How do firms learn? Evidence from corporate cash holdings during the COVID‐19 pandemic

Abstract

We examine the impact of COVID‐19 on US corporate cash holdings. Our findings suggest that greater pandemic exposure is associated with higher corporate cash holdings and that firms learn from prior experiences as they manage their cash policies. More specifically, the level of cash holdings in firms that experienced severe financial constraints during the 2008 credit crisis and firms with prior severe acute respiratory syndrome (SARS) and H1N1 exposure is significantly lower than that of firms with no prior epidemic or financial constraints experience. Overall, our findings support the learning behaviour of cash and contribute to corporate cash holdings literature by providing insights on the extent to which firms learn from prior experiences to manage their liquidity.

1. INTRODUCTION

The word “unprecedented” is rarely used properly. This time, it's being used properly – James Dimon – JP Morgan Chase on COVID‐19.

Corporate cash management has been the focus of academic studies and the popular press for several key reasons. First, the average cash balances in both the US and international firms have been increasing since the 1980s. According to Graham and Leary (2018), the average cash‐to‐assets ratio has increased from 8.67% in the 1970s to 13.23%, 16.73% and 22.64% in the 1980s, 1990s, and from 2001 through 2014, respectively. Second, as cash represents a valuable financing instrument, cash‐rich firms can deploy their reserves to take on profitable investment opportunities without costly external financing or to service their debts during an economic downturn (Acharya et al., 2007; Almeida et al., 2004; Ferreira & Vilela, 2004). Third, firms can also benefit from having healthy cash balances for risk management purposes, including using cash to reduce cash flow volatility (Acharya et al., 2007).

The topic of corporate cash management is brought to an even sharper focus amid the unprecedented and wide‐spreading impact of COVID‐19. According to the World Health Organization (WHO), there are over 535.86 million confirmed COVID‐19 cases worldwide as of 18 June 2022. The outbreak has been documented in 220 countries, areas and territories and COVID‐19 containment remains a challenge.1 With multiple shelter‐in‐place orders and closures of shopping malls, movie theatres, and gyms at the height of March 2020 when COVID‐19 was first declared a global pandemic, this virus has crippled the health of millions of individuals and hundreds of economies alike. Faced with a pandemic‐driven global supply chain disruption and prolonged sapping demands, US companies are scrambling to shore up corporate cash. S&P 500 companies, including Hilton, United Airlines, and Delta Airlines are reportedly selling loyalty points or miles to increase liquidity2 while Lennard Corp suspended land purchases to conserve its cash holdings.3 In another Wall Street Journal article, PNC Financial Services group reportedly sold its $13 billion stakes in Blackrock in May 2020 while Sanofi relinquished its entire one‐fifth stake in the drug maker Regeneron Pharmaceuticals Inc. to generate $11.7 billion in proceeds.4 According to data released by the Commerce Department, repatriation of foreign earnings surged to $124 billion in the first quarter of 2020 as US companies brought back foreign profits amid coronavirus.5

Such articles illustrate the urgency of building up cash reserve as COVID‐19 strikes global economies. As such, the need to understand how firms' exposure to COVID‐19 impacts cash reserve has never been greater. Furthermore, the necessity to comprehend whether and to what extent firms learn from prior experiences in managing their cash policies moving forward has never been more urgent as in today's state of the current pandemic. In this paper, we investigate how COVID‐19 influences cash holdings and the potential learning behaviour of firms. In particular, we examine the extent to which firms learn to manage their cash holdings from prior experiences, including the previous severe acute respiratory syndrome (SARS) and H1N1 epidemic and the 2008 financial crisis.6

Our findings suggest that greater pandemic exposure is associated with higher corporate cash holdings. In particular, we find that one‐standard‐deviation increase in COVID‐19 exposure is associated with a 2.44 percentage point or $32.2 million increase in cash holdings. Our finding of increased cash holdings during the pandemic is consistent with the ‘dash for cash’ phenomenon documented in Acharya and Steffen (2020) in which cash levels increased because of precautionary borrowings in the midst of COVID‐19. It should be noted that cash hoarding, while it may be beneficial for precautionary motives, could destroy shareholders' value. In particular, Darmouni and Siani (2022) document heightened bond issuance activities during the pandemic to the point that such activities crowd out bank loans. They also cautioned that these firms accumulate large and persistent amounts of liquid assets instead of using such funding for real investments. Ng et al. (2022) further show that a portion of cash holdings caused by managers' behavioural bias destroys shareholders' wealth.

While it is hardly questionable that the pandemic influences cash holdings, it is less clear whether prior experiences, including firms' exposure to the prior SARS/H1N1 epidemic and experience of severe financial constraints during the 2008 financial crisis, lead firms to increase or decrease their cash holdings systematically. Our results support the role of organisational learning in which firms learn from prior experiences as it manages cash policy moving forward. More specifically, firms with prior SARS/H1N1 exposure and those that experienced severe financial constraints during the 2008 credit crisis save less cash than their inexperienced peers do, perhaps due to the experienced firms' overestimation of their abilities to handle another health and financial‐related crisis. Overall, our findings support the learning behaviour of cash and contribute to corporate cash holdings literature by providing insights on the extent to which firms incorporate prior experiences in their liquidity management.

Our paper makes several contributions to the existing cash holdings and management literature on organisational learning. To the best of our knowledge, this paper is the first to empirically examine the learning behaviour of cash across multiple crises, including SARS/H1N1, the financial crisis of 2007–2009, and COVID‐19. As COVID‐19‐induced shock provides interesting setting to examine corporate liquidity management, our paper is related to several studies examining corporate payout and financing policies during the pandemic, including Acharya and Steffen (2020), Pettenuzzo et al. (2021), Hotchkiss et al. (2022), Darmouni and Siani (2022) and Ng et al. (2022). While Acharya and Steffen (2020) examine one particular subset of corporate financing – the level of daily credit line drawdown – during the early days of COVID‐19 (January 2020–June 2020), Hotchkiss et al. (2022) and Darmouni and Siani (2022) investigate the role of equity issuance and bond issuance, as firms respond to COVID‐induced cash flow shock. Meanwhile, Pettenuzzo et al. (2021) provide a helpful overview of corporate payout and financing decisions as strategies to preserve cash to outlast the pandemic and Ng et al. (2022) offer a behavioural explanation of negative managerial sentiments which lead to higher cash holdings under COVID‐19. Our paper, however, differs from the above studies in that we consider both the impact of firm‐level COVID‐19 risk in liquidity management and the learning behaviour of cash. More specifically, we bridge the gap between critical corporate policies such as cash holdings and organisational learning by testing the learning hypothesis in a corporate cash setting. While Chen et al. (2018, p. 680) allude to ‘the existence and benefit of learning effects’ in their empirical study of pre‐saved cash, and Pettenuzzo et al. (2021) document that ‘a prior corporate action7 made it more likely that a firm would follow up with another action’, suggesting a critical role of prior experience in explaining a firm's likelihood to follow up with other corporate strategies, we examine the role of organisational learning directly as we explicitly account for prior experiences and the extent to which such experiences impact cash policies. Thus, our paper makes an incremental contribution to better understand whether and how firms learn to manage cash.

The remainder of the paper is organised as follows. Section 2 reviews existing literature and presents hypothesis developments. Section 3 provides a description of the data, variables construction and methodology. Section 4 discusses our main results while Section 5 presents a battery of robustness checks. We conclude the paper in Section 6.

2. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

2.1. Corporate cash holdings

Since the 1980s, corporate cash management has been the focus of academic studies and the popular press for several key reasons. First, the average cash balances in US and international firms have been increasing since the 1980s. According to Graham and Leary (2018), the average cash‐to‐assets ratio has increased from 8.67% in the 1970s to 13.23%, 16.73% and 22.64% in the 1980s, 1990s, and from 2001 through 2014, respectively. Second, as cash represents a valuable financing instrument, cash‐rich firms can deploy their reserve to take on profitable investment opportunities without costly external financing or to service their debts during an economic downturn (Acharya et al., 2007; Almeida et al., 2004; Ferreira & Vilela, 2004). Third, firms can also benefit from having strong cash balances for risk management purposes, including using cash to reduce cash flow volatility (Acharya et al., 2007) or to signal resource independence to negotiate better prices in the M&A setting (Upadhyay & Zeng, 2017).

Following the increasing focus on cash policies, numerous studies, both theoretical and empirical, have examined motives and determinants of corporate cash holdings. On the one hand, the agency conflict between managers and its shareholders suggests that managers stockpile cash out of free cash flow for empire buildings and other self‐serving projects instead of paying dividends to shareholders (Dittmar et al., 2003; Dittmar & Mahrt‐Smith, 2007; Harford, 1999; Jensen, 1986; Masulis et al., 2009). On the other hand, cash holdings are attributed to the costly external financing motive, especially for financially constrained firms, and the precautionary motives in which firms hoard cash to hedge against economic downturn (Bates et al., 2009; Harford et al., 2014; Huang‐Meier et al., 2016; McLean, 2011; Opler et al., 1999).

As COVID‐19 exogenously induces cashflow shock for all firms and shutters hundreds of economies around the world, recent papers, including Acharya and Steffen (2020), Almeida (2021), Pettenuzzo et al. (2021), Hotchkiss et al. (2022), Darmouni and Siani (2022) and Ng et al. (2022), among others, investigate how firms' respond to COVID‐induced shock. Acharya and Steffen (2020) examine corporations' rush to undrawn credit lines, also known as the ‘dash for cash’ phenomenon in the midst of COVID‐19. Using daily credit line drawdown data, Acharya and Steffen (2020) highlight extreme precaution at the onset of the pandemic to the extent that ‘all firms drew down bank credit lines and raised cash levels’. Corporate ‘dash for cash’, however, is documented to be more prominent among low‐quality BBB‐rated firms. As BBB‐rated firms with low Z‐score face the most risk of becoming ‘fallen angels’ by being downgraded to non‐investment grade firms (BB‐rated or lower), such firms borrow significantly more from their credit lines to raise cash relative to comparable BB‐rated firms. In the later phase of the crisis when the US government offered more liquidity support via the establishment of various programs, including the Corporate Credit Facility, the Main Street Lending Facility, and the Paycheck Protection Program, Acharya and Steffen (2020) document the switch to capital markets to raise cash by the highest‐rated firms.

Using a case study of Ruth's Hospitality Group, Almeida (2021) investigates liquidity management during COVID‐19 and further corroborates evidence of precautionary borrowings. From Ruth's Hospitality Group annual report for the fiscal year end of 2020, the company's discussion of its credit line usage highlights the importance of precautionary borrowings as follows: ‘As a result of the COVID‐19 outbreak, we may be unable to secure additional liquidity. To improve our liquidity position, we have fully drawn the $115 million capacity of our revolving credit facility … The company borrowed the remaining available amount under our revolving credit facility as a precautionary measure in order to increase our cash position and preserve financial flexibility’.8 Corporations' ‘dash for cash’ is not only relevant to Ruth's Hospitality Group, but also widespread for others, especially those firms in heavily‐hit‐industries such as cruise lines, hotels, and airline companies. During the first 3 weeks of March 2020, more than 130 companies in Europe and the Americas borrowed at least $124 billion under revolving credit facilities.9 Among the first documented to resort to credit line borrowing are companies acutely hit by the outbreak such as Norwegian Cruise Lines and Hilton Worldwide as customers were forced to cancel travel plans. Subsequently, however, nearly every other industry followed in its ‘dash for cash’. To weather COVID‐19 related shutdowns, car maker General Motors announced its plan to draw down $16 billion from existing credit lines in its effort to ‘aggressively pursuing austerity measures’ to reserve cash10 while Ford drew down a total of $15.4 billion under its corporate credit facility to ‘offset the temporary working capital impacts of the coronavirus‐related production shutdowns and to preserve Ford's financial flexibility’.11

Hotchkiss et al. (2022) examine the role of external capital as companies respond to COVID‐19. Interestingly, young, small and riskier firms – those traditionally classified as financially constrained – appear to raise more capital (relative to assets) via equity issuance while its financially unconstrained peers resort to the debt markets to combat cashflow shortfalls. Examining lending activities between bank loans and debt markets during the first two quarters of 2020, Darmouni and Siani (2022) document heightened bank loan usage mostly among high‐yield firms and BBB‐rated firms. This finding is consistent with the ‘dash for cash’ in firms at the verge of being rated as high‐yield (‘fallen angels’) documented in Acharya and Steffen (2020). Darmouni and Siani (2022) also find that investment‐rated firms mostly prefer the debt markets to borrowing from their banks to the point that their preference for bond issuance crowd out bank loans. In particular, Darmouni and Siani (2022) document ‘over 40% of bond issuers leave their credit line untouched’ during the first quarter of 2020 and that ‘a large share of bond issuance is used to repay existing bank loans’. Darmouni and Siani's (2022) findings of the heightened use of bond issuance that crowd out bank loans during the first two quarters of 2020 further corroborate the role of external capital as financing sources for firms during COVID‐19 documented in Hotchkiss et al. (2022). Nevertheless, Darmouni and Siani (2022) caution readers of potential measurement errors as their data only captures quarterly debt outstandings. In the absence of daily firm‐level credit line drawdowns, the muted usage of bank loans as a financing source during the pandemic may be attributed to firms rushing to draw down their lines of credit during the first quarter of 2020 and repaying back all or most of their outstanding debts prior to the quarter end. It is important to note that Darmouni and Siani (2022) attribute the spike in bond issuance in 2020 to firms' preference for increased liquidity during the pandemic as opposed to real investment‐driven purposes during ordinary times. As such, the authors document greater increase in cash holdings, especially for firms in most affected sectors based on sector‐level employment changes since 2019, and further provide credence to the ‘precautionary borrowings’ hypothesis.

Pettenuzzo et al. (2021) investigate corporate payout and financing decisions during COVID‐19 in the aggregate. The authors find companies resorting to multiple strategies to preserve cash by suspending dividends and share repurchase programs and raising more capital via bond and equity issuance. Such corporate strategies as cutting dividends, pausing share repurchase programs, issuing bonds and stocks, while typically observed via a span of multiple months or years in an ordinary business cycle, now take place in a matter of days or weeks at the onset of COVID‐19. Interestingly, Pettenuzzo et al. (2021) find positive and statistically significant association between a firm's prior corporate action and its likelihood to engage in subsequent corporate strategies in all specifications. For instance, having cut dividends, stopped share repurchase programs or raised equity makes subsequent bond issuance decisions more likely. Similarly, prior corporate action is documented as an important predictor of firms' decisions to reduce dividends, suspend share repurchases, and raise new funding. While organisational learning is not a focus in Pettenuzzo et al. (2021), the authors suggest that ‘a prior corporate action made it more likely that a firm would follow up with another action’.

In international studies of corporate cash holdings, Honda and Uesugi (2021) provide evidence supporting the precautional motives of cash holdings in Japanese firms during COVID‐19. The authors document an increase in cash holdings since the start of the pandemic, especially in firms with greater sales volatility cash flow. Furthermore, Honda and Uesugi (2021) document greater sensitivity of cash holdings to cash flow during the first 3 months of the crisis and for financially constrained firms. Kwong and Zhang (2021) explore cash holdings during COVID‐19 outbreak for Asian‐listed‐firms and highlight the importance of liquidity buffers in time of crisis. Relative to larger firms headquartered in more developed parts of Asia, Kwong and Zhang (2021) find that smaller firms in emerging Asian locals experience greater barriers to secure external fundings.

While recent studies suggest ‘precautionary borrowings’ as the rationale for corporate cash hoarding, Ng et al. (2022) offer an alternative behavioural rationale observed during the pandemic. The authors find that working from home, catalysed by COVID‐19, depresses managerial sentiment and elevates managers' perceived risk, driving firms to build up cash holdings in times of heightened uncertainty. Interestingly, Ng et al. (2022) find that such excess cash build up often comes as a detriment to shareholders and destroys firms' values.

To the extent that firms hoard cash via increasing credit line drawdown as suggested by Acharya and Steffen (2020) and Almeida (2021), cutting payout policy and/or reducing R&D (Bliss et al., 2015; Pettenuzzo et al., 2021), issuing equities and bonds (Darmouni & Siani, 2022; Hotchkiss et al., 2022) either due to precautionary motives or depressed managerial sentiment (Ng et al., 2022), we predict a positive association between COVID‐19 exposure and cash holdings. As such, our first hypothesis is stated as follows:

There is a positive association between COVID‐19 exposure and cash holdings.

2.2. Organisational learning

As firms learn from past experience, Li (1995), Hébert et al. (2005) and Hoang and Rothaermel (2005) find experience and learning vary across different industries. Markides and Ittner (1994) and Lee and Caves (1998) document the role of country and culture in learning via international acquisitions. In particular, these studies find positive relations between US firms with prior international acquisition experience and increased short‐term abnormal stock return post‐acquisition. Such firms with prior acquisition experience also enjoy decreased volatility of post‐acquisition profits. Furthermore, the observed lower volatility of post‐mergers and acquisitions (M&As) profit is significantly more pronounced if such prior acquisition experience was in the same geographic area as the focal acquisition. Taken together, these studies suggest that learning from experience requires a certain level of specificity, and it can only be fostered if prior experience is closely related (Barkema et al., 1996, 1997; Barkema & Vermeulen, 1997; Cohen & Levinthal, 1990; Reuer et al., 2002; Shaver et al., 1997).

From the mid‐1990 and onward, several studies drawing from the transfer theory of cognitive psychology suggest that experience may not always be positive. As transferring old lessons to a new setting may not always result in a suitable solution, Haleblian and Finkelstein (1999) find a U‐shape relationship between experience as measured by the number of acquisitions taken and performance. Subsequent to the seminal paper by Haleblian and Finkelstein (1999), later studies explore the impact of a cumulative experience based on the focal acquisition (and not just the experience immediately prior to it) as well as the role of a heterogeneous versus homogenous experience in transferring learning into the focal acquisition (Finkelstein & Halebian, 2002; Hayward, 2002; Schijven & Barkema, 2007; Reuer et al., 2002). Thus, while earlier studies suggest the benefits of having prior experience, later studies document the need for a deeper examination of different types of experience and how such learning is being transferred since having prior experience can actually hurt firm performance.

More recent studies in the early 2000s provide mixed evidence on organisational learning. Fahlenbrach et al. (2012) empirically test two competing hypotheses, namely, the learning hypothesis and the risk culture hypothesis, as they examine the extent to which banks' performance during the 1998 crisis can predict performance in the subsequent 2008 financial crisis. The authors find evidence inconsistent with the learning hypothesis as they document that the worst performance banks during the 1998 crisis – those that stand to learn and gain the most – did not alter their business models and continued to perform poorly during the subsequent crisis. In the same vein, Yao et al. (2017) examine whether liability insurers learn from the 1980s crisis by investigating the link between insurers' performance during the 1980 crisis and the recent liability crisis in the early 2000s. Their findings also show inconsistencies to the learning hypothesis that ‘insurers that did poorly repeated their poor performance’. On the other hand, Luo (2005), Kau et al. (2008) and Aktas et al. (2011) provide evidence of CEO learning from market reaction around M&As. Specifically, Luo (2005) document a positive relationship between the likelihood of deal completion and market reaction to deal announcements, while Kau et al. (2008) find that managers pay attention to the market reaction in deciding whether a deal is consummated. Aktas et al. (2011) examine M&A bid premium and show that bid premium is a positive function of investor reactions to CEOs' previous deal announcements. Taken together, these studies document evidence of CEO learning via the lens of M&A activities in that CEOs acknowledge feedback from investors and dynamically adjust their acquisition biddings from deal to deal.

Even though empirical tests of the learning hypothesis are primarily concentrated within the M&A context, there are emerging studies documenting the role of past experiences in various areas of corporate policies. Malmendier and Nagel (2011) show that managers' risk‐taking is influenced by their prior experience of macroeconomic shock while Malmendier et al. (2011) document that CEOs' personal experiences such as those who grew up during the Great Depression and/or possess military experience have significant explanatory power for corporate financing decisions. In the cash management literature, Chen et al. (2018) examined the role of pre‐saved cash by financially constrained firms during the 2000 dot‐com and the 2008 financial crisis, they find that ‘firms that experienced the 2000 dot‐com crash and saved cash after that were less likely to default during the 2008 financial crisis’, consistent with the benefits of learning effects and organisational learning. More recently, Hassan et al. (2021a) examine firm‐level exposure to epidemic diseases such as COVID‐19, SARS, and H1N1 and find evidence that ‘firms with prior experience are somewhat more positive about the impact of the coronavirus on their business’ as they are able to learn from their experience and calibrate their expectation for COVID‐19 pandemic. While organisational learning is not a focus of the paper, Pettenuzzo et al. (2021) find that ‘a prior corporate action12 made it more likely that a firm would follow up with another action’, suggesting a critical role of prior experience in explaining a firm's likelihood to follow up with other corporate strategies. For instance, prior corporate actions such as the decision to suspend shares, buy back to preserve cash, or to issue equity and bonds to raise new capital ‘raised the likelihood that a firm subsequently suspended its dividend’. They also find that excluding the prior corporation action variable significantly reduces explanatory power over all corporate strategies to conserve cash.

While it is hardly questionable that the pandemic influences cash policy (Acharya & Steffen, 2020; Almeida, 2021; Darmouni & Siani, 2022; Hotchkiss et al., 2022; Ng et al., 2022; Pettenuzzo et al., 2021), it is less clear how firms learn to manage its liquidity amidst the mixed evidence in organisational learning literature on the impact of prior experiences. As such, in this study, we examine whether prior experiences such as the previous SARS/H1N1 epidemic exposure, and severe financial constraints during the 2008 credit crisis lead firms to systematically increase or decrease their cash hoard.

On the one hand, financially constrained firms that experienced hardship during the 2008 financial crisis may learn from the crisis and increase their precautionary savings to combat against the next crisis (Chen et al., 2018). Similarly, firms with prior adverse epidemic experience may be more motivated to increase their cash reserves for precautionary motives. To that extent, conditional on COVID‐19 exposure, we expect that firms with prior epidemic exposure or severe financial constraints experience during the 2008 financial crisis save more cash than their inexperienced firms do. On the other hand, prior experience may result in firms' overestimation of their ability to manage risk (Gervais & Odean, 2001), especially upon having survived the crisis. Facing COVID‐19 pandemic, such firms may appear to have less negative COVID‐19 related sentiment during their quarterly earnings calls (Hassan et al., 2021a). To this end, we hypothesise that firms experiencing severe financial constraints during the last 2008 financial crisis and those previously exposed to an epidemic may save less relative to their inexperienced peers. If firms do not exhibit any learning from the prior crisis, be it financial or health‐related, we would expect no statistical significance in the coefficients of the interaction terms between COVID‐19 exposure and firms' past experiences. Due to the potential opposing outcomes, our hypotheses are formalised in the null form as follows:

Conditional on COVID‐19 exposure, there is no significant association between firms that experience prior epidemic and cash holdings.

Conditional on COVID‐19 exposure, there is no significant association between firms that experience severe financial constraints during the 2008 financial crisis and cash holdings.

3. METHODOLOGY & DATA

3.1. Methodology

To empirically examine how firms manage their cash holdings amid COVID‐19 exposure, we adopt the cash model of Opler et al. (1999) augmented to include firm‐level epidemic variables constructed in Hassan et al. (2021a) as follows:

$$\begin{array}{cc}{\mathit{\text{CASH}}}_{i,t}& ={\beta }_0+{\beta }_1\mathit{\text{COVID}}\_{\mathit{\text{EXPOSURE}}}_{i,t}+{\beta }_2{\mathit{\text{SIZE}}}_{i,t}+{\beta }_3{\mathit{MB}}_{i,t}+{\beta }_4{\mathit{CF}}_{i,t}\hfill \\ & +{\beta }_5{\mathit{NWC}}_{i,t}+{\beta }_6{\mathit{\text{CAPEX}}}_{i,t}+{\beta }_7{\mathit{\text{LEVERAGE}}}_{i,t}\hfill \\ & +{\beta }_8\mathit{\text{INDUSTRY}}\_\mathit{\text{SIGMA}}+{\beta }_9\mathit{CF}\_{\mathit{\text{RISK}}}_{i,t}\hfill \\ & +{\beta }_{10}R\&D_{i,t}+{\beta }_{11}\mathit{\text{ACQUISITION}}\_{\mathit{EXP}}_{i,t}+{\epsilon }_{i,t},\hfill \end{array}$$ (1)

where CASH is measured as the ratio of cash and short‐term investments to net assets (book value of total assets minus cash and short‐term investments). We use data provided by Hassan et al. (2021a) to measure firm‐level exposure related to COVID‐19. COVID_EXPOSURE is measured by counting the frequency of keywords/synonyms related to the spread of COVID‐19 in conference call transcripts. To account for the differences in transcript lengths, the authors divide COVID‐19 related total words by the total number of words in each transcript. While this novel measure captures the time‐varying firm‐level COVID‐19 exposure via frequency counts of key words and synonyms related to the spread of COVID‐19 at quarterly earnings calls, what management discusses is arguably as important as how often certain key words appear in a transcript. Using a word pattern‐based algorithm to limit human judgement to a minimum, Hassan et al. (2021a) identify the underlying concerns of management and market participants by examining a total sample of 174,582 discussion snippets13 in 2020. At the onset of the outbreak in January 2020, general concerns unspecified to any specific topics or generic worries such as ‘There is no doubt that COVID‐19 is impacting our business’ account for approximately 60% of the discussion snippets analysed. Such general concerns voiced during the first quarter of 2020 recede in the subsequent quarters of 2020 and make up about 40% of all classified sentences. Out of the five common topics identified during quarterly earnings calls, demand impact and concerns regarding the sudden change in demand remain the most voiced topic (41.53%). Following in the second is supply shocks‐related concerns (32.39%), which include discussions related to supply chain, production and operations. Cost topics and financing concerns represent approximately 11.22% and 13.30%, respectively of the five commonly discussed topics during 2020. Lastly, a relatively small percentage of the discussion snippets (1.57%) is related to government support measures such as the Coronavirus Aid, Relief, and Economic Security (CARES) Act or the Paycheck Protection Program.

Prior studies document several determinants of corporate cash holdings, including firm size, growth opportunities, cash flows, net working capital, capital expenditures, leverage, industry‐level and firm‐level cash flow volatility, R&D investment, and acquisition expenses (Bates et al., 2009; Han & Qiu, 2007; Kim et al., 1998; Opler et al., 1999). As such, we include these factors as control variables in all of our regression models and provide descriptions of all variable constructions in Appendix 1: Table A1.

Our empirical models to test Hypotheses 2A and 2B are provided as follows:

$$\begin{array}{cc}{\mathit{\text{CASH}}}_{i,t}& ={\beta }_0+{\beta }_1\mathit{\text{COVID}}\_{\mathit{\text{EXPOSURE}}}_{i,t}+{\beta }_2\mathit{\text{PRIOR}}\_{\mathit{\text{EPID}}}_{i,t-n}\hfill \\ & +{\beta }_3\mathit{\text{COVID}}\_{\mathit{\text{EXPOSURE}}}_{i,t}*\mathit{\text{PRIOR}}\_{\mathit{\text{EPID}}}_{i,t-n}+{\beta }_4{\mathit{\text{SIZE}}}_{i,t}+{\beta }_5{\mathit{MB}}_{i,t}\hfill \\ & +{\beta }_6{\mathit{CF}}_{i,t}+{\beta }_7{\mathit{NWC}}_{i,t}+{\beta }_8{\mathit{\text{CAPEX}}}_{i,t}+{\beta }_9{\mathit{\text{LEVERAGE}}}_{i,t}\hfill \\ & +{\beta }_{10}\mathit{\text{INDUSTRY}}\_{\mathit{\text{SIGMA}}}_t+{\beta }_{11}\mathit{CF}\_{\mathit{\text{RISK}}}_{i,t}+{\beta }_{12}R\&D_{i,t}\hfill \\ & +{\beta }_{13}\mathit{\text{ACQUISITION}}\_{\mathit{EXP}}_{i,t}+{\epsilon }_{i,t},\hfill \end{array}$$ (2)

$$\begin{array}{cc}{\mathit{\text{CASH}}}_{i,t}& ={\beta }_0+{\beta }_1\mathit{\text{COVID}}\_{\mathit{\text{EXPOSURE}}}_{i,t}+{\beta }_2\mathit{\text{SIZE}}\_{\mathit{AGE}}_{i,t-n}+{\beta }_3\mathit{\text{COVID}}\_{\mathit{\text{EXPOSURE}}}_{i,t}\hfill \\ & *\mathit{\text{SIZE}}\_{\mathit{AGE}}_{i,t-n}+{\beta }_4{\mathit{\text{SIZE}}}_{i,t}+{\beta }_5{\mathit{MB}}_{i,t}+{\beta }_6{\mathit{CF}}_{i,t}+{\beta }_7{\mathit{NWC}}_{i,t}\hfill \\ & +{\beta }_8{\mathit{\text{CAPEX}}}_{i,t}+{\beta }_9{\mathit{\text{LEVERAGE}}}_{i,t}+{\beta }_{10}\mathit{\text{INDUSTRY}}\_{\mathit{\text{SIGMA}}}_t\hfill \\ & +{\beta }_{11}\mathit{CF}\_{\mathit{\text{RISK}}}_{i,t}+{\beta }_{12}R\&D_{i,t}+{\beta }_{13}\mathit{\text{ACQUISITION}}\_{\mathit{EXP}}_{i,t}+{\epsilon }_{i,t},\hfill \end{array}$$ (3)

where PRIOR_EPID measures the total number of times SARS and H1N1 were mentioned in a firm's earnings calls held at the peak of SARS and H1N1 outbreaks, scaled by the number of words in a transcript. The discussions of SARS and H1N1 in earnings conference calls peak in the first quarter of 2003 and the third quarter of 2009, respectively. To measure a firm's financial constraint during the 2008 financial crisis, we follow Hadlock and Pierce (2010) and construct $\mathit{\text{SIZE}}\_\mathit{AGE}$, calculated as $-0.737*\mathit{Log}\left(\mathit{\text{Total Assets}}\right)+0.043*\mathit{Log}{\left(\mathit{\text{Total Assets}}\right)}^2-0.040*\mathit{Age}$. Total assets are Winsorised at $4.5 billion and age is Winsorised at 37 years. Higher values of the $\mathit{\text{SIZE}}\_\mathit{AGE}$ index imply greater levels of financial constraint.

On the one hand, financially constrained firms that experienced hardship during the 2008 financial crisis may learn from the crisis and increase their precautionary savings to combat against the next crisis such as the COVID‐19 pandemic (Chen et al., 2018). Similarly, firms with prior adverse epidemic experience may be more motivated to increase their cash reserves. In such cases, the coefficients of the interaction terms between COVID‐19 and firms' past experiences (i.e., ${\beta }_3$) are expected to be positive. However, prior experiences may result in firms' overestimation of their preparedness to manage risk in the face of a pandemic (Gervais & Odean, 2001), driving such firms to have less negative COVID‐19 related sentiment during their quarterly earnings calls (Hassan et al., 2021a). Thus, firms that experienced financial constraints during the previous 2008 financial crisis or those that were previously exposed to an epidemic may not save as much as their inexperienced peers do. Thus, the coefficients ${\beta }_3$ are expected to be negative. If firms do not exhibit any learning from the prior crisis, be it financial or health‐related, we would expect no statistical significance in the coefficients of the interaction terms between COVID‐19 exposure and firms' past experiences. Due to the potential opposing outcomes, we have no ex ante prediction for the coefficients of the interaction terms.

3.2. Data and sample

Our study focuses on corporate cash holdings during COVID‐19 and the learning behaviour of US firms in corporate liquidity management for several reasons. First, focusing on US firms allows us to reduce any heterogeneity impacts of external variables such as country‐level laws and regulations as all US firms are governed similarly at the Federal level. Second, using US data enables us to examine a larger sample of over 3000 firms, thus allowing us to draw more accurate inferences on organisational learning as it applies to liquidity management.

Our primary data source for firm‐level risk is as described in Hassan et al. (2019, 2021a, 2021b).14 We obtain financial statement data from Compustat quarterly files for the firms announcing their earnings during 2020, 2021 and the first quarter of 2022. Since financial firms (North American Industry Classification System (NAICS) two‐digit code 52) likely carry cash for capital requirements rather than for economic reasons, and cash holdings in utility firms (NAICS two‐digit code 22) may be subject to regulatory requirements, we exclude firms in these industries. We also require that firms have positive sales and total assets. Our final sample includes 24,667 firm‐quarters with 3025 unique firms. To address the potential problem of outliers, we Winsorise all continuous variables at the 1st and 99th percentiles. To examine the effects of a prior epidemic, we employ the SARS outbreak in 2003 and the H1N1 outbreak in 2009 and test their effect on cash holdings.

We present the descriptive statistics of our key variables in Table 1. Overall, CASH has a mean (median) of 0.56 (0.15), indicating that an average (median) firm holds 56% (15%) of net book assets in cash. The large difference in mean and median for CASH is due to the influence of extreme outliers in which a few firms hold most of their net assets in cash, as discussed in Bates et al. (2009).

TABLE 1.

Descriptive statistics

Variable	N	Mean	Std. dev.	P25	Median	P75
CASH	24,667	0.56	1.22	0.06	0.15	0.41
LN_CASH	24,667	0.31	0.43	0.06	0.14	0.34
COVID_EXPOSURE	24,667	1.08	1.16	0.23	0.72	1.56
PRIOR_EPID a	24,667	0.05	0.18	0.00	0.00	0.00
SIZE_AGE a	24,667	−3.43	0.41	−3.68	−3.66	−3.31
SIZE	24,667	7.37	2.00	6.05	7.38	8.72
MB	24,667	2.48	2.33	1.14	1.65	2.86
CF	24,667	−0.05	0.25	−0.01	0.02	0.04
NWC	24,667	−0.06	0.36	−0.11	0.00	0.11
CAPEX	24,667	0.03	0.04	0.01	0.02	0.03
LEVERAGE	24,667	0.32	0.24	0.12	0.30	0.46
INDUSTRY_SIGMA	24,667	2.32	1.49	1.51	2.30	2.53
R&D	24,667	0.41	2.15	0.00	0.00	0.11
ACQUISITION_EXP	24,667	0.02	0.05	0.00	0.00	0.00
CF_VOL	24,667	0.07	0.18	0.01	0.02	0.04

Note: This table presents the summary statistics of variables used in our analyses. All variables are defined in Appendix 1: Table A1.

^a We present descriptive statistics for PRIOR_EPID and SIZE_AGE using continuous variable but convert these variables to dummy variables for our main tests.

To reduce the potential problem of large outliers, we follow Foley et al. (2007) and use Ln Cash Holdings, which equals the natural logarithm of (1 + CASH), as an alternative measure of cash. Aside from the two main proxies of cash holdings, we also re‐estimate all models in the robustness check section using two other alternative measures of cash, including the ratio of cash and short‐term investments to the book value of total assets as well as the change in cash holdings. In our untabulated descriptive statistics, on a dollar basis, a median US firm with an asset base of $1.596 billion holds approximately $189 million in cash.

The 25th and 75th percentile for COVID_EXPOSURE is 0.23 and 1.56, respectively, indicating that most firms in our sample are adversely affected by COVID‐19 as the virus continues to disrupt hundreds of economies and the health of millions of people around the world. On the contrary, the 25th, median, and 75th percentile of PRIOR_EPID is 0, suggesting that most of the sample firms did not face any adverse consequence during the 2003 SARS and the 2009 H1N1 outbreaks. As such, prior exposure to epidemic diseases might help or hurt firms' motivation to save cash in dealing with COVID‐19 depending on what they learn from past experiences. On the one hand, prior exposure may allow firms to learn from the experience and motivate them to increase their cash reserve in lines with the precautionary motives of cash. On the other hand, firms may overestimate their preparedness upon having survived the crisis and end up holding less ‘precautionary’ cash moving forward. Finally, to maintain consistency, we also use dummy variable for SIZE_AGE to capture the presence of financial constraints during the 2008 financial crisis.15

Table 2 presents the Pearson correlations between key variables. A simple correlation analysis shows that CASH is positively associated with COVID_EXPOSURE, albeit the correlation of 0.03 is relatively small in magnitude. The correlations between CASH and other determinants of cash holdings such as firm size, market‐to‐book, cash flow, net working capital, capital expenditure, etc. show results consistent with prior studies in corporate cash holdings. Interestingly, our prior experience variable $\mathit{\text{SIZE}}\_\mathit{AGE}$ is positively correlated to CASH while the other experience variable, $\mathit{\text{PRIOR}}\_\mathit{\text{EPID}}$ is negatively correlated to CASH.

TABLE 2.

Correlation matrix

	Pearson correlation coefficients, N = 24,667
		1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1	CASH
2	LN_CASH	0.94
3	COVID_EXPOSURE	0.03	0.04
4	PRIOR_EPID a	−0.03	−0.04	0.04
5	SIZE_AGE a	0.01	0.04	0.05	−0.04
6,	SIZE	−0.31	−0.38	−0.10	0.17	−0.36
7	MB	0.31	0.41	−0.04	0.01	0.02	−0.12
8	CF	−0.54	−0.51	−0.04	−0.03	0.00	0.22	−0.11
9	NWC	−0.36	−0.35	0.04	0.00	0.08	0.01	−0.12	0.31
10	CAPEX	0.05	0.05	−0.11	−0.03	−0.03	0.01	0.08	0.04	−0.04
11	LEVERAGE	−0.21	−0.26	0.03	0.00	−0.11	0.19	−0.11	0.10	−0.02	−0.02
12	INDUSTRY_SIGMA	0.11	0.15	0.01	−0.04	−0.01	−0.06	0.09	−0.05	−0.07	0.09	−0.13
13	R&D	0.53	0.48	0.01	−0.02	0.00	−0.20	0.10	−0.42	−0.24	0.02	−0.10	0.06
14	ACQUISITION_EXP	−0.06	−0.06	−0.06	−0.01	−0.01	0.02	0.05	0.07	0.04	−0.05	0.00	0.02	−0.04
15	CF_VOL	0.62	0.58	0.04	−0.04	0.00	−0.30	0.18	−0.42	−0.27	0.03	−0.09	0.09	0.48	−0.04

Note: This table presents the Pearson correlation of variables used in our analyses. Bold values indicate that the correlation coefficient is significantly different from zero at the p < 0.05 level (two‐tailed). All variables are defined in Appendix 1: Table A1.

^a We present correlation for PRIOR_EPID and SIZE_AGE using continuous variable but convert these variables to dummy variables for our main tests.

4. EMPIRICAL FINDINGS

4.1. Pandemic risks and corporate cash holdings

Table 3 presents the impact of COVID‐19 on corporate cash holdings. Columns (1) and (2) report the results of the estimating Equation (1) for COVID_EXPOSURE with CASH as a dependent variable, whereas columns (3) and (4) report the results with LN_CASH as a dependent variable. COVID_EXPOSURE is the frequency of keywords related to the spread of COVID‐19 in conference call transcripts as measured in Hassan et al. (2021a). To account for the difference in transcript lengths, Hassan et al. (2021a) divide COVID‐19 related total words by the total number of words in the transcript. Across all four specifications, the coefficients on COVID_EXPOSURE are positive and statistically significant. In column (1), the coefficient on COVID_EXPOSURE is positive (0.021) and statistically significant at the 5% level. Using the point estimate in column (1) to gauge the economic effect of COVID‐19, we find that a one‐standard‐deviation increase in COVID_EXPOSURE is associated with a 2.44 percentage point increase in cash holdings.16 On a dollar basis, a one‐standard‐deviation increase in COVID_EXPOSURE is associated with a $32.2 million increase in cash holdings.17 As such, our findings support hypothesis H1 of a positive association between COVID‐19 exposure and corporate cash holdings and our evidence further corroborates the COVID‐19 ‘dash for cash’ phenomenon documented in recent studies (Acharya & Steffen, 2020; Almeida, 2021; Darmouni & Siani, 2022; Hotchkiss et al., 2022; Pettenuzzo et al., 2021). Thus, the impact of the pandemic on corporate cash policy not only has statistical significance but also holds meaningful economic importance.

TABLE 3.

COVID‐19 exposure and cash holdings

Variable	Cash holdings		Ln cash holdings
	1	2	3	4
Intercept	0.544***	0.377**	0.384***	0.382***
	(9.60)	(2.22)	(17.61)	(7.00)
COVID_EXPOSURE	0.021**	0.017*	0.010***	0.009***
	(2.47)	(1.95)	(3.43)	(3.17)
SIZE	−0.050***	−0.052***	−0.033***	−0.034***
	(−8.43)	(−8.72)	(−14.74)	(−15.06)
MB	0.091***	0.082***	0.049***	0.044***
	(12.01)	(10.51)	(17.90)	(15.83)
CF	−1.113***	−1.084***	−0.366***	−0.345***
	(−9.94)	(−9.92)	(−11.29)	(−10.99)
NWC	−0.474***	−0.553***	−0.164***	−0.185***
	(−7.38)	(−7.96)	(−8.16)	(−8.92)
CAPEX	0.813**	1.319***	0.210*	0.511***
	(2.42)	(3.73)	(1.86)	(4.49)
LEVERAGE	−0.571***	−0.596***	−0.245***	−0.251***
	(−9.45)	(−9.41)	(−11.07)	(−11.11)
INDUSTRY_SIGMA	0.013**	−0.044***	0.012***	−0.016***
	(2.03)	(−5.56)	(5.20)	(−6.44)
R&D	0.110***	0.099***	0.030***	0.026***
	(9.11)	(8.31)	(10.00)	(9.02)
ACQUISITION_EXP	−0.543***	−0.571***	−0.256***	−0.292***
	(−5.38)	(−5.94)	(−5.91)	(−7.06)
CF_VOL	2.178***	2.066***	0.659***	0.621***
	(11.13)	(10.80)	(13.20)	(12.84)
Observations	24,667	24,667	24,667	24,667
R‐squared	0.58	0.59	0.60	0.62
Industry FE	No	Yes	No	Yes

Note: This table presents the results of estimating Equation (1) using OLS regression with Cash Holdings as dependent variable in columns (1) and (2) and Ln Cash Holdings as dependent variable in columns (3), and (4). CASH equals the ratio of cash and short‐term investments to net assets. LN_CASH equals the natural logarithm of One plus Cash Holdings. The COVID_EXPOSURE is the frequency of keywords related to spread of COVID‐19 in conference call transcripts. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

In addition, the coefficients on other control variables have signs and significance consistent with those documented in prior literature. For example, R&D investments and firm growth opportunities, proxied by market‐to‐book ratio, are positively related to the level of cash holdings, consistent with the precautionary motive for holding cash (Opler et al., 1999). The results also indicate that larger firms hold less cash due to economies of scale of holding cash (Bates et al., 2009). Similarly, acquisition expenditures, NWC and cash flow have negative and significant coefficients. A positive and significant coefficient on cash flow volatility is consistent with the explanation that uncertainty leads to situations in which firms have more outlays than expected. Therefore, one would expect firms with greater cashflow uncertainty to hold more cash (Opler et al., 1999). Finally, the coefficients on capital expenditure are positive and statistically significant.

4.2. Organisational learning and corporate cash holdings

Table 4 reports evidence on the impact of COVID‐19 on corporate cash holdings, conditional on whether a firm has experienced prior epidemics such as SARS or H1N1. If firms learn from their past epidemic experiences and subscribe to the ‘cash is king’ mentality, we expect firms with prior epidemic exposure to save more cash, given the benefits of having cash on hand in a crisis. However, to the extent that prior epidemic experience manifests overestimation in a firm's preparedness to manage risk in a health‐related crisis, especially upon surviving the crisis, we hypothesise that such firms are likely to scale back on their ‘precautionary’ cash holdings. Our main variable of interest is the interaction term between COVID_EXPOSURE and PRIOR_EPID, where PRIOR_EPID is the sum of the number of times SARS (H1N1) is mentioned in a firm's earnings calls held in 2003 (2009), scaled by the number of words in the transcript. For firms with no prior SARS/H1N1 exposure, the effect of COVID‐19 on cash is ${\beta }_1$. Using results from column (2) for interpretation, a one‐unit increase in COVID_EXPOSURE is associated with a 0.024 unit increase in cash. However, the coefficient on the COVID_EXPOSURE * PRIOR _EPID interaction term is negative, ranging from −0.054 to −0.019, and statistically significant at the 1% level across all four specifications. These findings imply that firms with prior SARS and H1N1 exposure consistently save less cash relative to firms with no prior epidemic exposures. Thus, we reject the null hypothesis in 2A and find significant negative association between firms' prior exposure to SARS and H1N1 epidemic and corporate cash holdings.

TABLE 4.

Prior epidemic experience and cash holdings

Variable	Cash holdings		Ln cash holdings
	1	2	3	4
Intercept	0.526***	0.346**	0.378***	0.371***
	(9.13)	(2.05)	(17.08)	(6.78)
COVID_EXPOSURE	0.027***	0.024**	0.012***	0.012***
	(2.88)	(2.51)	(3.91)	(3.81)
PRIOR_EPID	0.002	−0.015	0.010	0.002
	(0.05)	(−0.41)	(0.71)	(0.12)
COVID_EXPOSURE * PRIOR _EPID	−0.046***	−0.054***	−0.019***	−0.021***
	(−2.66)	(−2.96)	(−2.86)	(−3.03)
SIZE	−0.047***	−0.048***	−0.032***	−0.033***
	(−7.64)	(−7.74)	(−13.83)	(−13.96)
MB	0.091***	0.082***	0.049***	0.044***
	(12.01)	(10.54)	(17.88)	(15.85)
CF	−1.118***	−1.090***	−0.367***	−0.347***
	(−9.98)	(−9.99)	(−11.35)	(−11.10)
NWC	−0.474***	−0.555***	−0.164***	−0.186***
	(−7.38)	(−7.97)	(−8.16)	(−8.93)
CAPEX	0.808**	1.315***	0.211*	0.510***
	(2.41)	(3.72)	(1.87)	(4.50)
LEVERAGE	−0.573***	−0.598***	−0.245***	−0.252***
	(−9.45)	(−9.42)	(−11.07)	(−11.13)
INDUSTRY_SIGMA	0.012*	−0.045***	0.012***	−0.017***
	(1.96)	(−5.63)	(5.16)	(−6.53)
R&D	0.110***	0.099***	0.030***	0.026***
	(9.11)	(8.30)	(10.01)	(9.02)
ACQUISITION_EXP	−0.540***	−0.569***	−0.254***	−0.291***
	(−5.35)	(−5.91)	(−5.88)	(−7.02)
CF_VOL	2.174***	2.061***	0.658***	0.619***
	(11.12)	(10.79)	(13.19)	(12.83)
Observations	24,667	24,667	24,667	24,667
R‐squared	0.58	0.59	0.60	0.62
Industry FE	No	Yes	No	Yes

Note: This table presents the results of estimating Equation (2) using OLS regression with Cash Holdings as dependent variable in columns (1) and (2) and Ln Cash Holdings as dependent variable in columns (3), and (4). CASH equals the ratio of cash and short‐term investments to net assets. LN_CASH equals the natural logarithm of One plus Cash Holdings. Following Hassan et al. (2021a), PRIOR_EPID is a dummy variable equal to 1 for firms that have positive Prior_epidemic value and 0 otherwise. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

Given that there exists evidence of organisational learning in corporate cash policies in the way that firms with prior epidemic experiences tend to save less going forwards, perhaps due to the firms' overestimation of their preparedness to manage the next crisis, we ask if this learning behaviour of cash is exclusive to a health‐related crisis or whether firms may also learn to hoard less cash after surviving a different type of crisis. Thus, we next analyse the extent to which firms manage their cash holdings, conditional on whether firms had experienced severe financial constraints during the last 2008 credit crisis and present our findings in Table 5. Our main variable of interest is the interaction term between COVID_EXPOSURE and SIZE_AGE, where SIZE_AGE is a dummy variable equals to 1 for firms with SA index higher than the median SA index of −3.66 and 0 otherwise. We find that the coefficient on the COVID_EXPOSURE * SIZE_AGE interaction term is negative, ranging from −0.045 to −0.018, and statistically significant at the 1% level across all four specifications. These results, similar to that of Table 4, suggest that having experienced severe financial constraints and survived a financial‐related crisis such as the Great Recession influence firms to save systematically less cash relative to those firms with no or little financial hardship. As such, we reject the null hypothesis in 2B, and find significant negative association between firms that experienced severe financial constraints during the 2008 financial crisis and cash holdings.

TABLE 5.

Prior financial constraints experience and cash holdings

Variable	Cash holdings		Ln cash holdings
	1	2	3	4
Intercept	0.590***	0.434**	0.402***	0.405***
	(9.15)	(2.51)	(16.48)	(7.25)
COVID_EXPOSURE	0.036***	0.032***	0.016***	0.015***
	(3.31)	(2.93)	(4.51)	(4.37)
SIZE_AGE	−0.019	−0.019	−0.007	−0.008
	(−0.67)	(−0.68)	(−0.64)	(−0.69)
COVID_EXPOSURE * SIZE_AGE	−0.045***	−0.045***	−0.018***	−0.018***
	(−2.71)	(−2.74)	(−3.22)	(−3.24)
SIZE	−0.054***	−0.057***	−0.035***	−0.036***
	(−8.57)	(−8.89)	(−14.53)	(−14.96)
MB	0.090***	0.081***	0.049***	0.044***
	(11.97)	(10.45)	(17.86)	(15.78)
CF	−1.108***	−1.079***	−0.364***	−0.343***
	(−9.93)	(−9.92)	(−11.29)	(−11.00)
NWC	−0.471***	−0.550***	−0.162***	−0.184***
	(−7.33)	(−7.91)	(−8.10)	(−8.86)
CAPEX	0.795**	1.308***	0.203*	0.506***
	(2.38)	(3.72)	(1.81)	(4.48)
LEVERAGE	−0.579***	−0.605***	−0.248***	−0.255***
	(−9.47)	(−9.43)	(−11.14)	(−11.18)
INDUSTRY_SIGMA	0.012*	−0.045***	0.012***	−0.016***
	(1.91)	(−5.61)	(5.08)	(−6.50)
R&D	0.110***	0.099***	0.030***	0.026***
	(9.08)	(8.28)	(9.97)	(8.97)
ACQUISITION_EXP	−0.540***	−0.569***	−0.255***	−0.292***
	(−5.34)	(−5.90)	(−5.89)	(−7.04)
CF_VOL	2.158***	2.048***	0.651***	0.613***
	(11.08)	(10.75)	(13.12)	(12.76)
Observations	24,667	24,667	24,667	24,667
R‐squared	0.58	0.59	0.60	0.62
Industry FE	No	Yes	No	Yes

Note: This table presents the results of estimating Equation (3) using OLS regression with Cash Holdings as dependent variable in columns (1) and (2) and Ln Cash holdings as dependent variable in columns (3), and (4). CASH equals the ratio of cash and short‐term investments to net assets. LN_CASH equals the natural logarithm of One plus Cash Holdings. SIZE_AGE is a dummy variable equal to 1 for firms with SA index higher than the median (−3.66) and 0 otherwise. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

Taken together, the results in Tables 4, and 5 suggest that the effect of COVID‐19 on corporate cash holdings is a function of its prior experiences. More specifically, facing the pandemic, firms with prior SARS/H1N1 exposure and those that experienced severe financial constraint during the 2008 credit crisis consistently save less cash than their inexperienced peers do. Our results highlight the importance of organisational learning in general, and prior experiences in particular, in shaping firms' cash policies moving forward.

5. ROBUSTNESS CHECKS

To check for the sensitivity of our analysis, we re‐estimate all models using two alternative measures of corporate cash holdings. Following Harford et al. (2008) and Deshmukh et al. (2021), we use ∆ CASH, measured as the current year's cash holdings minus last year's cash holdings, as the first alternative measure for cash. Following Bates et al. (2009), we also use CASH/TA, calculated as the ratio of cash and short‐term investments to book value of total assets, as an alternative dependent variable of cash holdings. Moreover, we also include undrawn line of credit as an additional control variable because firms that lack access to external capital may hoard more cash, irrespective of whether they are exposed to COVID‐19. Thus, in additional to firm size, we use UNDRAWN_LOC 18 as control for access to external market. We also include dividends (DIV) as additional control variable given that dividend may also be related to cash holdings.19 The results of our sensitivity analysis are reported in Table 6 for Equation (1). Even when alternative corporate cash measures are employed, our findings suggest a positive and statistically significant relationship between COVID‐19 and cash holdings, evidenced by the positive COVID_EXPOSURE coefficients ranging from 0.005 to 0.008 across all cash proxies. Next, we re‐run Equations (2) and (3) to examine the impact of the pandemic on cash conditioning on firms' prior experiences. We report the estimation results in columns (1) and (2) for the dependent variable ∆ CASH and in columns (3) and (4) for CASH/TA in Table 7. Consistent with our main findings, firms that experienced severe financial constraints due to the 2008 credit crisis and firms with prior SARS/H1N1 exposure tend to hold less cash than their peers do, as evidenced by the negative and statistically significant coefficients of the interaction terms.

TABLE 6.

COVID‐19 exposure and alternate cash holdings measures

Variable	Change in Cash Holdings		Cash/Total Assets
	1	2	3	4
Intercept	0.050***	0.050***	0.294***	0.294***
	(4.97)	(4.97)	(23.96)	(23.96)
COVID_EXPOSURE	0.008***	0.008***	0.004***	0.004***
	(3.90)	(3.90)	(3.06)	(3.06)
SIZE	−0.005***	−0.005***	−0.020***	−0.020***
	(−4.61)	(−4.61)	(−15.41)	(−15.41)
MB	−0.002*	−0.002*	0.031***	0.031***
	(−1.85)	(−1.85)	(22.55)	(22.55)
CF	0.014	0.014	−0.162***	−0.162***
	(0.76)	(0.76)	(−11.96)	(−11.96)
NWC	−0.065***	−0.065***	−0.076***	−0.076***
	(−5.95)	(−5.95)	(−7.90)	(−7.90)
CAPEX	−0.263***	−0.263***	0.106*	0.106*
	(−3.50)	(−3.50)	(1.79)	(1.79)
LEVERAGE	−0.019*	−0.019*	−0.150***	−0.150***
	(−1.89)	(−1.89)	(−12.87)	(−12.87)
INDUSTRY_SIGMA	0.001	0.001	0.007***	0.007***
	(1.14)	(1.14)	(5.52)	(5.52)
R&D	0.001	0.001	0.010***	0.010***
	(0.18)	(0.18)	(7.50)	(7.50)
ACQUISITION_EXP	−0.693***	−0.693***	−0.156***	−0.156***
	(−13.82)	(−13.82)	(−6.26)	(−6.26)
CF_VOL	−0.113***	−0.113***	0.269***	0.269***
	(−3.07)	(−3.07)	(13.16)	(13.16)
UNDRAWN_LOC	0.055***	0.055***	−0.123***	−0.123***
	(2.82)	(2.82)	(−7.30)	(−7.30)
DIV	0.150*	0.150*	−1.051***	−1.051***
	(1.83)	(1.83)	(−7.79)	(−7.79)
Observations	24,667	24,667	24,667	24,667
R‐squared	0.017	0.018	0.57	0.60
Industry FE	No	Yes	No	Yes

Note: This table presents the results of estimating Equation (1) with alternate cash holding measures as the dependent variable. In columns (1) and (2), $∆\mathit{\text{CASH}}$ equals to current year cash holdings minus last year's cash holdings. In columns (3) and (4), CASH/TA equals to the ratio of cash and short‐term investments to book value total assets. The COVID_EXPOSURE is the frequency of keywords related to spread of COVID‐19 in conference call transcripts. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

TABLE 7.

Prior epidemic, prior financial constraints and alternate cash holdings measures

Variable	Change in cash holdings		Cash/total assets
	1	2	3	4
Intercept	0.032	0.022	0.297***	0.311***
	(1.39)	(0.89)	(11.23)	(11.68)
COVID_EXPOSURE	0.008***	0.010***	0.006***	0.008***
	(3.36)	(3.36)	(3.98)	(4.52)
PRIOR_EPID	0.014**		0.012
	(2.44)		(1.37)
COVID_EXPOSURE * PRIOR_EPID	−0.005		−0.013***
	(−1.30)		(−3.37)
SIZE_AGE		0.014**		−0.002
		(2.45)		(−0.27)
COVID_EXPOSURE * SIZE_AGE		−0.007*		−0.010***
		(−1.82)		(−3.21)
SIZE	−0.005***	−0.004***	−0.020***	−0.021***
	(−4.45)	(−3.60)	(−15.14)	(−15.83)
MB	−0.002*	−0.002	0.027***	0.027***
	(−1.71)	(−1.64)	(20.03)	(19.95)
CF	0.014	0.013	−0.149***	−0.147***
	(0.76)	(0.72)	(−11.46)	(−11.36)
NWC	−0.076***	−0.077***	−0.086***	−0.086***
	(−6.11)	(−6.15)	(−9.00)	(−8.92)
CAPEX	−0.270***	−0.273***	0.310***	0.305***
	(−3.24)	(−3.29)	(5.41)	(5.35)
LEVERAGE	−0.017	−0.016	−0.151***	−0.152***
	(−1.61)	(−1.55)	(−12.67)	(−12.75)
INDUSTRY_SIGMA	0.005	0.005	−0.008***	−0.008***
	(1.60)	(1.61)	(−6.73)	(−6.66)
R&D	0.001	0.001	0.009***	0.009***
	(0.19)	(0.19)	(6.64)	(6.61)
ACQUISITION_EXP	−0.686***	−0.685***	−0.183***	−0.183***
	(−13.57)	(−13.57)	(−7.55)	(−7.58)
CF_VOL	−0.114***	−0.113***	0.253***	0.251***
	(−3.09)	(−3.05)	(12.77)	(12.67)
UNDRAWN_LOC	0.051**	0.051***	−0.111***	−0.111***
	(2.57)	(2.58)	(−6.94)	(−6.94)
DIV	0.112	0.119	−0.974***	−0.958***
	(1.36)	(1.43)	(−7.22)	(−7.09)
Observations	24,667	24,667	24,667	24,667
R‐squared	0.02	0.02	0.59	0.60
Industry FE	Yes	Yes	Yes	Yes

Note: This table presents the results of estimating Equations (2) and (3) with alternate cash holding measures as the dependent variable. In columns (1) and (2), $∆\mathit{\text{CASH}}$ equals to current year cash holdings minus last year's cash holdings. In columns (3) and (4), CASH/TA equals to the ratio of cash and short‐term investments to book value total assets. PRIOR_EPID is a dummy variable equal to 1 for firms that have positive Prior epidemic value and 0 otherwise. SIZE_AGE is a dummy variable equal to 1 for firms with SA index higher than the median (−4.12) and 0 otherwise. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

As we document the positive association between this widespread pandemic and corporate cash holdings, a concern with drawing inferences from investigating the OLS association between the COVID_EXPOSURE and CASH is that firms may use keywords related to the spread of COVID‐19 in conference call transcripts to increase COVID_EXPOSURE measure in anticipation of future levels of corporate cash. This implies that causality can run both ways, from COVID_EXPOSURE to cash holdings and vice versa, which raises concerns for endogeneity. We address this endogeneity concern in several ways. First, we replace COVID_EXPOSURE with Lag_COVID_EXPOSURE in the OLS model and report the results in columns (1) and (2) of Table 8. The coefficients on Lag_COVID_EXPOSURE (ranging 0.030–0.038) are all positive and statistically significant at the 1% level across both specifications. We interpret this result as Lag_COVID_EXPOSURE having a causal effect on CASH. Second, in place of Hassan et al.'s (2021a) COVID‐19 firm level risk, we use state‐level quarterly death rate (per 100,000 population), reported by the Centers for Disease Control and Prevention (CDC) to proxy for pandemic risk.20 The advantage of using COVID‐19 state‐level death rate is that this exogenous variable should be unrelated to firms' cash policies while capturing risk factors associated with the pandemic. We replace COVID_EXPOSURE with DEATH_RATE in the OLS model and report the results in columns (3) and (4) of Table 8. The coefficients on DEATH_RATE are positive and statistically significant at the 1% level in columns (3) and (4), suggesting that increasing in COVID‐19 risk, as measured by state‐level death rate, is associated with an increase in cash holdings. Third, we address the endogeneity concern using the instrumental variable (IV) estimation method that employs the 2‐stage least squares regressions and report the results in Table 9. To implement this empirical strategy, we need an instrument that would affect COVID_EXPOSURE, but that is unrelated to CASH. Since COVID‐19 has affected some industries more severely than others (e.g., retail, airline, cruises, etc.), we use the 2‐digit NAICS code industry average of COVID_EXPOSURE as an instrument for a firm's COVID‐19 exposure. We first check for the relevance of our instrument by regressing COVID_EXPOSURE on Industry_Average_Covid_Exposure. Next, we use the predicted value of COVID_EXPOSURE from this first‐stage regression as an explanatory variable in the second stage, where we regress cash holdings on this variable along with other control variables. Since the predicted value of COVID_EXPOSURE from the first stage captures the variation in COVID_EXPOSURE using the Industry_Average_Covid_Exposure as an instrument, we use this variation to estimate the slope of COVID_EXPOSURE in the second stage.

TABLE 8.

Lag COVID‐19 exposure, COVID‐19 death and cash holdings.

Variable	Cash holdings
	1	2	3	4
Intercept	0.519***	0.354**	0.505***	0.226
	(−9.25)	(−2.09)	(8.02)	(1.13)
DEATH_RATE			0.001***	0.001***
			(3.15)	(3.24)
LAG_COVID_EXPOSURE	0.038***	0.030***
	−5.04	−3.86
SIZE	−0.049***	−0.052***	−0.048***	−0.046***
	(−8.32)	(−8.64)	(−6.93)	(−6.68)
MB	0.089***	0.080***	0.080***	0.072***
	−11.82	−10.33	(10.21)	(8.94)
CF	−1.109***	−1.081***	−1.577***	−1.535***
	(−9.87)	(−9.84)	(−7.50)	(−7.57)
NWC	−0.479***	−0.558***	−0.415***	−0.490***
	(−7.43)	(−7.98)	(−5.92)	(−6.32)
CAPEX	0.860***	1.359***	1.646***	2.116***
	−2.64	−3.96	(3.28)	(4.10)
LEVERAGE	−0.564***	−0.590***	−0.473***	−0.500***
	(−9.37)	(−9.30)	(−7.34)	(−7.39)
INDUSTRY_SIGMA	0.014**	−0.040***	0.016**	−0.034***
	−2.34	(−5.03)	(2.14)	(−3.28)
R&D	0.112***	0.101***	0.105***	0.096***
	−9.09	−8.31	(6.67)	(6.22)
ACQUISITION_EXP	−0.516***	−0.546***	−0.454***	−0.534***
	(−5.13)	(−5.69)	(−3.83)	(−4.62)
CF_VOL	2.196***	2.084***	2.263***	2.152***
	−10.89	−10.54	(10.35)	(9.90)
Observations	24,667	24,667	16,119	16,119
R‐squared	0.58	0.6	0.63	0.64
Industry FE	No	Yes	No	Yes

Note: This table presents the results of estimating Equation (1) using OLS regression with cash as the dependent variable. Cash holdings equals to the ratio of cash and short‐term investments to net assets. In columns (1), and (2), the independent variable is the lag of COVID_EXPOSURE ure. In columns (3), and (4), the independent variable represents the death rate in the state where the company is headquartered. COVID_EXPOSURE is the frequency of keywords related to spread of COVID‐19 in conference call transcripts. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

TABLE 9.

COVID‐19 exposure and cash holdings: IV regressions

Variable	IV regressions: First‐stage results	IV regressions: Second‐stage results
		Cash holdings	Ln Cash holdings
	1	2	3
Intercept	0.355***	0.244***	0.281***
	(3.17)	(2.61)	(8.19)
INDUSTRY_AVERAGE_COVID_EXPOSURE	0.980***
	(12.53)
IV_COVID_EXPOSURE		0.215***	0.076***
		(3.91)	(3.86)
SIZE	−0.051***	−0.039***	−0.029***
	(−6.56)	(−5.84)	(−11.85)
MB	−0.026***	0.094***	0.051***
	(−5.08)	(12.51)	(18.37)
CF	−0.081	−1.092***	−0.359***
	(−1.26)	(−9.75)	(−11.05)
NWC	0.105***	−0.508***	−0.176***
	(3.28)	(−7.68)	(−8.51)
CAPEX	−2.710***	1.511***	0.450***
	(−9.56)	(3.65)	(3.30)
LEVERAGE	0.246***	−0.624***	−0.263***
	(4.42)	(−9.67)	(−11.30)
INDUSTRY_SIGMA	0.054***	0.009	0.011***
	(7.52)	(1.36)	(4.47)
R&D	−0.017**	0.112***	0.030***
	(−2.19)	(9.32)	(10.28)
ACQUISITION_EXP	−1.441***	−0.289**	−0.168***
	(−8.46)	(−2.18)	(−3.23)
CF_VOL	0.164*	2.129***	0.642***
	(1.79)	(10.99)	(12.98)
Observations	24,667	24,667	24,667
R‐squared	0.08	0.58	0.60
First‐Stage F‐Statistics	505***
Durbin–Wu–Hausman Test
Ho: variables are exogenous		p = 0.00	p = 0.00

Note: This table presents the results of estimating Equation (1) using IV regressions. Cash holdings equals the ratio of cash and short‐term investments to net assets. Ln cash holdings equals the natural logarithm of one plus cash holdings. The COVID_EXPOSURE is the frequency of keywords related to spread of COVID‐19 in conference call transcripts. Column (1) reports the first‐stage results of the instrumental variable (IV) regressions. Columns (2) and (3) report the second‐stage results of the IV regressions with cash holding and ln cash holdings as the dependent variables respectively. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

Column (1) of Table 9 reports the first‐stage results of the IV regressions. The results in column (1) show that the selected instrument (Industry_Average_Covid_Exposure) is positive and statistically significant at the 1% level, implying that the instrument is relevant. Moreover, the first‐stage F‐statistics value of 505 indicates that the selected instrument (Industry_Average_Covid_Exposure) has significant power in explaining the endogenous variable (COVID_EXPOSURE). Columns (2) and (3) report the second‐stage results of the IV regressions where the Durbin–Wu–Hausman endogeneity test results substantiate our endogeneity concern, which implies that the OLS estimates are biased and inconsistent. In contrast, the IV model estimates are unbiased and consistent. Collectively, these tests indicate that our selected instrument is both valid and relevant. Consistent with the OLS and the lagged regression results, the 2‐stage least squares estimation results show that COVID_EXPOSURE is positively related to cash holdings.

Thus far, our result of a positive association between COVID‐19 risk and cash holdings‐ obtained from a battery of robustness tests – is consistent with the precautionary motive suggested in prior literature (Acharya & Steffen, 2020; Almeida, 2021; Darmouni & Siani, 2022; Hotchkiss et al., 2022; Opler et al., 1999; Pettenuzzo et al., 2021). Next, we test whether CEO's experience in dealing with prior crisis impacts the association between corporate cash holdings and its prior experiences. Our argument throughout the paper is that firms learn from prior experiences. Thus, whether the same management team during the prior experience still lead the company at the time of COVID‐19 plays an important role in a firm's learning behaviour of cash. If they do, they may use their experience in dealing with prior crises to navigate the present pandemic. Our results show that firms previously experienced SAR1/H1N1 or financial constraints during the 2007–2009 financial crisis hoard less cash than its peers do. To further isolate the learning effect from other confounding factors such as liquidity injection by the Federal Reserve in the early 2020, or depressed managerial sentiments as a result of work from home (Ng et al., 2022), and since it is not possible to track the movement of all company managers, we focus our attention on the CEO of the company. We use the natural departure of CEOs21 (such as those with natural retirement or departure caused by personal health issue) who have prior experience with SARS/H1N1 and the 2008 financial crisis as an exogenous shock to the firm's prior experience.22 This identification test helps us examine the extent to which CEO's experience in dealing with prior crises impacts corporate cash holdings during the pandemic.

Our findings in Table 10 show that firms with experienced CEOs, defined as those with tenure starting on or before the SARS/H1N1 epidemic or the last financial crisis of 2008 and continuing through the COVID‐19 pandemic, save significantly less cash than firms with inexperienced CEOs do (CEOs that depart prior to the COVID‐19 pandemic due to retirement or personal illness). More specifically, the coefficient on the interaction variable COVID_EXPOSURE * PRIOR_EPID (COVID_EXPOSURE * SIZE_AGE) is −0.092 (−0.102) and significant at the 5% (1%) level, respectively, for the sample of experienced CEOs while they are insignificant for the CEO sample with natural departure.23 The notion that firms led by the same CEOs as in prior crises hold less cash during COVID‐19 lends further credence to the crucial role that learning from prior experiences plays in shaping corporate cash policies.

TABLE 10.

CEO prior experience subsample analysis: Prior epidemic and prior financial constraints

Variable	Cash holdings (CEO_priorexp = 0)			Cash holdings (CEO_priorexp = 1)
	1	2	3	4	5	6
Intercept	0.428***	0.448***	0.419***	0.447***	0.438***	0.797***
	(5.05)	(4.85)	(3.91)	(4.02)	(3.41)	(4.65)
COVID_EXPOSURE	0.023**	0.025*	0.019	0.035*	0.038*	0.065**
	(2.06)	(1.76)	(1.41)	(1.85)	(1.65)	(2.01)
PRIOR_EPID		0.023			−0.100
		(0.83)			(−1.32)
COVID_EXPOSURE * PRIOR_EPID		−0.012			−0.092**
		(−0.63)			(−2.10)
SIZE_AGE			0.011			−0.063
			(0.34)			(−1.06)
COVID_EXPOSURE * SIZE_AGE			0.009			−0.102***
			(0.38)			(−2.86)
SIZE	−0.029***	−0.033***	−0.030***	−0.043***	−0.029*	−0.056***
	(−3.50)	(−3.95)	(−3.15)	(−2.78)	(−1.79)	(−3.28)
MB	0.013*	0.016**	0.016**	0.096***	0.087***	0.084***
	(1.89)	(2.21)	(2.27)	(6.74)	(5.80)	(5.71)
CF	−0.019	−0.014	−0.016	−1.011***	−1.037***	−0.974***
	(−0.13)	(−0.10)	(−0.11)	(−4.34)	(−4.58)	(−4.35)
NWC	−0.408***	−0.497***	−0.501***	−0.318***	−0.410***	−0.397***
	(−3.54)	(−3.94)	(−4.06)	(−3.40)	(−3.42)	(−3.34)
CAPEX	−0.102	−0.235	−0.232	1.393*	1.693**	1.584**
	(−0.29)	(−0.68)	(−0.67)	(1.90)	(2.28)	(2.15)
LEVERAGE	−0.287***	−0.308***	−0.310***	−0.628***	−0.667***	−0.706***
	(−4.38)	(−4.14)	(−4.24)	(−4.05)	(−4.15)	(−4.29)
INDUSTRY_SIGMA	−0.000	−0.002	−0.002	0.007	−0.038**	−0.039**
	(−0.01)	(−0.39)	(−0.33)	(0.60)	(−2.47)	(−2.51)
R&D	0.975***	0.910***	0.881***	0.076***	0.063***	0.063**
	(2.91)	(2.73)	(2.61)	(3.06)	(2.62)	(2.54)
ACQUISITION_EXP	−0.259***	−0.212**	−0.211**	−0.544*	−0.594**	−0.538**
	(−2.87)	(−2.53)	(−2.50)	(−1.82)	(−2.26)	(−2.08)
CF_VOL	1.134**	1.111**	1.121**	3.792***	3.553***	3.524***
	(2.34)	(2.40)	(2.44)	(6.71)	(6.43)	(6.34)
Observations	1952	1952	1952	5590	5590	5590
R‐squared	0.34	0.36	0.36	0.58	0.60	0.60
Industry FE	Yes	Yes	Yes	Yes	Yes	Yes

Note: This table presents the results of estimating Equations ((1), (2), (3)) using OLS regression with cash holding as the dependent variable. Cash holdings equals the ratio of cash and short‐term investments to net assets. PRIOR_EPID is a dummy variable equals to 1 for firms that have positive Prior_epidemic value and 0 otherwise. SIZE_AGE is a dummy variable equals to 1 for firms with SA index higher than the median and 0 otherwise. T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

The evidence summarised thus far indicates that companies exposed to COVID‐19 hold more cash, but those with prior experiences hold less cash compared to less experienced peers. Darmouni and Siani (2022) find that as bond yields fell, firms issued bonds to accumulate large and persistent amounts of liquid assets instead of investing. Ng et al. (2022) further show that a portion of cash holdings caused by managers' behavioural bias decreases the value. Thus, we examine the value of cash conditional on prior experiences. We follow the approach adopted by Bates et al. (2009) which uses a valuation regression developed by Fama and French (1998) and modifies it by introducing cash holdings as an independent variable. We augment the baseline model to include LN_CASH, SIZE_AGE (PRIOR_EPID) and an interaction variable of LN_CASH and SIZE_AGE (PRIOR_EPID) as follows:

$$\begin{array}{cc}{V}_{i,t}& =\alpha +{\beta }_1{E}_{i,t}+{\beta }_2{\mathit{dE}}_{i,t}+{\beta }_3{\mathit{dE}}_{i,t+2}+{\beta }_4{\mathit{dNA}}_{i,t}+{\beta }_5{\mathit{dNA}}_{i,t+2}\hfill \\ & +{\beta }_6{\mathit{RD}}_{i,t}+{\beta }_7{\mathit{dRD}}_{i,t}+{\beta }_8{\mathit{dRD}}_{i,t+2}+{\beta }_9{I}_{i,t}+{\beta }_{10}{\mathit{dI}}_{i,t}\hfill \\ & +{\beta }_{11}{\mathit{dI}}_{i,t+2}+{\beta }_{12}{\mathit{DIV}}_{i,t}+{\beta }_{13}{\mathit{\text{dDIV}}}_{i,t}+{\beta }_{14}{\mathit{\text{dDIV}}}_{i,t+2}+{\beta }_{15}{\mathit{dV}}_{i,t+2}\hfill \\ & +{\beta }_{16}\mathit{LN}\_{\mathit{\text{CASH}}}_{i,t}+{\beta }_{17}\mathit{\text{SIZE}}\_\mathit{AGE}\left(\mathit{\text{PRIOR}}\_\mathit{\text{EPID}}\right)+{\beta }_{17}\mathit{LN}\_{\mathit{\text{CASH}}}_{i,t}\hfill \\ & *\mathit{\text{SIZE}}\_\mathit{AGE}\left(\mathit{\text{PRIOR}}\_\mathit{\text{EPID}}\right)+{\epsilon }_{i,t},\hfill \end{array}$$ (4)

where Xt is the level of variable X in year t divided by the level of total assets in year t; dXt is the change in the level of X from year t − 2 to year t, Xt − Xt − 2; dXt + 2 is the change in the level of X from year t to year t + 2, Xt + 2 − Xt; V is the market value of the firm calculated at fiscal year‐end as the sum of the market value of equity, the book value of short‐term debt, and the book value of long‐term debt; E is earnings before extraordinary items plus interest, deferred tax credits, and investment tax credits; NA is net assets defined as total assets minus cash; RD is the research and development (R&D) expense; I is the interest expense; and DIV is dividends defined as common dividend paid. When R&D is missing, we set it equal to zero. LN_CASH equals the natural logarithm of One plus Cash Holdings. SIZE_AGE and PRIOR_EPID are prior experience variables, which capture the extent to which firm experiences severe financial constraints during the previous 2008 credit crisis and the severity of prior SARS/H1N1 exposure during the peak of the epidemic in 2003 and 2009.

Table 11 presents our findings on the impact of prior experience on the value of cash. Column (1) reports the baseline regression results that are consistent with those reported by prior studies. Column (2) includes the SIZE_AGE dummy and the interaction term between LN_CASH and SIZE_AGE. We find that the coefficient on the LN_CASH * SIZE_AGE interaction term is negative (−0.127) and statistically significant at the 5% level. Column (3) includes the PRIOR_EPID dummy and the interaction term between LN_CASH and PRIOR_EPID. We find that the coefficient on the LN_CASH * PRIOR_EPID interaction term is negative (−0.271) and statistically significant at the 1% level. There are two possible explanations for the results. On the one hand, these results are consistent with the behavioural explanation documented in Ng et al. (2022) in that for firms with prior exposure to economic shock and/or previous epidemics, if CEOs overreact by hoarding cash, this higher level of cash is associated with a lower value of cash as seen from the negative coefficients of the interaction terms. On the other hand, by combining regression results of both the level and the value of cash, we also consider an alternative interpretation in the case of managers learning from their prior experience and reducing their cash holdings in the first place. In this case, instead of hoarding cash, CEOs of experienced firms actually reduce the level of cash to mitigate the negative effect of cash holdings on shareholder value.24

TABLE 11.

Market value of cash holdings.

Variable	Market value of the firm
	1	2	3
Intercept	1.520***	1.672***	1.595***
	(16.14)	(14.91)	(15.79)
Et	5.269***	5.418***	5.379***
	(5.21)	(5.27)	(5.28)
dEt	−2.406***	−2.471***	−2.456***
	(−4.07)	(−4.11)	(−4.11)
dEt + 2	0.159*	0.156*	0.163*
	(1.82)	(1.78)	(1.86)
dNAt	1.128***	1.071***	1.129***
	(4.83)	(4.60)	(4.83)
dNAt + 2	0.167	0.149	0.153
	(1.34)	(1.20)	(1.24)
RDt	28.333***	28.317***	28.418***
	(12.26)	(12.17)	(12.29)
dRDt	−14.832***	−14.725***	−15.015***
	(−6.02)	(−5.95)	(−6.09)
dRDt + 2	1.231***	1.203***	1.197***
	(3.18)	(3.13)	(3.11)
It	−3.222	−3.873	−2.346
	(−0.39)	(−0.46)	(−0.28)
dIt	−26.205**	−25.686**	−26.086**
	(−2.52)	(−2.47)	(−2.53)
dIt + 2	0.014	0.070	0.065
	(0.01)	(0.07)	(0.06)
DIVt	13.076***	13.160***	13.449***
	(5.15)	(5.19)	(5.25)
dDIVt	−6.062***	−6.056***	−6.233***
	(−3.86)	(−3.84)	(−3.94)
dDIVt + 2	−1.082	−1.056	−1.022
	(−1.27)	(−1.23)	(−1.20)
dVt + 2	−0.003	−0.003	−0.003
	(−0.76)	(−0.75)	(−0.79)
LN_CASHt	0.434***	0.478***	0.456***
	(15.61)	(13.83)	(15.22)
SIZE_AGE		−0.462***
		(−2.70)
LN_CASHt * SIZE_AGE		−0.127**
		(−2.38)
PRIOR_EPID			−0.691***
			(−3.01)
LN_CASHt * PRIOR_EPID			−0.271***
			(−3.32)
Observations	24,667	24,667	24,667
R‐squared	0.21	0.22	0.22
Industry FE	Yes	Yes	Yes

Note: This table presents the OLS regression results for the market value of cash holdings. The dependent variable for the regressions is the market value of the firm in year t, Mt. For each independent variable X, Xt is the level in year t, divided by the level of total assets in year t; dXt is the change in the level of X from year t − 2 to year t, divided by total assets in year t ((Xt − Xt − 2)/At); dXt + 2 is the change in the level of X from year t + 2 to year t, divided by assets in year t ((Xt + 2 − Xt)/At). Column (1) presents results of the estimating Equation (4) for the baseline regression adopted by Bates et al. (2009). Columns (2) and (3) present results of the estimating Equation (4) augmented to include LN_CASH, SIZE_AGE (PRIOR_EPID) and an interaction variable of LN_CASH and SIZE_AGE (PRIOR_EPID) T‐statistics based on heteroscedasticity‐robust standard errors clustered by firms are reported in parentheses. Variables are defined in Appendix 1: Table A1. All continuous variables are Winsorised at the 1% and 99% levels. The significance levels at the 1%, 5%, and 10% are indicated with ***, **, and *, respectively.

6. CONCLUSION

In this paper, we investigate the impact of COVID‐19 on corporate cash holdings and the extent to which firms learn to manage their liquidity as a function of prior experience. Using firm‐level COVID‐19 exposure during 2020 and through the first quarter of 2022, we find a positive and significant relation between pandemic exposure and cash holdings. In particular, a one‐standard‐deviation increase in COVID‐19 exposure is associated with a 2.44 percentage point or $32.2 million increase in cash holdings. While there is little doubt that the current COVID‐19 pandemic influences corporate cash policy, how firms learn from prior experiences to manage their cash reserves is an important empirical question that has yet to be explored in prior studies. Our analysis lends support for a learning behaviour of cash in that prior experiences, including past crises of the Great Recession and the prior SARS/H1N1 epidemic exposure, lead firms to alter cash hoard systematically. In particular, facing the COVID‐19 pandemic, firms with prior SARS/H1N1 exposure save less cash relative to their peer firms with no or little SARS/H1N1 exposure. Additionally, firms that experienced severe financial constraint during the 2008 credit crisis have lower cash holdings than their financially unconstrained counterparts do. Our findings are robust via a battery of robustness tests that address endogeneity concerns and employ alternative measures of cash as well as COVID‐19 exposures. We also conduct additional sub‐sample analysis in further testing the learning behaviour of cash as well as the value of cash. This evidence suggest the existence of a learning effect on corporate cash holdings as a result of prior experiences.

Our paper makes several contributions to the existing cash holdings and the management literature on organisational learning. To the best of our knowledge, this paper is the first to empirically examine the learning behaviour of cash across multiple crises, including SARS/H1N1, the financial crisis of 2007–2009, and COVID‐19. Second, we bridge the gap between critical corporate policies such as cash holdings and organisational learning by testing the learning hypothesis in the corporate cash setting. Although this study provides evidence in support of organisational learning in corporate cash management, the channel(s) as to how prior experiences induce firms to save less cash, relative to their inexperienced peers, is not a focus of this study. As such, the need for deeper understanding of how and why past experiences drive firms' learning to lower cash holdings will certainly be an interesting and fruitful one to be explored in the future. Nevertheless, our findings of the extent to which firms learn from prior experiences to manage their liquidity downward are expected to have meaningful implications for corporate executives and board of directors in establishing cash policies and in achieving a better understanding of the learning behaviour of cash.

APPENDIX 1.

TABLE A1.

Variables construction

Variables	Description
CASH	(Cash + Marketable securities)/book value of net assets calculated as CHEQ/(ATQ‐CHEQ) where CHEQ is Cash and Short‐Term Investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
LN_CASH	Log Cash Holdings equals the natural logarithm of One plus Cash holdings
COVID_EXPOSURE	We use data provided by Hassan et al. (2021a). The Covid_exposure measure counts the frequency of keywords related to spread of COVID‐19 in conference call transcripts
PRIOR_EPID	Following Hassan et al. (2021a), Prior_epidemic is the sum of the number of times SARS (H1N1) is mentioned in firm i's earnings calls held in 2003 (2009) (measured at the peak of their outbreaks in 2003 (2009)), scaled by the number of words in the transcript. PRIOR_EPID is a dummy variable equal to 1 for firms that have positive Prior_epidemic value and 0 otherwise
SIZE_AGE	Following Hadlock and Pierce (2010), the SA index is defined as [−0.737 × log(TOTAL_ASSETS)] + [0.043 × log(TOTAL‐ASSETS)2] − (0.040 × AGE). TOTAL‐ASSETS are Winsorised at $4500 million and AGE is Winsorised at 37 years. Higher values of the SA index imply greater levels of financial constraint. SIZE_AGE is a dummy variable equal to 1 for firms with SA index higher than the median (−4.12) and 0 otherwise
SIZE	The natural logarithm of the book value of total assets calculated as Log (ATQ) where ATQ is total asset (Compustat quarterly data #44)
MB	Market value of assets over book value of assets calculated as (ATQ‐CEQQ + CSHOQ × PRCCQ)/ATQ where ATQ is total asset (Compustat quarterly data #44), CEQQ is total common equity (Compustat quarterly data #59), CSHOQ is common shares outstanding (Compustat quarterly data #61), and PRCCQ is stock price at the end of the quarter
CF	Operating income before depreciation, net of interest, taxes, and dividends/book value of net assets calculated as (OIBDPQ‐XINTQ‐DVY‐TXTQ + DPQ)/(ATQ‐CHEQ) where OIBDPQ is Operating Income Before Depreciation (Compustat quarterly data #21), XINTQ is interest expense (Compustat quarterly data #22), DVY is cash dividends (Compustat quarterly data #89), TXTQ is income taxes (Compustat quarterly data #6), DPQ is depreciation and amortisation (Compustat quarterly data #5), CHEQ is Cash and Short‐Term Investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
NWC	(Net working capital – Cash – Marketable securities)/book value of net assets calculated as (ACTQ‐LCTQ‐CHEQ)/(ATQ‐CHEQ) where ACTQ is current assets (Compustat quarterly data #40), LCTQ is current liabilities (Compustat quarterly data #49), CHEQ is cash and short‐term investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
CAPEX	Capital expenditures/book value of net assets computed as CAPXY/(ATQ‐CHEQ) where CAPXY is capital expenditures (Compustat quarterly data #90), CHEQ is Cash and Short‐Term Investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
LEVERAGE	(Long‐term debt + Debt in current liabilities)/Book value of assets calculated as (DLTTQ+DLCQ)/ATQ where DLTTQ is total long term debt (Compustat quarterly data #51), DLCQ is debt in current liabilities (Compustat quarterly data #45) and ATQ is total asset (Compustat quarterly data #44)
INDUSTRY_SIGMA	The mean of the standard deviation of cash flow to net asset ratio over 5 years of firms in the same 2‐digit NAICS code industry
R&D	Research and development expense/sale calculated as XRDQ/SALEQ where XRDQ is Research and development expense (Compustat quarterly data #4)and SALEQ is sales revenue (Compustat quarterly data #2)
ACQUISITION_EXP	Acquisition expense/Book value of net assets calculated as AQCY/(ATQ‐CHEQ) where as AQCY is Acquisitions (Compustat quarterly data #94), CHEQ is Cash and Short‐Term Investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
CF_VOL	The standard deviation of a firm's quarterly cash flow to net asset ratio over 12 quarters
UNDRAWN_LOC	Undrawn line of credit/Book value of net assets calculated as UNDRAWN/(ATQ‐CHEQ) where as UNDRAWN is remaining portion of revolving line of credit (Capital IQ data UNDRAWNCRDTPORTIONREVOLVINGCRDT), CHEQ is Cash and Short‐Term Investments (Compustat quarterly data #36) and ATQ is total asset (Compustat quarterly data #44)
CEO_PRIOREXP	The CEO_PRIOREXP is a dummy variable equal to 1 if the current CEO was also the CEO of the same company in 2003 and/or 2008, and 0 if a CEO has prior experience in 2003 and/or 2008 but voluntarily departed (natural retirement or departure caused by personal illness) before the COVID‐19 crisis
V	Market value of the firm calculated at fiscal year‐end as the sum of the market value of equity, the book value of short‐term debt, and the book value of long‐term debt
E	Earnings before extraordinary items plus interest, deferred tax credits, and investment tax credits
NA	Net assets defined as total assets minus cash
RD	Research and development (R&D) expense. When R&D is missing, we set it equal to zero
I	Interest expense
DIV	Dividends defined as common dividend paid

BC, B. & Simpson, T. (2022) How do firms learn? Evidence from corporate cash holdings during the COVID‐19 pandemic. Accounting & Finance, 00, 1–32. Available from: 10.1111/acfi.13031

DATA AVAILABILITY STATEMENT

Data derived from public domain resources. The primary data that support the findings of this study are available in Compustat at https://wrds‐www.wharton.upenn.edu and firm risk data is openly available at https://www.firmlevelrisk.com. Supplementary data sources are openly available and are as indicated in the paper.