The COVID-19 pandemic triggered a financial crisis in March 2020 that required extraordinary government intervention. Unlike during the financial crisis of 2007-08—when the government directly bailed out the banks—the banking sector did not receive firm-specific support or a TARP 2.0.
This contrast between these two government responses was obvious, leading to a new narrative by some banking observers: In the COVID-19 crisis, banks were the pillars on which the economic rescue stood, rather than beneficiaries of government handouts. One banking analyst summed it up: “During a significant shock to the system, the banks kind of were able to be leaned upon to support rather than the area that is being bailed out. It is just fundamentally a different place for them.”1 Or consider the view the former Comptroller of the Currency gave during congressional testimony in late 2020: “Just as banks have been a source of strength for citizens, businesses and communities during the pandemic, they will lead the national recovery as well.”2
Importantly, banks benefited indirectly from the cash that Congress pushed out to the American people.3 Government support allowed firms and families to keep paying their loans so that banks did not take losses.4
The fact that banks benefited from the fiscal response to COVID-19 is not bad. Indeed, it was the intended outcome, broadly speaking. The goal of the economic support was to bolster the economy—and banks taking large losses would have hurt, not helped, the recovery.
But ignoring the contributions these programs made to bank health is an important error of omission, in our view. The banks’ preferred narrative suggests that the inherent strength of banks, and that strength alone, explains their relatively good health coming out of a pandemic. In the worst case, these arguments could be used to advocate for weakening bank regulations that would put taxpayers at greater risk in future financial shocks.5
So how much did banks benefit from extraordinary government support? Determining a precise answer to that question is difficult because there are so many factors at play. There will be considerable uncertainty with any such calculation. But a few back-of-the-envelope calculations, done in very different ways, point to the same general answer: Banks received hundreds of billions of dollars in support from government programs. We estimate the amounts of government support in the range of roughly $100 billion to $300 billion.
In the rest of this note, we detail these calculations. First, we discuss what we call the “counterfactual” calculation. Then we discuss several alternative simple measures of the support provided to banks.
One logical way to calculate the support government programs provided banks is to project what bank loan losses would have been without the interventions. These projected losses can then be compared with the actual 2020 results. The difference between the projected and actual losses can then be interpreted as an estimate of the support banks received from the government relief programs.
We use a simple but reasonable method to estimate this counterfactual. A strong relationship exists between certain macroeconomic variables—such as the unemployment rate—and bank loan losses. Chart 1 shows the historical relationship between loan losses6 (measured as a percent of average loans, shown in blue) and the average unemployment rate (shown in orange) using quarterly data from 1985–2020. Loan losses for banks typically increase when the unemployment rate rises (early 1990s, early 2000s, and 2007-08 financial crisis) and then slowly decline as the unemployment rate falls.
Loading chart 1...
However, this relationship did not hold in 2020. There may be other factors at work, but at first approximation, government support to the economy would seem like an important cause for the breakdown in this relationship. The unemployment rate climbed more than 9 percentage points in the second quarter of 2020 (an increase larger than what occurred in the 2007-08 financial crisis), yet loan losses were largely unchanged in the quarter and then declined slightly over the rest of the year. Indeed, the intent of the government support at that point was to reduce the economic harm that such high unemployment would cause.
The relationship between loan losses and the unemployment rate can be more formally quantified using some basic statistical techniques. The result is a simple model that can be used to generate projected amounts of future loan losses for a specified change in the unemployment rate under historical conditions. This type of modeling is the same basic structure used in so called top-down stress test models.7 Additional details regarding our simple model can be found in the appendix. We acknowledge that our approach, as does all modeling of this type, abstracts from factors beyond the unemployment rate that could drive bank loan losses.
We can then use the model, along with the actual path of unemployment in 2020, to construct a counterfactual that shows what loan losses would have been had they simply followed the pre-COVID-19 historical relationship with unemployment. The left-hand side of Table 1 contains the actual and projected loan loss rates using this process. As noted above, actual loan losses ticked up fractionally in the second quarter of 2020 (moving from 0.53 percent to 0.56 percent) and then declined after that. The projected loan loss rate—based on the historical relationship between loan losses and unemployment—climbed markedly in the second quarter to over 3 percent and was still above 1 percent by the end of the year. Note that the projected amounts were three to six times as large as what actually occurred.
Actual and Projected Loan Loss Rates and Total Loan Losses
Source: Board of Governors and authors’ calculations
||Annualized loan loss rate (%)
||Quarterly losses ($b)
The projected loan loss rates can then be combined with the size of the total loan portfolio for the entire banking industry to generate an estimate of total loan losses (shown on the right-hand side of Table 1). Based on this simple model, projected loan losses—conditional on the realized values of unemployment that occurred in 2020—would have been nearly $170 billion over the last three quarters of the year. The actual loan losses recorded by the industry totaled only $38 billion. This implies that the government interventions may have provided the banking sector with the equivalent of roughly $130 billion of financial support in 2020. We find similar results if we restrict the sample to the nation’s largest banks.8
Moreover, this projection is dependent on the actual path of unemployment in 2020. While the rate surged in the second quarter (climbing to roughly 13 percent), it then declined for the remainder of the year. This improvement, though, was due in part to the government interventions. Unemployment may have stayed at or above 13 percent for the remainder of the year had the emergency actions not been taken.
Table 2 shows the results from the original projection based on the actual values of unemployment, along with two alternative scenarios. “UR flat” reports the projected amount of losses generated by the model if unemployment had remained at nearly 13 percent for the rest of 2020. In this case, projected loan losses would have totaled roughly $250 billion over the three quarters. “UR rising” shows the projected losses if the unemployment rate had continued to rise by another percentage point in both the third quarter and the fourth quarter to reach 15 percent. Losses under this alternative scenario would have been almost $270 billion, which suggests that the support for banks could have been as large as $230 billion had the unemployment rate reached 15 percent.
Actual and Projected Loan Losses Under Different Scenarios
Source: FRED, Federal Reserve Bank of St. Louis, and authors’ calculations
||Projected losses ($b)
||Projected losses ($b)
||Projected losses ($b)
These estimates might overstate the extent to which fiscal support prevented banks from suffering loan losses. The government also allowed certain borrowers to avoid making payments on their loans (that is, forbearance). To the degree this policy led to lower loan losses, our estimate is too high. As such, loan losses might be low now, but they might rise in the future once the forbearance policy is removed and banks can once again recognize the losses. However, some data suggest that the amount of losses related to forbearance may be limited.9
Alternative measures of the implied support
One way to gauge the reasonableness of the estimates we just made is to use different approaches and see if they generate similar results. In this section, we describe some alternative measures of the implied support received by the banking sector. Again, before describing them, we note that these calculations are simple and more back-of-the-envelope than precise estimates from more complicated measures. But they are illustrative. And we take some information from the fact that they provide measures relatively similar to our simplified statistical approach described above.
The first measure involves the charges taken by banks on a regular basis to cover potential loan losses. These charges are termed “provision expenses,” and they can be interpreted, at a very nontechnical level, as a forecast by bank management of what they think future loan losses will be. (More specifically, this expense is the amount needed to make the “allowance for loans and leases” adequate to absorb estimated future credit losses.) Chart 2 shows the quarterly provision expense for the banking industry (orange line) along with the corresponding amount of quarterly loan losses (blue), both measured as a percent of total loans for the period 2004 to 2021.10
Loading chart 2...
Banks themselves were directly anticipating significant losses from the COVID-19 crisis as the pandemic unfolded. The quarterly provision expense for the industry jumped from 0.14 percent to 0.48 percent during the first quarter of 2020 and then climbed another 8 basis points to 0.56 percent during the second quarter. Provision expenses of this magnitude indicated that bank management expected significant future loan losses for their institutions. A similar increase in provision expenses occurred in 2008 during the financial crisis. Quarterly loan losses subsequently rose during 2009 to 0.50 percent or more. A loss rate of 0.50 percent over the last three quarters of 2020—had it occurred—would have generated nearly $158 billion in losses based on the $10.5 trillion in loans that were outstanding at the beginning of the year.
Another approach is to use the expectations of equity analysts who monitor and assess the health of publicly traded banks. Chart 3 shows the relationship between annualized net charge-offs (blue line, the same series used in the counterfactual example) and the consensus estimate for net income (divided by assets) known as “return on assets,” or ROA (orange line). Coming into 2020, equity analysts were forecasting an industry-level ROA of 1.15 percent. Once the pandemic hit, though, they slashed their estimates by nearly two-thirds to a new level of just 0.48 percent. This dramatic decline of expected future income was consistent with what occurred at the start of the 2007-08 financial crisis. We note that this measure captures total net income, so it includes all revenue and expense items. Changes in the series are not due solely to changes in expectations regarding future loan losses. Still, it seems likely that a large portion of the decline in the consensus estimate was due to markedly higher projections of future loan losses, given the importance of the loan portfolio to banks. Despite the major change in analysts’ expectations, actual charge-offs hardly budged.
Loading chart 3...
Another measure from equity markets, called “SRISK,” is also consistent with the main message of our analysis. This measure seeks to calculate the expected capital shortfall of financial firms during a financial crisis.11 SRISK for publicly traded financial firms climbed to an all-time record in May 2020—higher even than the levels recorded during the 2007-08 financial crisis. One interpretation of this was that equity investors were valuing banks and other financial intermediaries at levels that implied the largest capital shortfalls seen in the past two decades. SRISK began to decline steadily following its peak in May, perhaps as investors realized that the financial sector benefited indirectly from the multiple fiscal and monetary actions. By March 2021, the measure was back to pre-COVID-19 levels.
Lastly, we note that the extraordinary government actions also included a host of emergency programs set up by the Federal Reserve to maintain the integrity and functioning of various financial markets. Absent this support, the market disruptions witnessed during the middle of March could easily have developed into even bigger and more sustained meltdowns, causing significant losses for banks that had the largest capital market operations, or even the failure of a key counterparty. Quantifying the losses from such outcomes is challenging, but the results from the official supervisory stress tests offer some insight. These tests include losses from a “global market shock” like what might have occurred had the capital and financial markets seized up. Banks with substantial trading, processing, and custodial operations were projected to have $95 billion in trading and counterparty losses from the shock included in the 2020 exercise. While it is impossible to know with certainty what might have happened to the financial system if the Fed’s actions hadn’t been taken, it is clear from the stress test results that losses from such disruptions could have been sizable.12
Comparing these results with those from a stress test
The methods and projected loan losses described in this note might sound similar to those from a stress test, particularly since we produced some projections recently using the Minneapolis Fed’s COVID-19 stress test tool.13
We briefly remind readers of the results from that test and then explain key differences between these two exercises. The COVID-19 stress test tool generated capital depletions for the 21 included banks that ranged from $460 billion to $640 billion, depending on the scenario chosen. The depletions reflected shocks that hit all bank operations. The stress shock, for example, reduced revenue from both lending and nonloan-related activities, along with capital reductions due to dividend payments. The total projected loan losses for the 21 banks ranged from $350 billion to $565 billion.
These amounts are higher than the numbers cited in this article for at least three reasons. First, the losses produced from the COVID-19 stress test tool were measured over nine quarters. We had only three quarters of actual data from 2020 to use for this note, and we estimated support only over that period, so our total is based on a much shorter period. Second, the loan loss models used in the stress test tool are driven by many macroeconomic factors, rather than relying on just the unemployment rate, as we do in our counterfactual. Finally, the scenarios used in the stress test tool allowed for changes in the economy that produce losses for banks much further into the future. For instance, unemployment spiked in the second projection quarter but then stayed at or above that level for another quarter before slowly declining. Other key variables—like the price indexes for houses and commercial real estate—didn’t hit their troughs until much later in the projection horizon.
For all of these reasons, and perhaps others, the COVID-19 stress test tool generated materially higher loan loss rates compared with the ones described in this note—with peak values ranging from 6 percent to 10 percent versus the 3 percent to 4 percent shown here. The bottom line is that the stress test exercise is different from the calculations in this note. At the same time, the differences suggest that our results could be a lower-bound estimate of how much support banks indirectly received from the government. If we had tried to look at a longer time period, as in a stress test, we might have observed additional shocks to banks that would have occurred absent government support, and our estimates of that support might have been much larger.
Discussing the results of our back-of-the-envelope exercise in the context of stress testing reinforces the ultimate point of our analysis. Few banks “failed” the exceptionally harsh COVID-19 stress test, and multiple rounds of the official Federal Reserve stress tests found that banks had sufficient capital and liquidity to withstand scenarios much worse than what actually happened during 2020.14 Banks look strong in this view. But those findings do not change the fact that banks benefited from the COVID-19 support programs, and the current strength of the banking sector is due in some part to the programs. We should not assume that the current condition of banks justifies policy or regulatory changes that could put taxpayers at increased risk of loss from banks in the future.
The results discussed above are constructed from the following data:
- Loan losses for commercial banks, measured using data from the Federal Reserve Board of Governors on quarterly net charge-offs (total loans, all banks, seasonally adjusted).
- Unemployment rate, as reported by the U.S. Bureau of Labor Statistics (retrieved from FRED, Federal Reserve Bank of St. Louis).
- Total loans for commercial banks, measured using data from the Board of Governors H.8 release (all commercial banks, seasonally adjusted).
The simple model is constructed by regressing the quarterly net charge-off rate on a constant, the lagged value of the charge-off rate, and the quarterly change in the unemployment rate. Estimating the model with data from Q2 1985 to Q1 2020 results in the following coefficients:
Estimated Coefficients for the Simple Loan Loss Model
The above coefficients are then used to generate projected loss rates for Q2 2020 to Q4 2020 using the following equation:
LRt = 0.07 + 0.93 * LRt-1 + 0.29 * UR change
Projected loss dollars are found by converting the annualized loss rates to quarterly values and then multiplying each one by the corresponding amount of total loans reported in the H.8 release. For example, Table 1 in the main text showed that loan losses in Q2 2020 were projected to be $88 billion:
Projected losses = (3.27 percent / 4) * $10,758 billion = $88 billion