Bank liquidity hoarding during the COVID-19 pandemic

Abstract

This paper examines the association between bank liquidity hoarding (BLH) and the COVID-19 pandemic. Using a sample of U.S. banks and applying fixed effect estimators, we reveal that banks rack up liquidity assets and liabilities when the pandemic escalates. Our finding holds with alternative BLH and COVID-19 proxies and is further validated by falsification tests. Additional analysis reveals that BLH improves bank stability by reducing earnings volatility, non-performing loans and the propensity to go bankrupt. This study supports the existing literature on BLH and economic adversities and expands our understanding of BLH during the COVID-19 pandemic.

1. Introduction

The onset of the COVID-19 pandemic has induced the closures and shrinkage of many non-essential businesses since firms are typically unprepared for a scenario when economic activities are forcefully restrained and customers shift their purchases to online platforms (Bartik et al., 2020; Donthu and Gustafsson, 2020). The pandemic-induced economic hardship incites a more significant need for external capital, and firms look to banks for financing support considering the liquidity creation function of the banking system to the economy (Berger et al., 2020; Li et al., 2020; Duan, Fan et al., 2021; Boubakri et al., 2023). Banks serve as a systemic stabilizer during this reverberated pandemic and are put under tremendous liquidity pressure due to a massively rising demand for capital in deposit withdrawal and credit forms. Another layer of lending risk also arises as firms, during COVID-19, seek loans primarily for survival rather than business expansion (Bartik et al., 2020). Such lending poses a foreseeable consequence of non-performing debts when the borrowers default because of the prolonged crisis (Taylor, 2022).

Relaxing banks' capital buffers could help; however, it does not sound like an ideal solution because recurring outbreaks likely deplete capital buffers. Demirgüç-Kunt et al. (2021) highlight that asset quality deterioration happens due to exhausting banks' capital buffers. The continually strict measures against the ongoing pandemic may lead to unknown aftermath for the banking system. The uncertainty literature emphasizes that banks' default risk increases with economic uncertainties (Bilgin et al., 2021); thus, banks must take action to revert this possibility by hoarding liquidity, especially among banks with weaker liquidity capacity (Duan, El Ghoul, et al., 2021; Berger et al., 2022). If liquidity creation by conventional banks exacerbates financial stability in high-income nations (Berger et al., 2019), BLH is a self-insured move to insulate banks from liquidity shocks and, through that, transmit a strong signal about their assured solvency to the market (Berrospide, 2021). Empirical evidence from the previous global financial crisis (GFC) in 2008 reveals that banks proactively hoarded liquidity and ceased interbank lending to cushion their exposure to subprime mortgage risks (Berrospide, 2021). Further back in time, U.S. banks racked up their voluntary excess reserves as a precautionary measure after the Great Depression.

Peculiar to the COVID-19 pandemic, banks have started accruing liquidity since 2019:Q2 and carried on more aggressively after the advent of the coronavirus in 2020:Q1, as shown in Fig. 1 . Empirical evidence suggests that banks hoard liquidity assets in the face of financial crises and uncertainties in anticipation of potential liquidity shocks that might otherwise force them to urgently sell liquidity assets at unexpectedly low prices (Li et al., 2020; Berger et al., 2022). Other motivations of BLH are to decrease their credit supply to minimize their exposure to heightened uncertainty and reduce federal fund deposits and interbank loans, consequently spreading liquidity risks to the banking system (de Haan and van den End, 2013; Berger et al., 2022). Furthermore, if the shock emanates when banks are unprepared, they might consider fire sales of their liquid assets as a last resort (de Haan and van den End, 2013).

Fig. 1.

Bank liquidity hoarding and total COVID-19 infected cases.

Given the importance of the banking system in orchestrating the economy, the failure of the banking system will incite a domino effect on all operating businesses. Lehman Brothers, a huge financial institution with roughly 25,000 employees at the time of default, is a salient example of a collapse whose effect lasts for years. Therefore, the idea of scrutinizing banks' defensive behaviors during the deadly COVID-19 pandemic is irresistible. Our study focuses on BLH behaviors during the unprecedented COVID-19 pandemic using the BLH measures of Berger et al. (2022). We document that, opposing the economic calm times, banks hoard more liquidity during the COVID-19 pandemic, and this hoarding occurs through both the asset and liability sides. Furthermore, we unveil the positive effect of BLH on bank stability during the pandemic. Specifically, although BLH does not improve bank profitability, it curtails banks' risks by lowering their earnings volatility, non-performing loans and the likelihood of bankruptcy.

Several theoretical contributions are highlighted in this study. While previous studies (e.g., Berrospide, 2021; Berger et al., 2022) have examined BLH in financial crises and uncertainties, this study is the first to explore BLH behavior during the COVID-19 pandemic, thus contributing to the emerging literature on the impact of COVID-19 on the economy and how banks respond to mitigate this impact (Li et al., 2021; Simoens and Vander Vennet, 2022). Moreover, although prior studies reached a consensus about the bank-destabilizing effect of COVID-19 and banks' inclination to hoard liquidity to encounter uncertainty, research on the connection between bank stability and BLH amid highly uncertain periods is currently limited. This study adds to the literature on the stability-enhancing effect of BLH behaviors by scrutinizing how bank stability, measured from different perspectives, varies in the cross-section of BLH and the COVID-19 pandemic.

The remaining of our paper is as follows. Section 2 presents the methodology and data. Section 3 reports and discusses empirical results. Section 4 concludes and provides some policy implications.

2. Methodology and data

2.1. Model design

To study BLH behaviors during the COVID-19 pandemic, we use the following model:

$$LiquidityHoardin{g}_{i,t}=\alpha +\beta COVI{D}_{i,t}+\gamma Control{s}_{i,t}+u_i+v_t+{\epsilon }_{i,t}$$ (1)

where Liquidity Hoarding proxies include LH (Total), LH (Assets), LH (Liabilities), and LH (OBS), following Berger et al. (2022). Precisely, LH (Assets), LH (Liabilities), and LH (OBS) capture liquidity hoarding from the asset, liability, and off-balance-sheet sides, respectively, while LH (Total) is the sum of three measures. Additionally, we use liquidity creation (LC) as an alternative measure of liquidity hoarding (Berger and Bouwman, 2009). The directional effect of LC is likely opposite to LH because the computation of some LH components is directly opposite to those of LC (see Berger et al., 2022). The magnitude of the COVID-19 contagion is measured by both dummy and continuous variables, namely Dum_CVD and Ln_CVD, respectively. Dum_CVD, taking the value of zero from 2018:Q3 to 2019:Q4 and one from 2020:Q1 to 2021:Q2, is designed to capture 6 quarters before and after the emergence of the COVID-19 pandemic. Ln_CVD, which reflects the severity of the pandemic, is the natural logarithm of the number of infected cases since 2020:Q1.1 Following the existing literature on COVID-19 and banking (e.g., Ashraf, 2020; Berger et al., 2021, 2022), we control bank size, size squares, capital, earnings, net interest income, bank competition level (HHI), bank-level population, operating cash flow, Tobin's Q, the standard deviation of daily stock returns, and GDP forecast dispersion. Bank and quarter fixed effects are included in models to account for characteristics peculiar to each bank and each point of time in our study. We cluster standard errors at the bank level. Variable definitions are presented in Table 1 .

Table 1.

Variable definitions.

Variables	Definitions	References
Bank liquidity hoarding
LH (Total)	Liquidity hoarding, scaled by gross total assets	Call reports, Berger et al. (2022)
LH (Assets)	Asset-side liquidity hoarding, scaled by gross total assets	Call reports, Berger et al. (2022)
LH (Liabilities)	Liability-side liquidity hoarding, scaled by gross total assets	Call reports, Berger et al. (2022)
LH (OBS)	Off-balance-sheet-side liquidity hoarding, scaled by gross total assets	Call reports, Berger et al. (2022)
LC	Liquidity creation, scaled by gross total assets	Call reports, Berger and Bouwman (2009)
COVID-19 variables
Dum_CVD	A dummy variable that equals one from 2020:Q1 to 2021:Q2, and zero from 2018:Q3 to 2019:Q4
Ln_CVD	The natural logarithm of total cases infected	JHU's Coronavirus Resource Center
PRE-COVID	A dummy variable that equals one if the quarter is between 2017:Q1–2018:Q4
Bank-level variables
SIZE	The natural logarithm of gross total assets	Call reports
SIZE_SQ	Size square	Call reports
CAPITAL	The ratio of capital over gross total assets	Call reports
EARNINGS	Income before taxes, provisions recognized in income over gross total assets	Call reports
HHI	Bank competition level (HHI), calculated as a deposit-weighted average of the Herfndahl–Hirschman index in all areas in which a bank has branch offices	Berger et al. (2022)
POPULATION	Bank-level population index calculated as the natural log of a weighted average of the population (in millions) in all areas in which a bank has a business, obtained from the Federal Reserve Bank of St. Louis.	Berger et al. (2022)
CASH FLOW	A state-level cross-sectional average of operating cash flows for each firm in quarter t divided by lagged total assets of each firm in the Compustat data whose headquarters is located in a corresponding state. Cash flow is calculated as the sum of earnings before extraordinary items and depreciation.	Compustat, Berger et al. (2022)
TOBIN'S Q	A state-level cross-sectional average of normalized Tobin's Q, defined as a firm-level Tobin's Q in quarter t normalized by a lagged total asset of each firm in the Compustat data whose headquarters is located in a corresponding state. Tobin's Q is defined as the market value of assets divided by the book value of assets.	Compustat, Berger et al. (2022)
SD (STOCK RETURN)	The standard deviation of daily value-weighted stock market returns from WRDS in quarter t.	CRSP, Berger et al. (2022)
GDP DISPERSION	Forecast dispersion of real GDP, defined as 75th percentile minus 25th percentile scaled by the absolute value of 75th percentile of expected real GDP growth in the next quarter from the Survey of Professional Forecasters of the Federal Reserve Bank of Philadelphia.	FED Philadelphia, Berger et al. (2022)
ROA	Net income over gross total assets	Call reports
sdROA	The standard deviation of ROA over the previous 4 quarters	Call reports
lnZSCORE	The natural logarithm of Berger et al. (2017) Z-score. Z-score is computed by summing a bank's mean ROA (Net income/Gross total assets) and its mean Capitalization ratio (Equity/Gross total assets), then divided by the standard deviation of ROA over the previous 4 quarters (t -3 to t)	Call reports
NPL	Non-performing loans to total loans	Call reports
BFE	Bank fixed-effects
TFE	Time fixed-effects

This table presents definitions of all main variables used in the analysis.

2.2. Sample

The Federal Reserve provides quarterly Call Reports (Reports of Condition and Income) for all commercial banks. Our baseline sample consists of 28,044 bank-quarter observations of the U.S. commercial banks from 2020:Q1 to 2021:Q2. Following Berger and Bouwman (2013) and Berger et al. (2017), we replace all observations with the ratio of total equity to total assets less than 1% by 1% to avoid distortion in ratios that contain equity. All financial ratios are winsorized at 1% level on the top and bottom percentiles to dampen the effects of outliers. We retrieve COVID-19 infections from Johns Hopkins University's Coronavirus Resource Center database. Descriptive statistics of variables used in this study and their pairwise correlation coefficients are shown in Table 2 .

Table 2.

Summary statistics and correlation coefficient matrix.

Panel A. Descriptive Statistics
Variable	N	Mean	SD	p25	p50	p75
LH (Total)	28,044	0.217	0.183	0.090	0.206	0.338
LH (Assets)	28,044	-0.031	0.146	-0.135	-0.045	0.058
LH (Liabilities)	28,044	0.309	0.077	0.269	0.321	0.362
LH (OBS)	28,044	-0.057	0.038	-0.078	-0.052	-0.029
LC	28,030	0.278	0.209	0.168	0.303	0.418
Ln_CVD	28,044	15.667	1.814	14.787	15.795	17.232
SIZE	28,044	12.620	1.370	11.692	12.474	13.368
SIZE SQ	28,044	161.151	36.032	136.703	155.607	178.697
CAPITAL	28,044	0.122	0.065	0.095	0.109	0.129
EARNINGS	28,044	0.014	0.012	0.008	0.013	0.018
HHI	28,044	0.167	0.110	0.098	0.141	0.202
POPULATION	28,044	1.970	0.848	1.386	1.820	2.541
CASH FLOW	28,044	-1.549	47.575	-0.204	-0.029	0.009
TOBIN'S Q	28,044	85.756	288.952	1.683	2.790	13.257
SD (VWRET)	28,044	0.015	0.010	0.010	0.010	0.020
GDP DISPERSION	28,044	0.362	0.072	0.289	0.356	0.439

Panel B. Pearson correlation coefficient matrix
Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
(1) LH (Total)	1.000
(2) LH (Assets)	0.865***	1.000
(3) LH (Liabilities)	0.445***	0.015***	1.000
(4) LH (OBS)	0.406***	0.408*	-0.303***	1.000
(5) LC	-0.519***	-0.751***	0.514***	-0.615***	1.000
(6) Ln_CVD	0.267***	0.264***	0.122***	0.003	-0.107***	1.000
(7) SIZE	-0.219***	-0.216***	0.035***	-0.278***	0.400***	0.044***	1.000
(8) SIZE SQ	-0.204***	-0.191***	0.021***	-0.257***	0.382***	0.042***	0.996***	1.000
(9) CAPITAL	0.027***	0.261***	-0.499***	0.187***	-0.669***	-0.047***	-0.209***	-0.187***	1.000
(10) EARNINGS	-0.162***	-0.009***	-0.264***	-0.069***	-0.197***	0.025***	0.109***	0.113***	0.498***	1.000
(11) HHI	0.065***	0.072***	0.025***	0.017***	-0.031***	0.005	-0.032***	-0.030***	-0.007***	0.031***	1.000
(12) POPULATION	0.071***	0.048***	0.043***	0.051***	-0.028***	0.005	0.192***	0.190***	-0.004	-0.058***	-0.073***	1.000
(13) CASH FLOW	0.003	0.003	0.001	0.002	-0.003	0.024***	-0.010***	-0.010***	-0.001	0.000	-0.003	-0.005	1.000
(14) TOBIN'S Q	-0.013***	-0.016***	-0.003	0.016***	0.021***	-0.019***	0.063***	0.064***	0.004	0.029***	-0.013***	0.218***	-0.106***	1.000
(15) SD (VWRET)	0.071***	0.067***	0.034***	0.006***	-0.026***	-0.039***	0.029***	0.028***	0.005	-0.004	-0.080***	0.016***	-0.059***	0.119***	1.000
(16) GDP DSPERSION	0.123***	0.122***	0.057***	-0.008***	-0.052***	0.694***	0.047***	0.045***	0.001	0.008***	-0.087***	0.018***	-0.021***	0.085***	0.333***	1.000

As shown in Panel A, Table 2, the mean value of LH (Total) is 0.217, implying that liquidity hoarded by banks accounts for roughly 21.7% of their gross total assets. The means of LH (Assets) and LH (OBS) are negative, whereas the mean of LH (Liabilities) is positive, suggesting the illiquidity of banks in our sample. The overall statistics are qualitatively consistent with those of Berger et al. (2022). The Pearson pairwise correlation coefficients of our variables are reported in Panel B, Table 2. Overall, these variables are strongly correlated; hence, including them in the model is necessary.

3. Results and discussion

3.1. Multivariate regressions

Table 3 reports the regression results of Model 1. First, we regress LH (Total) on our measures of COVID and obtain significantly positive coefficients of all COVID measures at 1% level, indicating that banks tend to rack up liquidity when the COVID-19 contagion is getting complicated. The effect remains positive when we replace LH (Total) with LH (Assets) and LH (Liabilities), whereas dissimilar impacts of Dum_CVD and Ln_CVD are observed when LH (OBS) is employed to measure BLH. This divergence may stem from the fact that Dum_CVD is to compare BLH before and after the advent of the COVID-19 pandemic, whereas Ln_CVD aims at observing the change in BLH during the health disaster. On the one hand, the adverse impact of Dum_CVD on LH (OBS) could be due to the government's massive support to small businesses through the banking system (e.g., CARES Act's Paycheck Protection Program) since the beginning of 2020:Q2, which ends up extending overall banks’ illiquid guarantees in the form of new credit lines on top of banks’ existing credit facilities (Beck and Keil, 2022). This, in turn, dampens the off-balance-sheet-side liquidity. On the other hand, regarding the positive coefficient of Ln_CVD, although banks exercise precautions during the pandemic by refraining from new credit issuance (with the exception of government support programs), the shortage of cash motivates current customers to exploit their already-granted credit lines, driving a sharp decline in banks’ unused commitments in the off-balance-sheet side (Greenwald et al., 2021; Li et al., 2022). Furthermore, this unprecedented health disaster endorses a greater use of liquid derivatives to hedge against unanticipated risks associated with the pandemic, thus magnifying BLH via the off-balance-sheet side (Berger et al., 2022). These two reasons explain the increase of LH (OBS) during the COVID-19 disaster and justify the positive coefficient of Ln_CVD. In general, the results indicate that BLH is performed via the asset and liability sides, while BLH via the off-balance-sheet side is sensitive to sampling periods.

Table 3.

Baseline multivariate analysis.

Variable	LH (Total)		LH (Assets)		LH (Liabilities)		LH (OBS)		LC
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Dum_CVD	0.139***		0.119***		0.025***		-0.004***		-0.104***
	(0.004)		(0.004)		(0.002)		(0.001)		(0.005)
Ln_CVD		0.027***		0.022***		0.004***		0.002***		-0.015***
		(0.001)		(0.001)		(0.000)		(0.000)		(0.001)
SIZE	0.144*	0.198*	0.316***	0.328***	-0.105***	-0.101***	-0.064***	-0.032*	-0.386***	-0.466***
	(0.082)	(0.115)	(0.073)	(0.103)	(0.024)	(0.021)	(0.014)	(0.017)	(0.096)	(0.126)
SIZE SQ	-0.007**	-0.009**	-0.013***	-0.013***	0.004***	0.003***	0.003***	0.001**	0.015***	0.017***
	(0.003)	(0.004)	(0.003)	(0.004)	(0.001)	(0.001)	(0.000)	(0.001)	(0.003)	(0.004)
CAPITAL	-0.081	-0.088	0.088	0.035	-0.186***	-0.165***	-0.015	0.010	-0.779***	-0.696***
	(0.102)	(0.137)	(0.090)	(0.120)	(0.034)	(0.030)	(0.022)	(0.028)	(0.121)	(0.167)
EARNING	-0.363***	-0.236***	-0.479***	-0.318***	0.111***	0.108***	-0.008	-0.005	0.499***	0.360***
	(0.086)	(0.082)	(0.074)	(0.073)	(0.037)	(0.029)	(0.015)	(0.014)	(0.089)	(0.085)
HHI	-0.013	0.001	-0.028	-0.018	0.011	0.018**	-0.004	0.001	0.041	0.032
	(0.033)	(0.030)	(0.029)	(0.028)	(0.008)	(0.008)	(0.006)	(0.005)	(0.030)	(0.027)
POPULATION	-0.080***	-0.068***	-0.083***	-0.067***	0.005	0.003	-0.001	-0.003	0.089***	0.076***
	(0.024)	(0.021)	(0.021)	(0.018)	(0.009)	(0.008)	(0.003)	(0.003)	(0.025)	(0.020)
CASH FLOW	-0.000	-0.000	-0.000	-0.000	0.000	0.000	0.000	-0.000	0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
TOBIN's Q	-0.000**	0.000*	-0.000***	0.000	0.000**	0.000*	0.000***	0.000**	0.000***	-0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
SD (STOCK RETURN)	1.535**	-0.115	1.984***	-0.050	1.126***	1.352***	-1.540***	-1.373***	-3.852***	-1.205*
	(0.658)	(0.716)	(0.571)	(0.614)	(0.278)	(0.261)	(0.214)	(0.201)	(0.700)	(0.725)
GDP DISPERSION	0.188***	0.104***	0.125***	0.054***	0.076***	0.067***	-0.012***	-0.017***	-0.030***	0.027***
	(0.005)	(0.007)	(0.004)	(0.006)	(0.002)	(0.002)	(0.001)	(0.001)	(0.006)	(0.007)
Constant	-0.506	-1.217	-1.881***	-2.270***	1.012***	0.972***	0.350***	0.118	2.730***	3.517***
	(0.580)	(0.835)	(0.522)	(0.744)	(0.168)	(0.146)	(0.098)	(0.125)	(0.706)	(0.919)
Obs	47,432	28,044	47,432	28,044	47,432	28,044	47,432	28,044	47,417	28,030
Adj R2	0.633	0.586	0.551	0.488	0.521	0.561	0.021	0.022	0.320	0.251
BFE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
QFE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

This table reports the estimates of Model 1 using different measures of BLH and COVID-19. For Columns 1, 3, 5, 7, 9: The sample spans from 2018:Q3 – 2021:Q2. For the remaining columns, the sample spans from 2020:Q1–2021:Q2. Robust standard errors are clustered at the bank level and are in parentheses. ***, **, * indicate significance at the 1%, 5%, and 10% levels, respectively.

Finally, we re-estimate Model 1 with LC as the dependent variable to validate our result. As expected, the liquidity creation role of banks decreases with the severity of the pandemic as suggested by the significantly negative coefficients of LC in both measures of COVID. Overall, our baseline model results show the increasing inclination of BLH when the COVID-19 situation heightens.

3.2. Falsification tests

We conduct falsification tests to validate our finding by assigning a placebo window in a pre-COVID period. Specifically, we employ two years of 2017:Q1–2018:Q4 as event windows for these falsification tests. Other model settings are similar to our main models. Based on Berger and Bouwman (2009), banks serve the role of liquidity creation during normal times. Therefore, if COVID indeed has a causal effect on Liquidity Hoarding, we expect the coefficients of PRE-COVID measures in the falsification tests to change their signs. The estimates, reported in Columns 1 to 5, Table 4 , show negative (positive) coefficients of variables LH (LC), implying that COVID does not affect banks' function of liquidity creation before the actual outbreak. This result corroborates a causal relationship between the COVID-19 pandemic and BLH.

Table 4.

Falsification test.

VARIABLES	LH (Total)	LH (Assets)	LH (Liabilities)	LH (OBS)	LC
	(1)	(2)	(3)	(4)	(5)
PRE-COVID (2018:Q1–2018:Q4)	-0.129***	-0.126***	-0.004***	0.000	0.123***
	(0.004)	(0.004)	(0.001)	(0.001)	(0.004)
SIZE	-0.322*	-0.291*	0.024	-0.074*	0.497**
	(0.168)	(0.153)	(0.053)	(0.040)	(0.233)
SIZE SQ	0.015**	0.014**	-0.001	0.003*	-0.022***
	(0.006)	(0.006)	(0.002)	(0.001)	(0.008)
CAPITAL	0.121	0.286	-0.113***	-0.036	-0.733***
	(0.228)	(0.199)	(0.022)	(0.038)	(0.209)
EARNING	-0.032	-0.070	0.004	0.045	-0.005
	(0.134)	(0.124)	(0.035)	(0.032)	(0.159)
HHI	-0.010	-0.000	-0.005	-0.001	-0.003
	(0.017)	(0.018)	(0.003)	(0.002)	(0.019)
POPULATION	0.036	0.015	0.009	0.009	-0.029
	(0.054)	(0.048)	(0.021)	(0.018)	(0.085)
CASH FLOW	0.000	-0.000	-0.000	0.000**	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
TOBIN's Q	-0.000***	-0.000***	0.000*	-0.000	0.000***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
SD (STOCK RETURN)	1.450***	0.905***	0.754***	-0.202***	0.141
	(0.218)	(0.191)	(0.062)	(0.060)	(0.211)
GDP DISPERSION	-0.193***	-0.145***	-0.065***	0.018***	0.067***
	(0.020)	(0.017)	(0.006)	(0.006)	(0.018)
Constant	1.902*	1.353	0.233	0.422	-2.308
	(1.132)	(1.022)	(0.361)	(0.264)	(1.542)
Obs	21,682	21,682	21,682	21,682	21,674
Adj R2	0.457	0.497	0.048	0.014	0.491
BFE	Yes	Yes	Yes	Yes	Yes
QFE	Yes	Yes	Yes	Yes	Yes

This table reports the estimates of the falsification test. We employ the period 2017:Q1–2018:Q4 as our event windows. Robust standard errors are clustered at the bank level and are in parentheses. ***, **, * indicate significance at the 1%, 5%, and 10% levels, respectively.

3.3. Liquidity hoarding and bank stability during COVID-19

COVID-19-induced uncertainty creates instability which clouds all entities in an economy. Banks’ stability is subject to such uncertainty because of their operating nature, which involves interactive transactions with various economic parties (Nguyen, 2021). This effect only pertains to conventional banks, which is the prevalent form of banking in the U.S. (Bilgin et al., 2021). Liquidity hoarding relieves banks from the fear of liquidity shocks that deteriorates their profitability and possibly pushes them to the brink of insolvency (de Haan and van den End, 2013; Duan, EI Ghoul et al., 2021). In this section, we investigate whether liquidity hoarding shelters banks from the COVID-19 adverse impact. Bank stability is captured by four proxies that examine different risk angles. The first proxy measures bank performance by the ratio of return on assets (ROA). The second proxy evaluates banks' earnings volatility by rolling standard deviation of ROA in the most recent twelve quarters (sdROA). The third proxy represents banks’ loan management through the non-performing loans to total loans ratio. Last but not least, the fourth proxy depicts banks' bankruptcy likelihood using Berger et al. (2017) Z-score (lnZSCORE). The results are reported in Table 5 .

Table 5.

Liquidity hoarding and bank stability.

VARIABLES	ROA	sd(ROA)	NPL	lnZSCORE
	(1)	(2)	(3)	(4)
Dum_CVD	-0.030***	0.039***	0.360***	-4.146***
	(0.004)	(0.006)	(0.048)	(0.677)
LH (Total)	-0.001*	0.000	0.011***	-0.090
	(0.000)	(0.001)	(0.004)	(0.070)
Dum_CVD * LH (Total)	-0.003***	-0.001***	-0.013***	0.199***
	(0.000)	(0.000)	(0.003)	(0.060)
SIZE	0.006***	-0.005***	-0.042***	1.052***
	(0.001)	(0.001)	(0.007)	(0.101)
SIZE SQ	-0.000***	0.000***	0.001***	-0.037***
	(0.000)	(0.000)	(0.000)	(0.004)
CAPITAL	0.014***	0.022***	-0.002	1.304***
	(0.003)	(0.002)	(0.011)	(0.221)
EARNING	0.406***	0.051***	-0.082*	-4.435***
	(0.013)	(0.008)	(0.042)	(0.610)
HHI	0.003***	-0.000	0.009	-0.042
	(0.000)	(0.001)	(0.007)	(0.089)
POPULATION	-0.000***	-0.000	-0.002***	0.022**
	(0.000)	(0.000)	(0.001)	(0.011)
CASH FLOW	0.000**	0.000	0.000*	-0.000
	(0.000)	(0.000)	(0.000)	(0.000)
TOBIN's Q	0.000	0.000	-0.000	-0.000
	(0.000)	(0.000)	(0.000)	(0.000)
SD (STOCK RETURN)	-1.095***	-0.012	-1.990***	30.455***
	(0.158)	(0.049)	(0.315)	(8.524)
GDP DISPERSION	0.012***	-0.001***	-0.012***	0.224***
	(0.001)	(0.000)	(0.002)	(0.057)
Obs	57,565	57,565	57,172	57,565
Adj R2	0.464	0.134	0.03	0.08
BFE	Yes	Yes	Yes	Yes
QFE	Yes	Yes	Yes	Yes

This table reports the estimates of the cross-sectional test of bank stability proxies on Dum_CVD, LH (Total) and their interaction. We employ the period 2018:Q3–2021:Q2 as our event windows. Robust standard errors are clustered at the bank level and are in parentheses. ***, **, * indicate significance at the 1%, 5%, and 10% levels, respectively.

As shown in Table 5, the adverse effect of COVID-19 is evident in all examined aspects of bank stability. To be specific, Dum_CVD negatively (positively) correlates with ROA and lnZSCORE (sdROA and NPL), indicating that the deadly pandemic deteriorates bank stability through weakened profitability, greater volatility of earnings, worsened non-performing loans and higher probabilities of bankruptcy. Turning to the moderating effect of LH (Total), Dum_CVD × LH (Total) is negatively correlated with ROA, which suggests a decline in earnings of liquidity hoarders amid the pandemic. This performance deficiency is a reflection of banks’ contractionary policies while continually accumulating liquidity to cope with the COVID-19 crisis. In the three remaining tests, Dum_CVD × LH (Total) is inversely correlated with sdROA and NPL while positively associates with lnZSCORE, revealing that liquidity hoarding during the pandemic relaxes banks' risk by reducing banks' earnings volatility, risky loans and the likelihood of bankruptcy, respectively.

Overall, although BLH is not beneficial for banks' profitability in the face of the health pandemic, it enhances bank stability by alleviating banks' earnings volatility, loan performance and bankruptcy threats.

4. Conclusions

In this paper, we study the BLH behavior during the COVID-19 pandemic and find that banks hoard liquidity to counter the increasing viral contagion. This result stands robust with various measures of BLH, COVID-19, and falsification tests. Moreover, we find evidence that BLH helps lower the volatility of banks' earnings, non-performing loans and the propensity of bankruptcy. Our finding is similar to previous studies concerning BLH and economic turbulence (e.g., de Haan and van den End, 2013; Berger et al., 2022).

Our results show the bright side of BLH as it assists the stabilization of the banking system when an abrupt health disaster takes place. It is worth noting that though liquidity hoarders solve their liquidity issue, such act aggravates other banks' solvency issue. Furthermore, BLH causes fewer liquidity assets circulating in the economy, thus causing malfunctioned short-term funding market. To resolve these potential problems while ensuring banks' financial soundness, strict regulatory requirements on banks' liquidity are necessary to mitigate excessive liquidity hoarding when the market turns unfavoured. In addition, policymakers should formulate a crisis management strategy for the capital market and place banks at the center of this strategy to accentuate banks' eminent role in response to economic adversities.

This study also offers a future research direction. Despite that BLH is a predictable motif at the onset of uncertainty, little is known about BLH in times of the COVID-19 pandemic. Future studies may find interest in dissecting and addressing specific BLH categories to understand banks' strategies to curtail their exposure to the ongoing health pandemic.

CRediT authorship contribution statement

Dung Viet Tran: Conceptualization, Data curation, Methodology, Software. Dien Giau Bui: Methodology, Visualization, Writing – review & editing. Cuong Nguyen: Formal analysis, Supervision, Writing – review & editing. Huy Viet Hoang: Formal analysis, Methodology, Validation, Writing – original draft.

Declaration of Competing Interest

None.

Data availability

• The authors do not have permission to share data.