage. If a bank is a low-cost bank, it will hold its loan; if it is a high-cost bank, it will sell its loan to the highest bidder.

A dynamic model

But the situation becomes more complex—and interesting—when we move to a more realistic dynamic scenario in which banks, buyers and lenders are able to evaluate one another’s behavior in previous transactions before deciding what to do in the next round of transactions. This opens the door to concerns about reputation; because of asymmetric information—banks know more about the risk level of their loan than do potential buyers—there is potential for adverse selection. Banks with high-quality loans are more likely to hold rather than sell them, leaving a market full of low-quality (lemon) loans. But knowing that high-quality banks tend to stay out of the market, a bank with a low-quality loan might act strategically by holding onto its loan in one round to create a (false) reputation that it is a high-quality bank.

We find that unlike the static model, which resulted in a clear and unequivocally efficient outcome, this dynamic model with adverse selection and reputational incentives generates “fragile” outcomes, in two senses. The first type of fragility is that it isn’t immediately clear whether reputation concerns will lead to good or bad results—in the jargon of economists, the model has “multiple equilibria”—so both outcomes are possible. The model’s second fragility is that small drops in collateral values can generate large and abrupt collapses in new issuances on the secondary loan market, collapses associated with increased inefficiency.

Thus, our dynamic model—with reputation concerns and also the adverse selection that occurs with asymmetric information—ends up providing a very good testing ground for real world policies that seek to mitigate dysfunction in secondary loan markets.

A deeper look at fragility

Concerns about reputation arise with repeated transactions because actors in the model economy can look to the past and make judgments about other actors before deciding whether to engage in another transaction, just as a customer would return to a store if previous purchases at that store seemed reasonably priced and of high quality. Knowing this, the store will try to offer products of good quality at reasonable prices, or at least try to convey that impression. In other words, it will attempt to build its reputation.

Similarly, a bank in our model will—in deciding whether to sell or hold its loan—bear in mind the effect of its action on its reputation. But our model demonstrates that it isn’t clear cut whether that reputational concern will result in good outcomes or bad. The dynamic model produces two mathematically correct solutions—equilibria—one good and one bad.

In the good outcome that we call the “positive reputational equilibrium,” high-quality loan originators have incentives to sell their loans at a current loss because they want to improve their reputation so that they can obtain higher prices in the future. In the bad outcome, the “negative reputational equilibrium,” loan originators who sell are perceived to have low-quality loans. That perception convinces banks with high-quality loans to hold onto them even if it isn’t profitable to do so. In this second outcome, then, the volume of loan issuances is smaller than in the good outcome, and (under specified conditions) market efficiency suffers.

The second type of fragility in this model economy is superficially similar to the first: A small change in a fundamental economic value—in this case, loan collateral—can generate a dramatic change in an aggregate market value: an abrupt collapse in loan issuances on the secondary market. This result is, of course, remarkably akin to the real world outcome during the recent financial crisis, and that helps form the base for our policy analysis.

The model’s ability to generate the latter type of fragility can be seen in the following two graphs. The first graph below (Figure 4) depicts the sell/hold decision threshold for high-quality banks (those with low-risk loans). According to the model’s mathematics, the curve represents the cut-off line for a bank in judging whether to sell a loan, depending on the market value of its collateral. At a collateral value of 4, banks with reputation levels below roughly 0.65 hold their loans and those with higher reputations sell. This means that if collateral values fall from 5 to 4, a large segment of banks—those with reputation levels roughly between 0.4 and 0.65, will decide to withdraw from the secondary loan market. Put otherwise, the graph illustrates that as collateral value falls, the adverse selection problem worsens, and only the lowest-quality banks (with highest-risk loans) remain in the market.

The second graph (Figure 5) displays the volume of lending trade, the fraction of all banks that sell their loans, as a function of collateral value. This shows that as the market value of a loan’s collateral (its default value) decreases from 1.3 to 1.1, the volume of trade collapses by half, from 60 percent of banks selling their loans on the secondary market to just 30 percent. (We also found that this second type of fragility doesn’t depend on whether the market is in the positive or negative reputational equilibrium. The secondary loan market can collapse regardless.)

Was policy effective?

While building this complex model of the secondary loan market is rewarding from a research perspective, contributing to the academic literature on both reputation concerns and financial market behavior, we believe it also has substantial value in allowing for evaluation of proposed and implemented policies that sought to address dysfunction in secondary loan markets. Rather than examining the details of these specific programs, we analyze two general policy types:

• Policies by which the government would purchase asset-backed securities at above-market value (similar to the TALF and to the Public-Private Partnership plan that was not enacted).
• Policies that decrease the costs of loans held to maturity (which include changes in the fed funds target rate and increased deposit insurance levels from the Federal Deposit Insurance Corp.).

Buying toxic ABS
When analyzed with our dynamic adverse selection model, policies under which the government would offer to purchase so-called toxic assets at prices above current market value would in all cases involve transfers to banks and imply that the government will make negative profits.

If prices offered to banks are below the level that prevails in our positive reputational equilibrium, market outcomes will not significantly change. Our model shows that banks with high-quality loans would enjoy no reputational gains by selling to government and would continue to stay out of the secondary market. Only banks with low-quality loans would sell to government, with no net benefit to the economy.

If, on the other hand, prices offered by government were sufficiently high, the purchase policy would leverage reputational incentives and could overcome adverse selection problems. Still, the government would, through its transfers to banks, lose money in this effort to unfreeze the market.

Lowering rates
We then look at policies of lowering interest rates so as to decrease costs of holding loans to maturity. If the government reduces current interest rates and leaves future rates unchanged, our model shows, the policy will aggravate the lemons problem in secondary loan markets by encouraging banks with high-quality loans to retain rather than sell their loans. If, on the other hand, the government leaves current rates unchanged but commits to reducing future rates, it can improve current allocations but will make later allocations less efficient by increasing banks’ incentives for holding onto their loans. And, of course, in the future the government would face strong incentives not to hold to its earlier commitment to reduce rates and thereby increase allocation inefficiency.

Other policy options
An alternative policy that we analyze with the model is forced asset sales. Under this policy (not proposed), government would randomly select banks and require them to sell their loans. The policy would by force generate a pool of loans in secondary markets, just as requiring home mortgage owners to purchase home insurance ensures a wide risk pool. However, this standard solution to adverse selection problems would come at a cost of loan misallocation: In some instances, low-cost banks would be forced to sell their loans, reducing the market’s overall efficiency in terms of comparative advantage.

Another alternative would be for the government to commit to purchasing assets in the future at prices contingent on signals about loan value. Our model shows that such a policy would support the positive reputational equilibrium, meaning that reputation concerns would overcome adverse selection problems and result in efficient market allocations. The feasibility of such a policy deserves further research, but would necessitate a model in which governments can commit but private parties cannot.

Conclusion

The volume of new loan issuances dropped sharply in the secondary loan market during the recent financial crisis, and U.S. policymakers responded with a variety of proposals aimed at restoring normal market function, including purchase of assets at above-market prices and reducing the costs of holding loans to maturity.

We have built a model of the secondary loan market in which its primary economic function is to allocate loans to those institutions—originators or secondary owners—that have a comparative advantage in holding and managing them. Because loan originators are better informed than potential purchasers about their loan quality, the markets suffer from adverse selection. We use a dynamic adverse selection model of the secondary loan market to determine whether reputational incentives improve or aggravate market outcomes.

Our model has fragile outcomes in the sense that it generates sudden collapses in new issuance volume due to small changes in collateral value. Such collateral drops and market collapses, associated with increased market inefficiency, resemble those seen empirically in late 2007 during the U.S. financial crisis.

We therefore use the model to analyze programs that were proposed and in some cases implemented by policymakers to address loan market dysfunction and find that they do little to resolve the market’s inherent adverse selection problem. We conclude that, unfortunately, these policies were (or would have been) most likely ineffective, and possibly even counterproductive, and we suggest options that may be more successful in addressing future market crises of this sort. Such findings have direct bearing on proposals now under consideration vis-à-vis regulatory design for segments of the financial industry that are currently subject to little oversight and regulation.

Little is certain about the United States’ fiscal future beyond this: Given foreseeable trends in economic growth, future tax revenues will not cover forecasted mandatory and discretionary expenditures; therefore, a large and growing budget deficit is highly probable.² While policymakers may be able to enact modest spending cutbacks, they will undoubtedly need to consider options for raising taxes as well.

Unfortunately, when they do so, they will face a further unpleasant economic reality: Taxes often introduce distortions and inefficiencies that depress economic activity. Indeed, taxes generally undercut the incentive to generate the income on which they are levied.

This economic policy paper addresses that quandary by offering an option with a number of appealing features:

• It allows governments to raise revenues without the labor-discouraging distortion common to income taxes.
• Its elimination of economic distortion contributes to economic activity and well-being.
• Because it allows citizens free choice to opt for an alternative tax arrangement, it is politically viable.

To be specific, this paper suggests that a tax buyout program could achieve the goal of raising revenues without distorting work incentives and thereby diminishing economic activity. The buyout is a contract between the government and individual citizens whereby each person has the option in each tax period to pay a fixed price in exchange for a set reduction in his or her marginal tax rate for a given period (say, one year).

We call it a “buyout” because it allows individuals who purchase the contract to effectively pay off a percentage of their regular (and distortionary) taxes with a lump-sum payment to the tax collection authority. Participation is voluntary and involves no risk from the individual citizen’s point of view: Only those who would gain from entering the contract in a given period (after any uncertainty about their labor income is resolved) will do so.

This paper begins by discussing the distortion problem addressed by this plan, including background from related research. It then describes the model we developed to analyze how a tax buyout program would work in a dynamic macroeconomy, including quantitative estimates of the impact such a program might have on the U.S. economy during a time of high fiscal pressures such as those now present. We conclude with suggestions of further issues that should be addressed to make tax buyouts a concrete policy option—an important goal in a period of substantial and growing fiscal deficits.

The views expressed here are ours, and not necessarily those of others in the Federal Reserve System.

Background and description of tax buyouts

The idea of a tax buyout focuses on an issue that is central to economic analysis: the disincentive effect of taxation. Taxes are sometimes imposed on activities that society wishes to discourage, such as smoking or pollution; in such cases, the disincentive is intentional. But when a government seeks to generate revenue by imposing taxes on a worker’s earnings, it spites itself. A tax on labor income discourages work because the worker knows that each hour of labor will generate less take-home pay. The resulting decrease in work effort leads to less economic output, which in turn leads to a lower tax base—undercutting the revenue generation that is the very goal of imposing such taxes. That distortion of economic incentives is a key challenge to tax policy, and to economic research as well.

In an effort to understand how to design a better tax system, British economist and Nobel laureate James Mirrlees analyzed these “labour-discouraging effects” in a classic 1971 paper, and much of modern “optimal fiscal policy” research—including ours—is based on his model.³ Mirrlees recognized that any labor tax system must cope with “asymmetric” information: A citizen knows more about his or her ability to work than does the government. Given that asymmetry, how do policymakers design a tax system that provides maximum incentive to work and minimal distortion to labor supply, economic growth and revenue generation? Mirrlees’ solution was to design a system that is “incentive compatible,” meaning that it must give workers a pecuniary incentive to reveal their true work abilities—that is, it must be designed such that workers’ self-interest will induce them to provide as much labor as they can.

That, indeed, is the intent of our buyout proposal: To establish a tax scheme that draws forth as much work effort as possible by offering individuals the chance to purchase a “buyout” contract that decreases their marginal tax rate. And our challenge is to determine whether there is, in fact, a contract price low enough to draw people into the program, but high enough to generate sufficient revenue to fund the buyout scheme and other government expenditures. Our analysis suggests that such a program is not only feasible, but also quantitatively significant and politically viable.

It should be noted that other economists have suggested related ideas. During a milder economic downturn in the early 1990s, Harvard economists Alberto Alesina and Philippe Weil proposed a two- (or more) tiered tax schedule under which taxpayers could purchase a lower marginal tax rate.⁴ “The tax payers who select to ‘buy’ the reduction in the marginal rate, i.e., who choose the new tax schedule, will be the most productive workers: under the new tax schedule they will work and consume more. … [The] introduction of the second tax schedule does not reduce total tax revenues. More generally, additional revenue-neutral or revenue-increasing Pareto improvements can be achieved.”

Similarly, in a 1994 paper, the University of Michigan’s Joel Slemrod and his co-authors investigated a two-bracket income tax structure and found that “a second tax bracket allows the lower marginal tax rate on high-wage people to coax out … greater labor supply from the most productive segment of society, with the increased tax revenue used to lower the tax burden of the least productive segment. Although the calculated optimal tax system features declining marginal tax rates, it still generally features increasing average tax rates, so that it is progressive but not graduated, in the standard sense of these terms.”⁵

This research provides important background but does not include several elements that are potentially important to fully evaluate the impact of these schemes. These previous models are static—analyzing economies at just one point in time—and the people acting in these models are essentially identical to one another in every way but work ability. Our research extends this idea into an economy that is dynamic (it evolves over time) and incorporates “heterogeneous agents” (meaning that people in our model vary substantially in numerous characteristics relevant to labor supply, income and taxes). In addition, we look at the broad macroeconomy and also at the idea of tax schedules that are “nonlinear” (tax rates for different income brackets can vary dramatically—tax rate graphs are curves, not straight lines). We believe that this appraisal renders our buyout scheme a pragmatic proposal that, with refinement, could be used to address current challenges in fiscal policy.

Step-by-step analysis

We first conduct an abstract exercise with a mathematical model of a national macroeconomy to see if the tax buyout idea is sensible at a theoretical level. Do the basic relationships among critical variables in our model economy—tax rates, labor supply, consumption levels and the like—result in the buyout plan inducing enough extra labor, and therefore extra output and tax revenue, to more than pay for itself? The answer is yes.

While our model is designed to represent crucial economic incentives and relationships, like all such models, it abstracts from reality in a number of respects. Nonetheless, in our analysis we take a step-by-step approach to incorporating increasing levels of realism into the model and at each step evaluate whether the important result of a tax buyout drawing forth additional work effort can be achieved.

We begin with a very basic model: an economy that examines static relationships between individuals and government with a constant tax rate and perfect information. By “perfect” information, we mean that the work capability of every individual is public knowledge: Tax collectors know how much work everyone is able to perform, so pretending to be disabled to avoid work and collect government insurance benefits isn’t an option. In this (unrealistic) case, our model demonstrates that the offer of a contract to reduce an individual’s tax rate in exchange for a set contract payment will be accepted by everyone in the population, will increase total labor and well-being, and will leave government revenues unchanged.

To understand the logic of this result, consider a simplified example. There are two workers: Alice and Ben. Alice earns a high labor income, while Ben earns a low one. Both pay taxes, and the more they earn, the more taxes they pay. If the government knows their ability exactly, it can offer them a tax buyout contract that involves Alice and Ben paying a fixed amount (high for Alice and low for Ben) in exchange for a reduction of their tax rate to zero. If the fixed amounts are chosen equal to the pre-buyout tax receipts, the government will not lose money from the contract.

But what incentive do Alice and Ben have to buy into this program?

The key difference between standard taxation and the tax buyout program is that in the tax buyout, the government asks for a fixed amount. So if Alice or Ben works an extra unit (another hour, day or week, say), she or he is the sole beneficiary of the extra revenue—they no longer have to pay a portion of it to the government as under standard taxation. This increases their incentive to work and thus will increase their income and, ultimately, their well-being.

In other words, by lowering an individual’s tax rate, the tax buyout removes what Mirrlees called the “labour-discouraging effect” of labor income taxes, thereby eliminating the inefficiency due to distortionary taxation. That releases a surplus that can then be shared by both government and individuals.

We then extend this to the more realistic scenario considered by Mirrlees in which information about work ability is imperfect, or “asymmetric”: The government does not know how much work every person is capable of. Can a tax buyout program create the incentive compatibility that Mirrlees showed is necessary?

Going back to the Alice and Ben example, now the government does not know which one is the more productive worker and thus cannot offer a tailored contract (a high price for Alice and a low one for Ben), but instead must offer a single contract. In this case, we find that the buyout contracts can nonetheless be priced at a level high enough to generate positive revenues for the government, but low enough to attract enough individuals to buy them.

In the Alice and Ben example, if the government offers the buyout at the price equal to the pretax liabilities of Alice, then Alice will take the buyout (and this will increase her labor effort and well-being), while Ben will typically not take it, and so his welfare will be unaffected. Still, the buyout is socially desirable because part of the population gains, while another part does not lose.

Some might contend that because Ben, a poor person, is unaffected, while Alice, who is rich, is gaining from the buyout, the program could increase the gap between rich and poor—an arguably unfair outcome. In the paper, we argue that it is possible to construct buyout schemes in which all people, including poor workers, can be made better off by the introduction of the buyout, even when they don’t participate in it directly. The idea again is that the contract generates a surplus that can be shared. With a properly designed buyout plan, the government can receive and redistribute some of this additional surplus so that the entire population benefits, not just the most productive.

Real-world relevance?

The step-by-step analytical modeling demonstrates that the tax buyout idea has substantial theoretical merit. But that leaves aside the issue of quantitative importance. That is, given actual levels and distributions of economic and demographic variables (such as household earnings, wealth levels, tax and interest rates, life span and retirement length), would a tax buyout program have any real dollars-and-cents impact on a multitrillion-dollar economy? Or is this merely an interesting academic proposition without practical application?

To answer this question, we write a more detailed artificial model economy with overlapping generations of heterogeneous (in terms of abilities and luck) households who make labor decisions, consume and accumulate wealth over their lifetimes. We then put this model through a process called “calibration”—essentially, setting the model’s parameters so that its basic predictions capture aspects of actual U.S. households that we think are crucial for our policy experiment.

In particular, we calibrate our model to ensure that

1. households in the model have the same wealth and labor earnings distribution as households in actual U.S. data for 2006, and
2. the shape of the tax function (i.e., the equation that assigns a household’s tax liability as a function of its total earnings and family composition) is consistent with actual U.S. tax code.⁶

A key parameter for our model economy is the so-called Frisch elasticity of labor supply, a measure of how much workers change their labor supply in response to a change in wages (or taxes), keeping everything else (including their wealth) constant.

This parameter is crucial for our question because if workers are not very responsive to wage or tax changes, then taxes are not very distortionary—that is, tax increases or decreases hardly affect overall welfare and labor supply. If that’s the case, the tax buyout, which operates through reduction of distortions, will not yield large benefits.

A very large literature in economics has tried to estimate Frisch elasticity, but economists are still uncertain. In our work, we start by considering a value that lies in the middle range of existing estimates, but we also experiment with different values.⁷

Generating answers

After this calibration process, we run the model through many computer simulations to generate numerical answers for the questions we’re interested in:

• What percentage of people will purchase a buyout contract at a given price for a specified reduction in their tax rate?
• What effect will that have on the hours of work they supply?
• How will that affect government tax revenue?
• To what degree will this change in labor supply (through a reduction in tax distortion) alter the nation’s economic output?

Our strategy is to consider an economy with a set level of government spending and no tax buyout plan (for example, the U.S. economy before the recent financial crisis), which then unexpectedly faces a 20 percent jump in public expenditures, due, say, to a financial sector bailout or sharply higher Medicare costs (the U.S. economy post-crisis). We then consider two scenarios: one without the buyout offer and one with it.

In particular, we consider the following buyout option: Each citizen has the option of reducing his or her labor income taxes by 5 percent for one year by paying the government the fixed price of $4,500. The contract is very simple to understand and to accept or reject. 

An example may help to make the option more concrete. Consider again our friends Alice and Ben. We’ll assume that Alice, the more productive worker, earns a labor income of $100,000, while Ben’s labor income is $30,000. At the time of filing her taxes, Alice would find it advantageous to accept the buyout because her take-home pay will be $500 higher. In contrast, Ben will not buy the contract because doing so would actually reduce his take-home pay by $3,000.⁸

Note that accepting or rejecting the buyout would not involve any additional risk for either Ben or Alice (the decision is taken at the time of filing taxes), but the essential element is that Ben and Alice know that the buyout is an option at the beginning of the year, when they decide how much to work. Notice that if Alice knows of the buyout option, she will in general work harder, because she can retain more of the additional dollars she earns, and her additional work is the key social and private benefit of the buyout.

In both cases (with and without buyout), we assume that the government will raise taxes to finance the additional expenditures so that the budget is balanced in every period. By comparing those scenarios, we can judge the quantitative impact of a variety of buyout plans. And because we use a dynamic model, we’re able to estimate results over a span of 20 years.

Quantitative results

In our first experimental run-through, we find that in the scenario without the buyout, taxes as a fraction of total income need to rise (in order to balance the budget) from roughly 21 percent to 26 percent. With the tax buyout option, however, taxes would rise to just 24.5 percent.

Given that government expenditures are identical in both scenarios, why would buyouts result in lower average taxes? Because, according to the model, over 8 percent of the population will purchase the buyout contracts, thereby generating additional government revenue. This transformation of part of government revenues from a tax that distorts labor decisions to a lump-sum payment that does not is the essence of the tax buyout contract. And it does so in a revenue-neutral fashion without making anyone worse off.

The reduction in work-supply distortion—the decrease in what Mirrlees called the “labour-discouraging effects” of income taxes—is quantitatively important. Labor supply with the tax buyout scheme is 0.33 percent higher than without it because those who buy the contracts choose to work harder (since their marginal tax rate is lower). Moreover, those people tend to be the most industrious workers, so there is an increase in average labor productivity. Therefore, while national economic output (or alternatively, national income) drops because taxes had to increase to fund higher government spending, it drops less with the buyout program, about 1 percent less. Due to higher overall taxes, wealth and consumption decrease, but the decrease is less severe with buyout contracts.

Changing assumptions

We then run the model under a few different scenarios, changing the size of the buyout, making its price age-dependent and altering the estimate of worker responsiveness to wage changes. The table below shows the results, compared with the results in the baseline scenario, reported in the first row.

As the table indicates, increasing the tax buyout size (or, alternatively, the tax rate reduction) from 5 percent to 10 percent (column 1, rows 1 and 2) means nearly a tripling in price (from $4,500 to $12,900) and half as many buyers. As expected, reducing the size (row 3) lowers the contract price and increases program participation. The larger buyout scenario still has a significant impact on GDP; the smaller buyout less so.

Tax Buyout Scenarios

Interestingly, if the price of the buyout contract is varied according to the purchaser’s age (row 4), similar to life insurance pricing, it will attract more buyers and generate a bit more revenue. This is because older people have higher wages on average, would benefit more from the reduction of distortion provided by the tax buyout and, hence, are willing to pay a higher price.

As discussed previously, a crucial parameter for evaluating the effectiveness of the buyout is the Frisch elasticity of labor supply. In the table’s last row (5), we show results when we consider a low elasticity value. In this more conservative case, the benefits of the buyout are smaller than in the baseline case but remain significant, with gains in GDP exceeding half of 1 percent.

Finally, we looked at how things change over time to get a sense of which types of people are most likely to buy the contract, not just now but in the future. This is one of the clear advantages of using a dynamic rather than a static model. One way of looking at a tax buyout is that it’s an opportunity to buy, for a fixed price, a subsidy on one’s labor income. And because the subsidy is calculated as a percentage of income, the benefits are greatest for those who earn—or expect to earn—high labor income. Bottom line: The people most likely to buy the buyout contract now or in the future are

• high-wage (and therefore older) people
• people who are patient (because they value the possibility of earning a lot in the future) and
• people with little wealth (because lower wealth induces individuals to work harder).

Our computer simulations find considerable differences over time among people. The types of individuals just listed would significantly benefit from introduction of a tax buyout program even when they don’t participate in it initially. An obvious example is young people: Even if they are not buying into the contract now, they will probably earn higher labor income when they’re older and therefore be more likely to participate. The program’s existence, and the possibility of (literally) buying into it in the future, is highly valued. Thus, in a dynamic world evaluated over the long run, the benefits of a tax buyout program spread well beyond the fraction of people who participate in it at any single point in time.

Further work needed

Before a buyout program is designed and implemented, a number of concerns call for further investigation. By the same token, several promising possibilities could lead to significant improvements in buyout strategy.

The first concern is what economists call a “general equilibrium effect.” One consequence of introducing a tax buyout program is that prices (in particular, wages) will change, and perhaps in a direction that is disadvantageous for some. Specifically, the tax buyout’s reduction in incentive distortion will result in a labor supply increase. That could reduce wage levels in general and hurt in particular the low-wage, low-productivity people who are least likely to buy the contract. This effect deserves quantitative investigation because its impact likely depends on factors not considered in our model, such as the openness of capital and labor markets.

Another concern arises in regard to the distribution of high and low labor income within the total population. The issues here are complex, but they come down to two basic questions: Would the program benefit only high earners, rendering it socially less desirable and politically unpalatable? As we discussed earlier, a possible solution to this issue is to accompany the buyout program with a redistribution policy (financed by the buyout itself) to assist low earners.

And secondly, are there so many high-labor-income people in the population, or people of such high labor income, that offering them the chance to lower their tax bill would significantly undercut general tax revenues? Future research should therefore investigate the benefits of limited buyouts, in which a person’s gain from tax reduction is limited to a specified multiple of the contract price. For instance, what if the tax benefits for a buyout contract were limited to, say, twice the contract purchase price? What labor supply, tax revenue and GDP impact would such a program have?

On the more encouraging side, there are many directions in which this buyout idea could be extended to reduce labor effort distortions still further. For example, varying the contract pricing schedule for individuals of high and low work ability could have a beneficial impact. Another possibility: In our current setup, we assume completely asymmetric information, meaning that the government knows essentially nothing about individuals’ work abilities. In reality, of course, the government knows quite a bit about its citizens—education levels and earning history, for example—and could alter contract prices accordingly.

Third, it seems likely that labor supply elasticity—again, sensitivity to changes in wage levels—differs among individuals: Some people will respond more than others to a $5 wage hike. In our quantitative experiments, we plug in just one value for the entire population, but in fact, people with high elasticity would be more likely to buy tax-reducing contracts, leading to higher program participation. And lastly, the tax buyout idea could be expanded to capital income—stock dividends, for instance—and further analysis should estimate the combined effects of buyout programs offered for both labor and capital income.

Conclusion

We believe that a tax buyout initiative is a promising means of addressing likely revenue shortfalls in the United States. By offering citizens the opportunity to decrease their marginal tax rate in return for a fixed payment, governments could reduce the negative impact that labor income taxes have on labor supply decisions, thereby increasing total work effort, raising overall economic output and well-being, and generating higher tax revenues.

Our initial analyses suggest that tax buyout programs can have significant quantitative importance in a national economy, especially at a time when high fiscal needs call for high levels of distortionary taxation. Prior to designing such a program for public implementation, a number of concerns should be addressed and several possibilities for improvement considered. Also, the effects and consequences of such a scheme could be evaluated with alternative methods, for example, by running small-scale experiments such as introducing the buyout for local and state taxes in small communities.

In the mid-2000s, we—as investors, home buyers and bank lenders—collectively bet that house prices would not fall by 30 percent in most major metropolitan areas in three years. We were wrong. This mismatch between our expectations and our realizations was the ultimate source of the financial crisis of 2007-09.

The Congress of the United States is currently considering legislation to restructure financial regulation. However, no matter how well-written or how well-intentioned the legislation may be, no law can completely eliminate the kinds of collective investor and regulator mistakes that lead to financial crises. These mistakes have taken place periodically for centuries. They will certainly do so again. And once these crises happen, there are strong economic forces that lead policymakers—for the best of reasons—to bail out financial firms. In other words, no legislation can completely eliminate bailouts. Any new financial regulatory structure must keep this reality in mind.

In this paper, I describe an approach to financial regulation that takes as given the inevitability of bailouts. The basis of the approach is that the magnitude of bailouts can be limited by taxes on financial institutions. I arrive at this conclusion about the usefulness of taxes by thinking through an analogy that I’ll develop at some length. I will argue that, knowing bailouts are inevitable, financial institutions fail to internalize all the risks that their investment decisions impose on society. Economists would say that bailouts thereby create a risk “externality.” There is nearly a century of economic thought about how to deal with externalities of various sorts—and the usual answer is through taxation. I will suggest that the logic that argues for taxation to deal with other externalities is exactly applicable in this case as well. The views expressed here are mine, and not necessarily those of others in the Federal Reserve System.

The size of the optimal tax for any given financial institution may depend on a host of risk-related attributes, and so may be difficult for supervisors to calculate and implement. I suggest a possible alternative: a market-based method to compute the appropriate magnitude of the tax. Roughly speaking, for a particular institution, the government should sell bonds that pay a variable coupon linked to the size of the transfers (that is, bailouts or other government assistance) received by the institution or its owners. The prices of these coupons will reflect the market’s judgment of that institution’s risk profile and can therefore be used to set the size of the tax that should be imposed.

It is important to distinguish my notion of a risk externality from two other types of externalities that are mentioned in discussions of bank regulation. One of these is a systemic externality. The failure of a given Bank X may affect the profitability of many other firms in the economy even though Bank X has no direct contracts with those firms. In this sense, any decision by Bank X that increases its probability of failure has a systemic implication, because it also increases the expected losses by the entire financial—and indeed economic—system.²

My notion of a risk externality is also distinct from what might be termed a fire sale externality. During financial crises, many financial institutions may have to sell assets or collateral at the same time. These simultaneous sales will put downward pressures on the assets’ prices. A given financial institution will not internalize the impact of its sales on the price of other institutions’ assets.³

I downplay these two externalities because governments typically eliminate their effects through targeted interventions during financial crises. Governments can correct a systemic externality by preventing the failures of financial firms through bailouts.⁴ Governments can stop fire sales of assets by purchasing the assets or being willing to treat them as collateral in loans to the relevant firms. Indeed, in the recent financial crisis, the United States government and the Federal Reserve System intervened precisely to address externalities of these kinds (although admittedly not in the case of Lehman Brothers).⁵

The inevitability of bailouts

In the crisis of 2007-09, governments made large transfers to claimants of financial institutions. Some onlookers have argued that future legislation should seek to eliminate these payments. In my view, many such payments are unavoidable in the context of a financial crisis. These payments assure depositors and debt holders that their financial interests in the relevant financial institutions will be backed by the government. Why does government provide such assurance? There are several reasons, but I believe that the most important concerns the prevention of “runs.”

Imagine that Bank X needs $100 billion of one-day loans to survive. This means that for a given lender to be willing to make a $1 billion, one-day loan to Bank X, that lender has to believe that Bank X will get another $99 billion in one-day loans. In this situation, Bank X could fail simply because every possible lender believes correctly that no other lenders are willing to lend to Bank X. Such a crisis of confidence can occur regardless of the true condition of Bank X.

This story is hardly a new one. It’s exactly why we have deposit insurance: to prevent runs by reassuring bank depositors that their money is safe. But the story has huge consequences for how governments operate. In a financial crisis, there is a tremendous sense of uncertainty. There are some truly insolvent financial firms out there—but no one knows for sure which they are. And during a crisis, the panic in the air means that any institution—even one with solid fundamentals—may be subjected to a run if its investors lose confidence in its solvency.

In such an atmosphere, contagion effects become extremely powerful. Even a slight loss by one short-term creditor can lead all short-term lenders to rush to the safety of Treasury bills. Such flight would endanger the survival of key financial institutions, even if they are fundamentally sound. Governments cannot risk such systemic collapse, and so during times of crisis, they end up providing debt guarantees for financial institutions. Thus, policymakers inevitably resort to bailouts even when they have explicitly resolved, in the strongest possible terms, to let firms fail.⁶

Many observers of the events of September 2008 have emphasized the need for better resolution mechanisms. Different people mean different things by this, but most want to impose losses on debt holders. I’m not opposed to faster resolutions of bankruptcies. But I do not believe that better resolution mechanisms will end bailouts. No matter what mechanisms we legislate now to impose losses on creditors, Congress, or some agency acting on Congress’ behalf, will block them when we next face a financial crisis. And Congress will do so for a very good reason: to forestall a run on the key players in the financial system.⁷

Debt guarantees result in excessively risky, inefficient investment

I have argued above that government concerns about runs make debt guarantees (that is to say, bailouts) inevitable at least in severe financial crises. Here, I argue that these guarantees lead to inefficient investments by financial institutions.

Imagine for a moment that we live in a world without bailouts, so that the government does not provide debt guarantees or deposit insurance. If a financial institution decided to increase the risk level of its investment portfolio, its debt holders and depositors would face a greater risk of loss. By way of compensation for that greater risk, they’d demand a higher yield. As a result, in the absence of government guarantees, financial institutions would find it more costly to obtain debt financing for highly risky investments than for less risky ones. This effect, on the margin, would curb a firm’s appetite for risk. It would have an especially powerful effect on highly leveraged financial institutions, because high debt-to-asset levels mean higher risk of being unable to fulfill debt obligations.

But now return to the real world, with deposit insurance and debt guarantees, and the inevitability of government bailouts. Even if they only kick in during financial crises, these guarantees change this natural market relationship between risk and cost. The depositors and debt holders are now partially insulated from increases in investment risk, and so they do not demand a sufficiently high yield from riskier firms. Financial institutions are no longer as deterred from undertaking risky ventures by the high costs of debt finance. And this missing deterrence is especially relevant for firms that are highly leveraged, because they should be paying out especially high yields on their debts.

Note that the problem here is created by the expectations of depositors and debt holders, not the expectations of the financial institution itself. Because the depositors and debt holders sometimes expect to receive a bailout, they accept a lower yield on their investments. The financial institution is then able to finance high-risk, high-return investments at low cost. The institution itself does not care why the funds are so cheap.

In this way, the expectation of bailouts leads to too much capital being allocated toward overly risky ventures. These misallocations of capital don’t create the collective mistakes in predictions that generate financial crises. But the misallocations do mean that society loses a lot from those mistakes—far more than is efficient.

An externality analogy

The problem I’ve just described of bailout inevitability and the relationship between debt guarantees and inefficient investment is well-known. Less understood, perhaps, is how closely related it is to a standard policy issue in economics: pollution.⁸

Think about a firm with a factory. The firm has to make a decision about how much output to produce at the factory. In doing so, it trades off the revenue gain associated with expanding production against the costs of producing that extra output. Unless required by law, it does not take into account the environmental cost associated with any pollution generated by the factory.

The pollution is an externality: It is a cost borne by society that is external to the purely market considerations that shape the firm’s decision. The presence of this externality means that the firm will choose to overproduce according to society’s standards because the firm’s costs are lower than the full societal costs.

Now return to the decision problem of a financial institution that is financed in part with guaranteed debt. As we have seen, the debt guarantee implies that taxpayers absorb some risk of the financial institution’s investments, allowing the institution to ignore that risk when choosing among investments. Hence, it is ignoring some portion of the costs of its decisions, and will therefore choose to overproduce high-risk investments.

Notice the analogy between the financial institution and the polluting firm. The firm increases production because it can ignore some costs that are borne by society. Similarly, the financial institution increases the risk level of its investments because the government guarantee allows it to ignore some costs (in the form of risk) that are borne by society. Debt guarantees create a risk externality. This connection is a useful one, because economists know a lot about how to design policies to address externalities like pollution. We can apply those lessons to great effect when thinking about optimal financial regulation.

The externality analogy and a tax solution

In this section, I use the externality analogy to develop an appropriate regulatory response to the risk externality created by government guarantees. Again, consider a firm with a factory that generates pollution. It is reasonable to presume that the firm can influence the pollution level in many different ways, including the following:

1. The amount of time that the firm runs the factory during the workweek.
2. The kinds of antipollution technology used at the factory.
3. The kind of energy used by the firm to run the factory.

Potentially, the government could regulate the firm’s pollution levels by controlling any or all of these choices. However, to do so, the government has to choose how to trade off these three (and other) factors against one another. Among other considerations, the government’s decision will be influenced by cost. If antipollution technology is cheap, the government may simply require the firm to invest in that. If antipollution technology is expensive, the government may require the firm to switch to using natural gas instead of coal. These trade-offs require the government to acquire a tremendous amount of firm-specific information and perform cost-minimization exercises for each and every factory. Such a task is clearly infeasible for any government to perform on a national level for all relevant industries.

The solution to this difficulty is to regulate the amount of pollution produced by the firm, not how the firm produces that pollution. The problem is that pollution has a social cost that the firm does not internalize when choosing its level of production. However, the firm will choose the efficient level of pollution if it is required to pay for its full social cost. More concretely, suppose that the firm is told, before choosing its level of production, that the government will

1. Measure the amount of pollution that the firm generates.
2. Charge the firm a tax that is exactly equal to the social cost of that quantity of pollution.

This policy generates a tax schedule that translates the amount of pollution generated into an amount paid by the firm. If the firm knows that it faces this tax schedule, its costs of production will include the social cost of pollution. In this way, what was external to the firm becomes internal. As a result, the firm will choose the socially efficient level of production. Just as importantly, it will automatically choose to produce that pollution—and its other outputs—in a cost-minimizing fashion.

This (well-known) solution to the pollution problem has an exact analog in the risk externality problem generated by debt guarantees. Currently, regulators are trying to combat the risk externality by having distinct regulations for financial institution capital, liquidity and incentive compensation.⁹ All of these measures are likely to mitigate the inefficiencies created by risk externalities. Again, though, the optimal trade-off between these various measures is likely to depend on a host of firm-specific information that will be hard to acquire.

For example, regulators are considering requiring financial firms to defer payments of incentive compensation. Such deferrals would make the compensation plans less attractive to employees. Firms will therefore have to increase their average wage bill to retain employees, by amounts that depend on subtle characteristics of both firms and workers.

The pollution analogy suggests how regulators can sidestep these difficult choices. Instead of regulating all of the financial institution’s decisions, the government should tax the financial institution for the amount of extra risk that it produces. In this scenario, the financial institution is told that after it chooses its investments, the government will

1. Estimate the expected present value¹⁰ of the net payments made by government—the cost of “pollution,” if you will—to the financial institution or its stakeholders.
2. Charge the financial institution a tax that is exactly equal to the above estimate.

This policy generates a tax schedule that translates the financial institution’s choices into an amount paid by the firm. This amount equals the extra cost borne by the taxpayers, appropriately adjusted for risk and the time value of money. If the financial institution knows that it faces this tax schedule, its private costs of financing an investment are now equalized to the social costs of doing so. Its investment choices will be efficient—as will its choices of capital, liquidity and incentive compensation, factors that current reform proposals address in a less precise manner.

Using markets to compute the right tax

Calculating the appropriate tax for a polluting firm requires measuring the quantity of emissions and then pricing those emissions for the costs they impose on society. The former is beyond my ken, but certainly within the expertise of environmental engineers. The latter can be (and in some cases, is being) accomplished via market mechanisms (such as carbon taxes or “cap and trade” emissions markets).

Similarly, computing the appropriate tax for a financial institution with debt guarantees requires measuring the quantity of taxpayer risk and then pricing that risk. The latter can be accomplished through options markets, which are designed specifically to price risk accurately. But how should the government go about measuring the quantity of risk? There are at least two possible methods, one that relies on regulatory monitoring and another that depends on markets.

Quantifying by government
If the government can observe the payoffs of the financial institution’s asset portfolio, then this problem is (at least conceptually) straightforward.¹¹ Good information about how well a financial institution’s investments actually do over time, or are likely to do, provides a clear picture of risk levels inherent in the firm’s investment decisions.

But good information may not be readily available. Put more technically, the probability distribution of the financial institution’s asset portfolio’s payoffs may be private information, known only to the institution (or its employees), not to government supervisors. In this situation, many different attributes of the financial institution may inform the supervisor about its assets’ payoff probability distribution. The supervisory authority should use all of these financial institution attributes to arrive at an estimate of the quantity of risk.

There is a useful analogy in private insurance markets. Consider the pricing of homeowners’ insurance. The insurer would like to link the insurance premium to how well the homeowner takes care of the home. But this is impossible without constant monitoring of homeowner behavior, an infeasible task. The pricing of insurance therefore ends up depending on various clues that have proven reliable guides to how homeowners treat their homes. Thus, the insurance price will be based partly on whether the home has a fire extinguisher. More subtly, the premium may also depend on the homeowner’s driving record, since good drivers also tend to be good homeowners.

This same logic applies to the regulation of financial institutions. Suppose two financial institutions both use incentive compensation plans for their investment managers. However, one institution defers managerial bonuses and the other does not. It is natural, then, for the government supervisor to presume that managers of the latter institution will choose investment projects with more extreme risk. That presumption should be reflected in the supervisor’s judgment about the quantity of taxpayer risk and, ultimately, in the tax paid by the financial institution.¹²

A market-based approach to quantifying risk
This kind of analysis seems daunting however, because it is likely to require monitoring an enormous number of financial institution attributes. For this reason, I believe that a market-based approach is at least complementary and possibly superior.¹³

Here’s what I have in mind. Suppose that, for every relevant financial institution, the government issues a “rescue bond.” The rescue bond pays a variable coupon equal to 1/1,000 of the transfers actually made from the taxpayer to the financial institution or its stakeholders. (I pick 1/1,000 out of the air; any fixed fraction will do.) Much of the time, this coupon will be zero. However, just like the financial institution’s stakeholders, the owners of the rescue bond will occasionally receive a large payment. In theory, or in a perfectly functioning market, the price of this bond is exactly equal to the 1/1,000 of the expected discounted value of the transfers to the financial institution’s stakeholders. Thus, the government should charge the financial institution a tax equal to 1,000 times the price of the bond.

Notice that this approach could be used for a wide variety of financial institutions, including nonbanks. In principle, the government need not figure out in advance which institutions are systemically important and which are not. Instead, the market would provide this information through the pricing of rescue bonds.

Markets for rescue bonds may prove to be thin and illiquid. In these circumstances, it would be inappropriate to rely only on market measures to compute the appropriate taxes. However, even when they are imperfect, market measures would contain valuable information that should be an input into the supervisory process.¹⁴

Conclusion

In this note, I’ve argued that to prevent runs, governments provide debt guarantees to firms in the financial sector. These guarantees create a risk externality, as those firms do not bear the full costs of their investment choices. Regulation should control and, if possible, eliminate that externality, because it leads to inefficiently risky investment.

There are numerous proposals for financial regulatory reform in the wake of the events of 2007-09. Several proposals, such as leverage caps, capital requirements and controls on incentive compensation, can help mitigate the risk externality problem. However, it may well be difficult for a government to figure out the optimal trade-off among these proposals on a firm-by-firm basis. Instead, a well-designed tax system can entirely eliminate the risk externality generated by debt guarantees to financial institutions. Figuring out the right tax may be complicated, but the task can be eased using appropriate information from financial markets.

Endnotes

References

Brunnermeier, Markus, and Yuliy Sannikov. 2009. “A Macroeconomic Model with a Financial Sector.” Working Paper (November): Princeton University.

Chari, V. V., and Patrick J. Kehoe. 2009. “Bailouts, Time Inconsistency and Optimal Regulation.” Unpublished paper (November): Federal Reserve Bank of Minneapolis.

Clement, Douglas, and Christopher Phelan. 2009. “Incentive Compensation in the Banking Industry: Insights from Economic Theory.” Economic Policy Paper 09-1 (December): Federal Reserve Bank of Minneapolis.

Feldman, Ron J. 2010. “Forcing Financial Institution Change Through Credible Recovery/Resolution Plans: An Alternative to Plan-Now/Implement-Later Living Wills.” Economic Policy Paper 10-2 (May): Federal Reserve Bank of Minneapolis.

Flannery, Mark J. 2010. “What to do about TBTF?” Unpublished working paper, (April): University of Florida.

Haldane, Andrew G. 2010. “The $100 Billion Question.” Comments at the Institute of Regulation & Risk (March 30): Hong Kong.

Hart, Oliver, and Luigi Zingales. 2009. “To Regulate Finance, Try the Market.”

Jeanne, Olivier. 2008. “Dealing with Credit Booms and Busts: The Case for Prudential Taxation.” Working Paper (November): Johns Hopkins University.

Jeanne, Olivier, and Anton Korinek. 2010. “Managing Credit Booms and Busts: A Pigouvian Taxation Approach.” Working Paper (April).

Kocherlakota, Narayana R. 2010. “Taxing Risk.” Comments at the Economic Club of Minnesota (May 10): Minneapolis, Minn. Video of speech.

Phelan, Christopher. 2009. “A Simple Model of Bank Employee Compensation.” Working Paper 676 (December): Federal Reserve Bank of Minneapolis.

Stern, Gary H., and Ron J. Feldman. 2004. Too Big to Fail: The Hazards of Bank Bailouts. Brookings Institution.

Wall, Larry D. 1997. “Taking Note of the Deposit Insurance Fund: A Plan for the FDIC to Issue Capital Notes.” Economic Review (First quarter): Federal Reserve Bank of Atlanta.

Appendix: Three Possible Concerns with the Market-Based Approach

In this appendix, I discuss three possible concerns with the market-based approach.

• Suppose a government makes a transfer to a financial institution, but that transfer is then paid to a given lender to fulfill an obligation. Should that transfer be credited to the lender’s rescue bond or to the financial institution’s rescue bond?
• How would the market-based approach account for financial institution assets that are transferred to government in exchange for bailout transfers?
• How would the market-based approach deal with the issue of financial institution decisions made over time, rather than the static analysis offered above?

Borrower/lender ambiguity
Suppose Bank A borrows from Bank B, and the government guarantees that B will receive its payments. During a crisis, this guarantee could be implemented in one of two ways: The government could pay A and then A pays B, or the government could simply pay B directly. Under my proposed market-based approach, the different choices would manifest in different outcomes for the owners of rescue bonds. If B is paid indirectly through a government payment to A, then the owners of A’s rescue bond receive a coupon payment. If the government pays B directly, then the owners of B’s rescue bond receive a coupon payment.

Fortunately, how the government resolves this ambiguity does not affect the efficacy of the rescue bonds. The loan from Bank B to Bank A is a transaction that offers benefits to both banks. The expected transfer from the government—regardless of whether it’s made to A or B—is distorting because it increases the joint benefits of the loan transaction. To eliminate the distortion, the government needs to levy a tax that cancels out those joint benefits—and that tax can fall on A, B or both of them. Indeed, in principle, there would be no efficiency losses if the government were to levy the tax on a wholly distinct third party C. The presence of the tax would still undo the distorting effects of the subsidy by providing an incentive for C to pay A and B not to undertake the loan.

Assets in exchange
When governments make transfers to debt holders or depositors, they often receive some of the financial institution’s assets in exchange. Rescue bonds should be based on the net transfer to debt holders, taking account of those exchanged assets, not the gross transfer. However, valuing the assets received by governments in exchange may well be difficult, especially during the heart of the crisis.

This problem can be addressed by keeping track of the payments received from the assets exchanged. Thus, suppose the government receives a bundle of mortgages from a struggling bank. It ends up holding those mortgages for a year before selling them. During the year, it receives payments from the homeowners, and at the end of the year, it receives a final payment. The financial institution’s rescue bond should reflect these payments by paying a negative coupon equal to 1/1,000 of these payments.

Of course, bonds with negative coupons do create difficulties. (How should the government collect from bondholders?) To deal with this issue, the government can require the rescue bond to pay a relatively large positive fixed coupon C (instead of a zero coupon) when no transfers are made to or received from the financial institution. The coupon goes up by 1/1,000 of any transfer made to the financial institution or down by 1/1,000 of any transfer received from it. The appropriate tax is then 1,000 times the difference in price between the rescue bond and a bond with fixed coupon C.

Sequential choices
In the earlier discussion, I described how to compute the tax in a static context, in which the financial institution is making a single investment choice. How should the tax be adjusted in light of new investment choices or in light of information about past choices?

Here, forward markets can play a useful role. Suppose a rescue bond is issued in 2010, and there is a forward market for 2011 delivery of the bond. When 2011 arrives, the spot price of the rescue bond may be higher or lower than the forward price set in 2010. If the spot price is higher, we can conclude that there has been an unexpected increase in the value of the transfers to be received by the financial institution. If the spot price is lower, there has been an unexpected decrease.

These changes in prices can be used to align the financial institution’s private incentives with social ones. Specifically, in 2011, the financial institution should be charged a tax equal to 1,000 times the difference between the 2011 spot price of its rescue bond and the price set in 2010 for 2011 delivery of the rescue bond. This tax will ensure that the financial institution internalizes the impact of new information about its choices and actions on future government transfers.

Note that, as time unfolds, the annual tax may well be negative. In this case, market participants have received information that the financial institution is less exposed to the risk of failure than had been anticipated earlier. The government is implicitly subsidizing the financial institution for reducing its risk profile.

Governments were not prepared for the potential failure of systemically important financial institutions (SIFI). SIFI failures and near deaths threatened to spill over to other financial firms, markets and the economy at large. Governments bailed out SIFI creditors to quash these spillovers. Creditors anticipating future bailouts will not check the risk-taking of SIFIs. Too much risk-taking by SIFIs means we will face another financial crisis absent effective reform.

In response, policymakers and analysts have called for “living wills” for financial institutions.¹ A living will for humans stipulates in advance the medical steps a terminally ill patient wants. A living will for financial institutions, under many popular conceptions, stipulates in advance the steps government and firms will take to produce an orderly resolution at a fatally weak financial institution.² Orderly resolutions obviate the need for bailouts. Living wills seem responsive to the prior lack of preparedness.

Unfortunately, such planning-focused living wills are not credible and will not prevent spillovers.³ First, absent countervailing forces, SIFIs have organized themselves and operate in ways that produce spillovers and bailouts. Neither firm nor government can fix those problems at the last minute. Second, under the living will analogy, the firm draws up the living will. SIFIs have no incentive to draw up credible plans. Finally, creditors will not view living wills drawn up in private as real threats to future bailouts. Thus, creditor discipline will remain too weak.

Resolution plans with the following three attributes offer a more credible tool to reduce expectations of government bailouts. The plans should do the following:

1. Force changes to firm structure and operation and supervisory powers long before firms become financially weak (as some observers have also noted)⁴;
2. Be prepared by supervisors based on firm input; and
3. Allow for transparency of plan-induced changes to outsiders.

Such resolution plans also offer important advantages over alternative reforms such as banning certain activities from banking organizations or breaking up SIFIs. And there is some evidence that credible resolution planning can make the fallout from financial crises less severe.

Before I discuss attributes of credible resolution planning, I will briefly describe living will reforms more generally.

The Gist of Living Wills

There is no shared understanding for what precisely constitutes a living will, hence, the multitude of names.

In general, proponents view the wills as tools to facilitate resolutions during grave times with minimal spillovers to the economy. Many versions of living wills would contain information such as the following:

• Catalogs and/or lists of key information needed during a crisis (e.g., lists of assets, lists of subsidiaries and their locations, summaries of key laws governing financial institution failure);
• Responses to “what if” scenarios (e.g., how would the financial institution raise more capital or liquidity, including from the sale of assets?); and
• Assessments on the state of firm readiness (e.g., capabilities of management information systems, “scores” assessing the ability of the plan to facilitate recovery/resolution).

Living will proposals may require parallel efforts by both the financial institution and the regulator/supervisor. The government agent would also have to consider their capabilities (e.g., legal, personnel, access to financial resources) to facilitate an orderly recovery or wind-down of the firm.

I now turn to the three attributes I think resolution plans must have to prove credible. The absence of such traits makes many living will proposals ineffective.

Ex Ante Change

The decisions that put financial institutions at risk for failure often occur over many years. In contrast, the firm can fall into unexpected, even fatal, financial distress in a very short period of time. Rapid decline makes it impossible for the financial institution to execute unplanned, significant actions to save itself. A financial institution that does not have effective databases to value and account for its assets cannot build such systems quickly. Likewise, firms cannot rapidly undo practices embedded in the firm’s structure or operations that hinder recovery or resolution. How quickly can a firm untangle thousands of interconnected subsidiaries used to own, fund and manage assets?

Supervisors, for their part, cannot suddenly acquire the legal powers or staff to facilitate an orderly resolution over a weekend.

To be credible, resolution plans must lead to changes in financial institutions’ and supervisors’ operations before a crisis hits. How might such a credible resolution plan process work?

Before a financial crisis, bank supervisors would identify those attributes of financial institution operations, structure and funding that curtail effective, bailout-free recovery or resolution. The recent crisis suggests several attributes of SIFI operations that may need ex ante changes, including but not limited to: (1) information systems that either do not adequately capture key information on assets or are not accessible by business lines in the firm, (2) legal structure and geographic scope of firm operations that make resolution difficult to implement, and (3) interfirm funding or hedging strategies that make it difficult to sell off valuable parts of the firm to facilitate recovery.

The identification would feed supervisory action plans for the firm. The plans would identify appropriate changes to scale and scope of the SIFI’s activities. Focusing on ex ante change saves the public from outdated plans or government agencies consulting heretofore dormant thousand-page plans during a crisis.

Supervisory Preparation

Financial institutions do not have incentives to plan for their demise. In particular, financial institutions would not spend resources determining how their demise spills over to others, let alone take steps to limit such fallout. Such activities impose costs but no benefits to the owners and employees of the financial institutions.

Government agents should act on behalf of society as a whole and seek to account for the spillovers from financial institution failure.

The respective incentives of government agents and financial institutions suggest that government supervisors/regulators prepare the resolution plan. This role seems particularly sensible if the plans will lead to important changes in firm and supervisor operations.

Supervisory leadership in plan development does not exclude financial institutions from the process. Information provided by the financial institutions will prove the critical raw input for plan development. But supervisors must determine what information to gather and identify the implications from that information.

Transparency

Observers have raised doubts about the ability and desire of financial institution supervisors to act in the public interest. At a minimum, critics note that supervisors did not prevent the type of risk-taking that led to the financial crisis. Others have taken a more critical view, arguing that supervisors are “captured” by the financial institutions they supervise. Captured governments made matters worse by, for example, approving firm mergers that raised both the systemic risk and political power of financial institutions. These critics would not find resolution plans driven by supervisors a credible tool to address systemic risk.

The government should address such credibility concerns head on even if policymakers disagree with the underlying arguments. If creditors of financial institutions agree that resolution planning is not credible, they will not view the planning process as putting themselves at greater risk of loss. As a result, creditors will continue to expect bailouts and will underprice financial institution risk-taking. Society will have too much financial institution risk if the price is too cheap.

Requiring resolution plan transparency will make the plans credible to creditors, leading to better pricing of risk. What form should transparency take? Congress could require supervisors to report on changes required at specific financial institutions. Policymakers could also bring in experts outside the supervisory process to review resolution planning effectiveness. In a more comprehensive approach, Congress could require supervisors to identify changes they have mandated in financial institution organization, activity, operation or size to reduce systemic risk. I have previously called for a system of “macroprudential ratings” that would achieve this goal.⁵

Comparing Options to Credible Resolution Plans

I have emphasized using resolution plans to reduce systemic risk by changing financial institution and supervisory operations. Some argue that society could accomplish the same outcome by limiting or banning the involvement of financial institutions in activities or operations that pose too much systemic risk. “Breaking up” firms also would force significant change to firm operations. This rule-based approach seems to have the benefits of the resolution plan through a credible, more straightforward approach.

The resolution approach is superior, because it generates better information on precisely what activities and operations of financial institutions and supervisors need reform. Observers and supervisors have a general sense of the type and level of activity posing systemic risk. But there is not yet common agreement on the relative contribution of a given activity to systemic risk. Does the assumption of certain asset concentrations pose the largest contribution to systemic risk of a firm? Or is it participation in select securities transactions? Perhaps the manner in which a firm organizes itself internationally poses the largest contribution to systemic risk. A legislative rule may target activities that analysts ultimately conclude pose second-order systemic risk concerns. Incorrect targeting could have serious opportunity costs, as it encourages policymakers to “let down their guard.” The resolution plan approach, in contrast, would rely on the acquisition of detailed private information on a firm-specific basis before identifying activity and size restrictions.

In addition, a rule-based approach often offers “one bite at the apple.” Firms will find ways to comply with the rule while still posing the same amount of ex ante systemic risk. The resolution plan approach requires repeated interaction to respond. Finally, resolution planning has the advantage of testing the new resolution regimes that many hope will facilitate imposition of losses on creditors.

Would Resolution Planning Make A Difference?

Legislative proposals before Congress would require preparation of resolution plans for systemically important financial institutions. The proposals are fairly general and generally provide some discretion to the government entity that would implement the requirement. The Senate Banking Committee proposal does put explicit rules around using the plans to require divestiture.

I think legislators could make the plans a credible check against systemic risk by ensuring that they have the three attributes discussed above. I make that claim not just on the logic already articulated, but also on examples where government-forced changes to firms and markets have made resolution and recovery more orderly.

For example, in July 2008, the FDIC issued a final rule requiring the largest insured banks to take steps allowing the FDIC to differentiate between insured and uninsured deposits.⁶ The FDIC argued that it could provide unnecessary insurance protection to uninsured creditors absent such changes. The banking industry largely opposed the rule. Some in the industry argued that changes to systems should only occur once a bank faced difficulty. Otherwise, the rule would require “unnecessary” changes in operations for healthy banks. The FDIC rejected this strategy recognizing that planning in advance a la living wills was insufficient. Action, even if costly, must occur before firms get into trouble.

Issues around legal structure, intercompany funding and hedging, information systems and legal powers of resolution raised concerns of spillovers for firms like AIG and Lehman Brothers.⁷ The resolution planning I have advocated would take on such issues directly before a crisis. It seems reasonable to conjecture that such planning could have reduced the potential spillovers from the demise of these firms.

Forcing change for firms and supervisors will not be easy, particularly for financial institutions with operations and supervisors located around the globe. Absent resolution planning, however, it is not clear how hurdles to orderly recovery and resolution will be overcome. We must take steps in good times to reduce huge bailouts in bad.

Global financial markets have recently experienced their most significant crisis since the Great Depression, and they remain under considerable, if diminished, stress. The Federal Reserve System pumped trillions of dollars into the U.S. economy in a determined though imperfect effort to restore liquidity and steady markets. Now that markets are beginning to stabilize, the Fed is considering how best to reabsorb some of that liquidity so that the vastly expanded money supply doesn’t create inflation as the economy slowly revives.

This brief note reviews three broad strategies that are now under consideration for managing monetary reserves in the United States: (1) paying interest on excess reserves held at the Fed,¹ (2) managing interest rates on short-term deposits at the Fed, and (3) selling back financial assets such as mortgage-backed securities. It then suggests that several potentially large dangers associated with the first two strategies have been overlooked, while a possible weakness of asset sales—currently, the least-favored of the three strategies—has been overstated. The best course, I assert, is a careful blend of all three approaches, with strong emphasis on gradual sales of assets.²

It should be noted that these views are the author’s alone, not necessarily those of others in the Federal Reserve.

Preface to the discussion

Before I elaborate on this argument, it is important to emphasize that from the standpoint of economic theory, these strategies are essentially identical. That is, if markets were “complete” in the sense that no practical impediments—imperfect flow of information, differing access to markets, frictions in market functioning—prevented their smooth operation, then any and all of these strategies would have equal efficacy. Such impediments may affect market outcomes over the very short term, but are unlikely to be very important over a horizon of two or three years.

Consequently, the question of which strategy is pursued is not of fundamental significance to the course of the macroeconomy. The choice of one or the other will neither create nor prevent a Great Depression or Great Inflation. Nevertheless, the question of how best to manage reserves is not unimportant. Because markets are not complete and since frictions do exist, each strategy can have a significant effect on both the economy and the conduct of future policy. Careful consideration is therefore in order.

Background

The U.S. and world economies have undergone and continue to undergo a severe financial crisis. The dramatic increases in various yield spreads (such as that between interest rates banks charge one another for loans, the LIBOR, and rates on comparable-term Treasury bills—the so-called TED spread) in the fall of 2008 together with a dramatic increase in the likelihood of failure of a number of financial institutions have led to large changes in the scale and scope of central bank activities throughout the world.

In particular, the Federal Reserve System has increased its balance sheet to a historically unprecedented size relative to the economy. The accompanying chart shows the Fed’s balance sheet liabilities. As is apparent, the balance sheet has risen dramatically since the summer of 2008. From the end of August 2008 to the middle of January 2010, deposits held by depository institutions at the Fed have risen from less than $20 billion to over $1.1 trillion. Excess reserves have risen over the same period from roughly $2 billion to over $1 trillion.

It is reasonable to suppose that banks have been willing to hold such a large volume of assets at the Fed because short-term interest rates offered by the market have declined to essentially zero, while the Fed has been paying interest at an annualized rate of 25 basis points (0.25 percent) on balances held at the Fed.

The markets and the Fed are concerned about the implications of this large stock of reserves on the course of future monetary policy and future inflation. The primary source of this concern is that, if the stock of “high-powered” money—that is, currency and reserves that are very liquid—is maintained at current levels, the stock of broader monetary aggregates—that is, the money supply as a whole—will rise dramatically.

Indeed, if the reserve-to-deposit ratio (that is, the “reserve ratio”) returns to its precrisis levels, we can expect broad money aggregates to increase by a factor of roughly 50,³ which is highly likely to lead financial markets to expect a large increase in inflation.

This reasoning suggests that the Federal Reserve System must plan for an exit strategy if and when banks choose to reduce their holdings of excess reserves.

Three strategies

To repeat, three broad strategies are being considered that would allow the Fed to manage reserves over the next few years. Two of them focus on inducing banks to maintain reserves at proper levels. The third focuses on selling off financial assets that were purchased by the Fed to stabilize asset markets and inject liquidity into the financial system.⁴ (The Fed is also considering another tool: reverse repurchase agreements. Reverse repos are, for all practical purposes, similar to strategies that effectively pay interest on reserves.)

First, consider those strategies that focus on managing reserves by inducing banks to maintain healthy levels of reserves at the Fed. If banks draw down their excess reserves too quickly as they seek to use these funds for alternative purposes (such as making personal loans or business investments), the economy will be flooded with more liquidity than it may be able to handle at still weak levels of macroeconomic activity, giving rise to increased inflation as, in the oft-used phrase, “too many dollars chase after too few goods.” Managing reserve flows can therefore contain money supply growth to levels that will encourage economic growth without overheating price levels. Two Fed strategies focus on management of reserves.

Strategy 1: Raising interest rates on overnight reserves

Currently, U.S. banks hold more than $1.1 trillion of reserves with the Federal Reserve System. To restrict excessive flow of reserves back into the economy, the Fed could increase the interest rate it pays on these reserves.⁵ Doing so would not only discourage banks from draining their reserve holdings, but would also exert upward pressure on broader market interest rates, since only rates higher than the overnight reserve rate would attract bank funds. In addition, paying interest on reserves is supported by economic theory as a means of reducing monetary inefficiencies, a concept referred to as “the Friedman rule.”⁶

While this strategy, discussed at some length by Fed Chairman Ben Bernanke in recent congressional testimony,⁷ is attractive in several ways, it also has one very evident drawback. Because the Fed has never held such a large volume of reserves and has only recently been authorized to pay interest on them, neither the Fed nor financial markets has much experience in communicating and interpreting the signals conveyed by adjusting those rates.⁸

The consequent uncertainty about the intent and effect of any such rate changes may induce a great deal of volatility in the quantity of reserves. And by the same token, large rate changes may be required to induce the flows sought by the Fed. In either case, large movements may well be interpreted by markets as changes in the course of future Fed policy and would therefore hamper the Fed as it pursues its dual mandate of low inflation and maximum employment. The volatility due to this signaling problem is inherent to the transient nature of reserve holdings—banks can withdraw or deposit them literally overnight, and interest rates could conceivably change just as rapidly.

The Fed is well aware of this volatility issue and has therefore suggested that it may instead favor using a different tool to contain the flow of reserves.

Strategy 2: Offering banks higher interest rates on short-term deposits

This second strategy avoids the volatility provoked by the signaling problems of reserve interest rates, and the Fed seems most intent on pursuing it. With this mechanism, the Fed would offer depository institutions term deposits, similar to certificates of deposit purchased by bank customers, by creating a term deposit facility that would auction large blocks of term deposits at attractive interest rates on balances held for periods of three months, six months, or longer. This tool would convert excess reserves into short-term deposits that could not be used for very short-term investments or loans and therefore would not immediately increase the money supply. The Fed will begin testing transactions of this nature this spring.

Again, strategy 2 holds one clear advantage over strategy 1, the benefit that explains why the Fed may well prefer it. Because term deposits lock up reserves for several months, rather than days, they would stimulate far less volatility than overnight reserves and may therefore lead to less confusion than movements in the overnight reserve rate and what those rate movements might or might not signal about long-term Fed policy.

Unrecognized concerns

There are, however, two significant concerns inherent to both of these strategies: interest rate risk and rollover risk, considerations that have been largely overlooked in debate over the proper exit strategy. These considerations suggest that a third strategy, asset sales, may be the preferred approach.

Interest rate risk

Banks generally prefer to borrow short term and lend long term, because short-term interest rates (such as the rates banks pay their account depositors) are generally lower than the yield they expect to receive on their longer-term investments and loans (such as mortgages). But the difference between short- and long-term rates can shift dramatically, and this so-called interest rate risk is a concern to banks in the current climate of likely but hard-to-predict tightening of monetary policy.

If the Fed were to pursue either strategy 1 or strategy 2, it too would expose itself to considerable interest rate risk. If interest rates turn out to be higher than now anticipated by markets, either strategy would be quite costly. So while borrowing short and lending long is superficially attractive—since short-term rates are customarily lower—this approach could well end up being quite costly if interest rate volatility requires a sequence of expensive short-term borrowing. 

This risk is neither far-fetched nor negligible: Imagine a short-term interest rate of 4 percent on the approximately $1 trillion of excess reserves. This $40 billion cost is greater than the “profits” the Fed now turns over to the Treasury each year.

Rollover risk

A second concern is the risk inherent to using short-term claims to fund long-term investments. When long-term investments are backed by shorter-term debts that have to be rolled over—that is, they mature, and the borrower has to seek a new loan—there is real potential that the borrower will not find willing lenders if confidence in that borrower’s creditworthiness has eroded. Such risk is the cause of bank runs, but it affects firms, governments, and central banks as well.

A salient example was seen in March 2008, when Bear Stearns was unable to obtain short-term financing because rumors about liquidity problems began to spread through financial markets and investors lost confidence. Though adequately capitalized by the Fed’s supervisory standards, Bear Stearns nearly collapsed because the liquidity problems became self-fulfilling. Similar problems undermined Lehman Brothers and other firms during the recent crisis. Such rollover risk was not at all anticipated by the experienced financial analysts at these firms, or by the counterparties involved.

Emerging market economies also borrow short and lend long. That is, they have assets that mature over long periods of time and fund them with a sequence of short-term loans. But as seen in Mexico in 1994, for instance, lenders can quickly lose confidence in the ability of a government to manage its economy and refuse to roll over loans made to the nation. For Mexico and other emerging countries, such sovereign debt crises can cause the near collapse of the national economy and threaten international financial stability as well.

The use of strategy 1 or 2 would similarly expose the Fed to rollover risk, in effect, because it amounts to using short-term debts—excess reserves or term deposits—to fund long-term assets. Of course, the Fed differs from private firms and emerging markets in that it can “create” money to finance its debts. And indeed, that ability may well lead to hubris on the part of policymakers—similar to that seen among financial managers in the current crisis who were clearly overconfident in their ability to obtain financing.

Regardless of such self-assurance on the part of policymakers, if market participants lose confidence in the Fed’s ability to obtain funds from lenders, the Fed would have to pay very high interest rates to obtain short-term debt. That in itself would constrain monetary policy, and this is the main source of risk from both strategies 1 and 2: A self-fulfilling, high-inflation equilibrium in which expectations that the Fed will pursue lax monetary policy because banks demand a high-inflation premium will lead banks to demand that high-inflation premium.

Strategy 3: Gradual sales of financial assets

Given these potential risks, both acknowledged and largely overlooked, of strategies 1 and 2, a third strategy—the sale of securities now held by the Fed—bears fuller consideration than it has heretofore received.

As of the end of March 2010, the Fed will have purchased $300 billion in Treasury securities, $1.25 trillion of mortgage-backed securities (MBS) guaranteed by federal agencies, and about $175 billion of other agency debt securities. Gradual sales of these assets would decrease the quantity of reserves held by the Fed, because purchasers would draw upon these reserves in making purchases. It would also reduce the overall size of the Fed’s balance sheet.

Many observers fret, however, that such asset sales would depress prices of these securities if sales occur over a short period. As The Economist put it recently, banks and others “All worry that the eventual sale of these securities could wreak havoc.”⁹ The housing market in particular could be adversely affected, some fear, if MBS were dumped quickly and mortgage rates jumped. It’s notable that in his February 2010 testimony, Chairman Bernanke sought to reassure Congress and the public about such concerns with the explicit statement: “I do not currently anticipate that the Federal Reserve will sell any of its security holdings in the near term.”¹⁰

Given the risks of strategies 1 and 2, however, the potential impact of selling off Fed assets bears closer scrutiny. Would it truly wreak havoc, as many assume? If the Fed manages asset sales judiciously, it is my judgment that such an outcome is highly unlikely.

How segmented are financial markets?

The extent to which asset prices would be depressed depends on how segmented financial markets are—that is, how difficult it would be for investors to shift assets from one market to another over a reasonable time frame. Bear in mind that the volume of traded financial assets of all kinds in world markets is on the order of $200 trillion.¹¹ Given this extremely large financial market, if the Fed were to sell $1 trillion in assets over a period of time, the odds are small that such sales would have a big effect on global financial markets, assuming assets can be shifted among markets with relative ease.

Is it really valid to make this 1-to-200 comparison, that is, to compare the volume of the assets that the Fed holds on its balance sheet—Treasury securities, mortgage-backed securities, and the like—to the entire stock of financial capital? If financial markets are very segmented from one another—in other words, if markets for individual securities are affected very heavily by supply/demand considerations in that specific, narrowly defined market—then such a comparison is not valid. 

For instance, it’s likely that the market for securities backed by credit card receivables is a distinct market with a small outstanding stock. If so, trades in that market will have a disproportionate effect on prices. But if market participants have the ability to shift holdings across a wide variety of asset types—from mortgage-backed securities to securities backed by credit card receivables, and vice versa, for instance—then the impact on prices in individual markets is going to be muted. “Segmented markets” therefore reflects the notion that markets for financial securities are very narrow and very specific, so supply/demand conditions in that particular market have a huge effect on prices.

How much segmentation truly exists among such markets? If you look at them minute by minute, it’s clear that markets are highly segmented; but over longer time horizons, such as two or three years, people clearly have time to shift their portfolios. The question ultimately is, where is that break point where market segmentation ceases to be quantitatively and materially important? While there is little empirical research on this question with regard to mortgage-backed securities, for instance, it is my judgment that over a reasonable time frame, market participants would be able to absorb the sale of the Fed’s security holdings without significant price impact.

A prudent strategy

Therefore, it seems prudent to pursue a strategy that centers on a preannounced program of slow but steady sales of long-term government debt and mortgage-backed securities that market participants anticipate will happen over a period of time. Such a program would be unlikely to have a substantially depressing effect on these prices.

Because sales of Fed securities would be gradual during that ongoing divestiture, the Fed may also employ to a limited extent strategies focused on reserve interest rate levels—that is, strategies 1 and 2. Financial markets would understand that the Fed is relying primarily on asset sales to temper inflation, so less attention would be directed to reserve interest rates as signals of monetary policy; and with less reliance on strategies inherently subject to interest rate risk and rollover risk, the Fed would lower its vulnerability to unforeseen financial trends and the hubris that can affect policymakers as well as financial managers.

Such a joint strategy is likely to have the highest probability of success in draining reserves with minimal risk.

Again, it bears emphasis that this debate and the ultimate choice of how the Fed reduces the size of its reserves is not a primary and fundamental macroeconomic issue. While significant, it is not as great a concern as the bailouts and issues of moral hazard raised by past, and possibly future, support for financial institutions deemed too big to fail. Still, policymakers should understand that the debate over the proper method of drawing down monetary reserves as part of the Fed’s exit strategy has often overlooked concerns of considerable import and exaggerated risks that should prove minimal if a prudent strategy is wisely implemented.

Endnotes

These are difficult questions, but fortunately, the use of some standard tools of economic theory can provide useful insights. In particular, the theory of mechanism design, whose architects were honored in 2007 with the Nobel prize in economics, sheds light on how institutional rules can be designed to achieve best outcomes.

Mechanism design relies on the notion that rules must be “incentive compatible”—that is, the rules provided should channel people’s natural self-interest toward the institutional goal. Otherwise, people tend to game the system by withholding information or being dishonest. The tools of mechanism design are particularly germane to questions of incentive compensation—indeed, the term itself mirrors incentive compatibility.

So, what can mechanism design theory tell us about how a bank should design an incentive compensation system that will best achieve its goals, and would such an incentive system conflict with the goals of society as a whole? To address these issues we’ve considered an economic framework of how banks (and similar financial institutions) structure job contracts with bank employees.³

This framework allows us to evaluate a number of relevant questions, including:

• Should compensation be a constant salary, or should it depend on outcomes of investments made by the employee on the bank’s behalf?
• What is the best timing for incentive payments: as soon as the investment is made or deferred until investment outcomes are known?
• And how does government support for banks (through the safety net provided by deposit insurance and otherwise) affect how banks choose to compensate their employees?

The results of our analysis suggest weaknesses of current practices and policies, and also trade-offs that are inherent to the bargain made when government protection is provided for private financial institutions. Yet the analysis also reveals a potential solution to the dilemma: providing proper incentives to banks by charging debt default insurance premiums that depend on the compensation structure they choose.

In the following discussion of framework and analysis, three points are explored: the structure of a socially optimal compensation system; the structure of a compensation system that is privately optimal, given the reality of government-guaranteed bank debt; and policy interventions that can lead from the second structure to the first—that is, from privately to socially optimal compensation.

Regarding policy interventions, we note that if policymakers consider the potential solution just mentioned (charging banks insurance premiums that vary according to compensation structure) either unwise or impractical, then it may indeed be useful for government to regulate bank compensation more directly.

It should be emphasized that like all economic models, our simple framework is a very abstract version of reality. Employee motivation depends on many things in addition to financial compensation, for example, and in determining compensation systems, bank owners and managers consider factors not explicitly accounted for in our analysis. That said, we believe this analysis will prove helpful in redesigning compensation systems in the financial industry.

It should also be noted that these are preliminary findings and are the authors’ views alone, not necessarily those of the Federal Reserve.

A bank and its employees

Consider the issues facing a bank that wants to maximize its expected profits. It does so, in part, by paying its employees the minimum amount necessary to motivate them to take the job, search for potentially profitable investment opportunities, and invest in good risky projects (those that can be expected to provide economic returns higher than a safe asset, like a risk-free Treasury bond) while avoiding bad risky projects that can be expected to return less.

The bank’s objective of maximizing expected return is constrained by the fact that its employees must be induced to act in the interest of the bank. That is, employees have their own goals and their own means to achieve them. They can choose to take the bank job or pursue another potentially lucrative job offer. If they take the bank job, they can choose to actively search for investment opportunities (as the bank wishes) or simply set bank funds in a risk-free Treasury bond with modest return (which involves less work). And if they actually do search for investment opportunities, they can then decide whether to fund any given opportunity they find.

Our framework stipulates that some information and actions are private (known only to the employee) and others are public (known to the bank, the government, and others). Whether the employee invests in a risky project or a safe asset, for example, is public information, but whether the employee actually searches for investment projects at all is private—after all, the employee could falsely inform the bank that he or she searched but found nothing worthwhile so simply invested in the safe asset, per bank instructions.

Further, while we assume that the bank can observe the outcome of a risky project, the employee alone knows the likelihood of any given outcome for a project prior to investing in it. This is where incentive compatibility comes in. The bank must create a compensation system that will provide incentives for employees to act in the bank’s interests: working actively, investing in those projects that should be funded by the bank (good risks), and turning down those projects that shouldn’t (bad risks).

From the bank’s perspective, arriving at the best employment contract is a question of balancing the costs and benefits of paying an employee a higher or lower wage for this outcome or that outcome, and now versus later, taking into account how these characteristics of an employment contract affect the bank’s profits and the employee’s incentives.

Fixed salary?

As with most modern economics, our framework is expressed mathematically, and analysis of its properties provides clear answers as to the nature of optimal incentive compensation.

Would a fixed payment system provide good incentives? That is, would bank employees be properly motivated by an annual salary as in a standard white-collar contract? Our analysis suggests that the simple answer is no. When finding and evaluating investment opportunities requires effort (as we assume), a fixed salary does not provide any financial motivation for an employee to do more than park funds in the safe Treasury bonds. Remember that whether employees actively search for investment projects is private information; they could avoid work by falsely telling their boss that they searched but couldn’t find anything worthwhile. In other words, a fixed payment wouldn’t be incentive compatible.

The question then is: What incentive structure will convince an employee to actually search for potential investment projects and, furthermore, to invest only in the good ones, not the bad ones?

Skin in the game

As noted, to encourage workers to pursue good investment opportunities, the bank must reward employees for not simply parking the bank’s funds in the safe asset. Would a bonus system achieve this: providing a fixed salary, but with a bonus if the employee invests bank funds in what he or she says is a worthwhile investment? Again, the answer is no; this system isn’t incentive compatible either. The bonus for finding a project would indeed give the employee a financial incentive to search for and invest in risky projects rather than risk-free bonds, but it would provide no specific motivation to avoid bad investments, as the bank wishes.

To encourage employees properly, our analysis shows, compensation must depend not only on whether the employee finds a worthwhile investment project, but also on that project’s returns. Specifically, an employee needs to be especially rewarded for those outcomes that are more likely under good projects than bad projects. By having something personal at stake (their compensation), employee incentives will be aligned with the bank’s.

But the issue is more subtle than this. If good projects are simply those with a higher probability of success, then an optimal compensation scheme will just reward successful outcomes. However, it is entirely possible that a spectacularly high return is more likely under an especially risky “shoot for the moon” bad project (which the bank would not want the employee to invest in) than it is under a good project. In that case, an optimal compensation scheme may reward reasonably good returns and punish both bad and spectacularly good returns. Overall, an optimal compensation scheme may depend on rather complicated risk characteristics of both the type of projects banks want their employees to invest in and the type that banks want them to avoid.

Delaying payment

How does this play out over time? More explicitly, how should management time the payment of incentives in order to achieve optimal results? This question goes to the policy option of “clawbacks” and other types of deferred compensation intended to ensure that employees keep an eye on long-term outcomes, not just short-run profits.

Here the answer depends on specifically what motivates employees. Bank employees care about the timing of wage payments. A job with low pay for a decade followed by large but uncertain payments is simply less desirable to an employee than one with a smoother payment schedule; to make such a job sufficiently enticing, the employee will have to receive extra compensation, on average. An optimal employment contract takes this into account. In our mathematical treatment, we show that if the employee’s desire for smoother payments over time is sufficiently strong, requiring delayed payments isn’t the optimal solution.

Limited liability

The previous analysis (which we will call scenario one to contrast with later scenarios) assumed unlimited liability on the part of the bank. That is, it assumed that all the costs of investing in a project and of paying the employee are paid by the bank, and that all the benefits of a project’s return accrue to the bank. This implies that a compensation policy seen as optimal from the perspective of the bank will also be optimal for society in general.

But in current practice, of course, banks have limited liability—they don’t face all the costs. Indeed, a major motivation behind public debate over incentive compensation is the conflicts among three parties who bear different shares of the bank’s costs and benefits: stockholders (who implicitly determine compensation schemes by choosing bank directors and managers), debt holders (including bondholders and depositors), and the government, through its implicit and explicit support by way of bailouts, deposit insurance, and the like.

Analysis of three more scenarios with different relationships between bank stockholders and bondholders, and different roles for government, provides further insight on optimal compensation for bank employees.

Consider a second scenario in which government plays no role and bank stockholders pay interest to bondholders at a rate that varies according to default likelihood and degree. Our analysis shows that if government does not guarantee debt (implicitly or explicitly), the existence of bank debt and the limited liability it implies for stockholders would in no way inhibit stockholders from choosing the efficient employee compensation plan. That is, the bank would still choose the efficient compensation plan.

The central reason for this is that the interest rate stockholders must pay bondholders for debt not guaranteed by government will depend on the likelihood of default and on the degree of default if it occurs. If stockholders choose a compensation plan that increases the likelihood of default, or increases by how much the bank defaults, they pay the full cost of this choice by paying bondholders a higher interest rate. In essence, the company’s owners are paying the full costs of their risk-taking.

This suggests a third possible scenario, where the government guarantees bank debt, but charges an appropriate insurance premium, similar to the varying interest rate in scenario two. In this case, as well, the bank will choose the efficient compensation plan if the insurance premium is set so that the government’s expected profits from deposit insurance equal zero for any given compensation scheme.

In this scenario, if the government adjusts its insurance premium to take into account the bank’s compensation plan, it can mimic how a bank’s interest payment on its nonguaranteed debt would adjust to take into account the bank’s compensation plan. Once again, this would ensure that the company’s owners are paying the full costs of the risks they are motivating their employees to take.

A suboptimal scenario

But the situation is far different in a fourth scenario where debt is guaranteed by the government and insurance premiums are nonexistent, fixed, or unrelated to incentive compensation. In this case, the bank’s bondholders (or depositors) no longer care about employee compensation because they will be paid regardless of how the bank’s investments turn out and regardless of how much or when the bank pays its employees—the government will make them whole. In particular, the interest rate on debt that bank stockholders have to pay will no longer depend on how the bank chooses to compensate its employees.

This difference in the bank’s situation (from no government protection to debt guarantee by government) can actually change the bank’s instructions to employees from discouraging to promoting the funding of bad investments. This aspect of the moral hazard problem is well known. But even if we assume that banks still instruct employees to invest only in good projects, what effect would a government guarantee of bank debt have on how they choose to compensate those employees?

What we find is that a government guarantee of debt will encourage a bank to promise higher wages to employees when default occurs than it would without such guarantees. Why? A promise to pay an employee a higher wage next year if the bank defaults on its debts costs the bank’s shareholders and bondholders nothing; should default occur, the shareholders are wiped out regardless and the bondholders are protected by the government. But the bank’s promise of a higher wage encourages the employee to take the job in the first place. In effect, promises to pay the employee in the event of default are a way of shifting the bank’s wage bill onto the government.

Perhaps more troubling is that this effect at least partially shows up before a default actually occurs, but where default appears likely to occur in the future. Again, in such a situation, the benefit to the bank of promising the employee a high wage is that it helps convince the employee to take the job. If the bank ends up not defaulting, the cost of this high wage will be paid by the stockholders. But if it does default, the wage cost will be paid by the government since high wages today mean the bank will have less money to pay off its bondholders tomorrow. 

Scenario review

The incentive distortion caused by government guarantees is seen clearly through a brief review of these four scenarios (see Table 1).

The original scenario was one of no debt and unlimited liability for stockholders. That resulted in an optimal compensation scheme because the bank paid its full costs and accrued its full benefits.

The second scenario was one of limited stockholder liability, but where debt was not guaranteed by government and the interest rate stockholders must pay to bondholders for this nonguaranteed debt depended on the likelihood of default and on the degree of default if it occurs. Here again, the bank paid its full costs and accrued its full benefits, and therefore chooses an efficient compensation plan.

The third scenario was one with limited stockholder liability and government guaranteed debt, but where the government charged a varying insurance premium depending on the compensation plan chosen by the stockholders. Again, in this third scenario, the bank chooses the efficient compensation scheme.

It is only in the fourth scenario, a government guarantee of bank debt with default insurance premiums that are fixed, nonexistent, or unrelated to compensation, that the bank’s stockholders will choose an inefficient compensation scheme because government essentially subsidizes employee wages, and does so more when things go badly than when things go well.

At the root of the incentive compensation problem, then, is government guarantee of bank debt. In the two scenarios without government guarantees, the stockholders, bondholders, managers, and employees of the bank reach an optimal solution. With the government safety net, compensation structures are distorted toward excessive risk-taking.

However, if government charges banks an insurance premium that depends on the compensation structure banks choose, then optimal compensation packages and a government safety net are compatible.

Thus, proper incentive compensation structures can best be achieved not through direct government regulatory oversight of incentive compensation packages, but by government providing proper incentives to banks through risk-varying debt default insurance premiums.

Key points

Economic theory provides discipline for intuition, and mechanism design theory can thus provide useful intellectual structure for designing incentive compensation systems that lead to better outcomes for banks, and thereby society. As regulations are considered regarding these compensation structures, some lessons from our analysis may prove useful. Here then, in summary, are some of the key findings provided by our simple economic framework:

• To be motivated sufficiently, employees must be exposed financially to the outcome of the investment projects they choose to invest bank funds in. A fixed salary, or even a base salary plus bonus, would not prevent employees from investing in bad projects.
• To encourage workers to pursue good investment opportunities and avoid bad investments, contracts must especially reward employees for outcomes that are more likely under good projects than bad projects. Such efficient contracts may depend on rather complicated risk characteristics of both the type of projects banks want their employees to invest in and the type that banks want them to avoid.
• Delaying compensation to prevent focus on short-term outcomes alone may not be a good solution.
• Government guarantees of financial institution debt may perversely encourage dangerous levels of risk-taking and the offloading of employee compensation to the government.
• If government charges banks a varying insurance premium that reflects the risk levels of bank employee investment choices, then optimal compensation packages and a government safety net are compatible.
• If policymakers consider charging banks insurance premiums that vary according to compensation structure either unwise or impractical, then it may be useful for government to regulate bank compensation more directly.
• Government regulators should pay particular attention to situations where a bank not in default is likely to be so in the future. In such cases, the optimal compensation package from the bank’s standpoint will pay the employee more than is socially optimal.

Table 1:

Economic efficiency of compensation packages under different scenarios

Endnotes