key: cord-0826468-069w3v2d
authors: Apergis, Nicholas; Danuletiu, Dan; Xu, Bing
title: CDS spreads and COVID-19 pandemic
date: 2021-11-18
journal: Journal of International Financial Markets, Institutions and Money
DOI: 10.1016/j.intfin.2021.101433
sha: 990579f1d2d5e8b1b63581a40526c2028ae0506e
doc_id: 826468
cord_uid: 069w3v2d

US corporate credit default swap (CDS) spreads have significantly increased since the beginning of the COVID-19 global pandemic. This paper shows that the magnitude of the pandemic measured by the number of COVID-19 cases and deaths both in the US and globally are positively linked to the CDS spreads, an effect both economically and statistically significant. However, there is a significant heterogenous effects across sectors, in which banking, travel & leisure, transportation, airlines, and restaurants are the hardest hit sectors. The analysis also documents that the COVID-19 pandemic increases corporate CDS spreads via increased firms’ distress levels transmission channels.

Corporate credit default swaps (CDS) are credit derivative contracts designed to allow investors to buy protection to swap/transfer credit risks on bonds or loans. The CDS buyer pays a premium to the seller to compensate the buyer if a specific credit event occurs, typically the default of a third debtor. Just like in insurance contracts, the buyer pays a periodic premium to a CDS seller in return for compensation should the credit event (accident) occurs. The natural buyers of corporate CDS would be large bondholders or banks that wish to enhance the creditworthiness of their loan portfolios. The price or spread of CDS reflects the probability of a firm failing to repay their debts in full, serving as an important indicator of firms' credit risks, especially during periods of uncertainty (Chiaramonte and Casu, 2013) . Therefore, a better understanding of the nature of CDS prices and their determinants are crucial to many decision makers in macroeconomic policy, investment/production decisions, consumption, management risks, and portfolio management. The ability to accurately gauge the CDS spreads can support both retail and institutional investors on the design of effective investment strategies i.e., to assess and select firms to collaborate with or invest in or adjust their investment portfolios if certain firms/sectors become unstable. In addition, CDS spreads can also act as an early warning indicator to alarm and assist managers to prevent the occurrence of credit failure.

The recent literature has contributed substantially to comprehend the causes of CDS spread fluctuations. The typical strategy is to search factors that could alter a firm's value and its ability to meet debt repayments, and hence, its default probability. Most studies find that CDS spreads respond to various firm-specific and macroeconomic variables, such as credit rating, profitability, liquidity, stock returns and their volatility (Abid and Naifar, 2006; Corò et al., 2013) . It is also widely accepted that CDS spreads are affected by macroeconomic conditions and the business cycle (Kim et al., 2017; Pereira et al., 2018) . The ongoing COVID-19 pandemic is not only a health crisis, but also a social, economic, and political catastrophe. The global economy has been suffering from the effects and many economies have been led into recession (Ludvigson et al., 2020) . According to IMF's World Economic Outlook, the global economy is predicted to suffer a 4.4% contraction in global GDP, which is far worse than the 2008-2009 financial crisis (-0.1%). The US unemployment rate reached to 15% for the second quarter of 2020 4 ; the UK economy suffered a shocking 10% contractions 5 and Japan economy diminished 4.8% in 2020. 6 Hence, it is not surprising that researchers have begun to link the impact of COVID-19 on CDS spreads. The worsened financial soundness on both macro-level and firm-level can increase the risk of default in debt, therefore, the spread of CDS can respond positively to the severity of COVID-19 pandemic (Bosio et al., 2020; Didier et al., 2020) .

Hence, it is not surprising that researchers have begun to link the impact of COVID-19 on CDS spreads, as the repercussions of this crisis greatly affect countries throughout the globe (Bosio et al., 2020; Didier et al., 2020) . The goal of our paper aims to build upon on this strand of the literature. In particular, it empirically examines the relationship between US corporate CDS spreads and COVID-19 pandemic shocks, while accounting for heterogeneity and transmission channels.

The COVID-19 pandemic can transmit its effect onto the CDS market through its impact on firms' future cash flows and discount rate channels. The radial steps imposed by administrations (e.g., lockdowns, border shutdowns, social distancing, travel restrictions) to halt the spread of the pandemic have sparked, among other things, tremendous undesirable 4 https://fred.stlouisfed.org/series/UNRATE 5 https://www.independent.co.uk/news/business/uk-economy-latest-gdp-recession-ons-b1801277.html 6 https://www.bbc.co.uk/news/business-56066065#:~:text=Japan's%20economy%20surged%20in%20the,negative%20growth%20for%20the%20year.& text=Japan's%20economy%20shrank%204.8%25%20over,its%20first%20contraction%20since%202009. consequences on demand, firms' profitability and growth, income, and productivity. The COVID-19 pandemic has sparked a massive spike in uncertainty and causing stock market volatility risen to levels that have rarely been seen since the Great Moderation (Baker et al., 2020) . The behavioural finance and psychology literature argues that economic agents behave differently in times of turbulences and fear as opposed to periods of prosperity and calm (Akerlof and Shiller, 2009 ). To be more specific, the unprecedented level of uncertainty caused by the pandemic, and various measures (e.g., lockdowns, curfew, travel restrictions) to halt the spread have caused, among other things, significant and far-reaching impacts on demand, reductions in firms' profitability and growth, income, employment, and productivity. For instance, managers may fail to identify best investment opportunities to standstill on new business formation, R&D expenditures, human capital investments due to unwillingness to bear risks, or equally, due to a shift in their preferences (Bloom et al., 2007; Caglayan and Xu, 2016a; Baker et al., 2020) . Fear and panic triggered by COVID-19 can reduce firms' profitability and growth from the demand side via a steady decline in customers' physical attendances (Liu et al., 2021) .

Households may increase their precautionary savings and smooth their consumptions in response to positive pandemic shocks as they perceive a greater chance of future unemployment (Jordà et al., 2020) . Furthermore, the interconnectedness among financial institutions, firms, households, and governments can accelerate the transmission effects of the pandemic shocks into the CDS market (e.g., Brei et al., 2020; Barua and Barua, 2020) . For instances, banks face withdrawals of deposits, funds and credit from businesses and households because their deteriorated current financial conditions and anticipated forthcoming cash flow concerns caused by the sustained pandemic shocks. On the other hand, firms/households with serious liquidity problem are more prone to default will reflect onto the banking system later causing banks holding lower quality assets (e.g., Bosio et al., 2020; Didier et al., 2020) . Taking all these together, the COVID-19 can have substantial negative impact on the financial institutions' ability to offering credits to private sector, efficiency and operational performance, and adversely affects the stability of the financial system (Caglayan and Xu, 2016b; Baum et al., 2020) .

Furthermore, there is ample evidence on financial institutions' reluctance to extend loans when uncertainty reaches extreme levels, because, in general, lenders can suffer more from asymmetric information issues that affecting their judgements on the expected returns from lending and enlarged monitoring costs as businesses are more prone to default (Caglayan and Xu, 2016a) . Furthermore, pandemic shocks can adversely affect CDS prices through the discount rate channel as monetary policy makers tend to carry out aggressive expansionary policies to combat the recession and kept interest rate low (Bernanke et al., 1997) . All these, in return, will impact firms' default probability and ultimately CDS spreads.

Our paper incorporates important innovations relative to previous literature. First, by employing both global and local COVID proxies to measure the severeness of the pandemic, it provides new insights on the turmoil effect of COVID-19 on the US corporate CDS market.

Despite all businesses are likely to experience significant economic losses due to the pandemic, it is crucial to reflect the heterogeneous effects across sectors. Furthermore, there is limited attention on searching for different transmission channels of COVID-19 pandemic on the spread of CDS.

To carry out our investigation, the analysis constructs a large panel of US corporate CDS data collected from Bloomberg. The final dataset that we employ in our analysis is comprised of 386 firms with a sample of 66,392 daily observations. The analysis covers the period between February 2020 to September 2020. The findings can be summarized as follows. First, we show that the magnitude of the pandemic is positively linked to the US corporate CDS spreads, an effect both economically and statistically significant. The results highlight the heterogenous effects across sectors, in which banking, travel & leisure, transportation, airlines, and restaurants are the hardest hit sectors in the US. By identifying the heterogeneity of sectoral sensitivities to the pandemic implies that some sectors can provide a channel for diversification during further waves of pandemic cases and fatalities. For example, if the cases and deaths are expected to increase in the future, investors may use this information to devise their investment strategies by taking short positions in banking, transportation, airlines, restaurants, and long positions in media, technology, telecommunications, pharmaceutical, information and data technology firms. Furthermore, the empirical results confirm the prediction that the pandemic adversely affects the CDS market through several transmission channels. To be more specific, the analysis provides strong supportive evidence of pandemic measures for increasing firms' likelihood of default considerably.

The rest of the paper is organised as follows. Section 2 reviews a brief review of prior literature. Section 3 presents the econometric methodology and describes the data employed.

Sections 4 and 5 present the empirical results, robustness checks and discuss the economic significance of the findings. Finally, Section 6 concludes.

The investigation relates to several strands of earlier studies that have examined the CDS markets. This section highlights some of the major studies along these lines.

Researchers have expended a considerable effort on identifying the key determinants of CDS spreads. Most studies find that both firm-and market-specific factors are useful in explaining CDS prices. Abid and Naifar (2006) , for example, find that credit ratings, CDS contract maturity, stock volatility, risk-free interest rates, and yield curves explain 60% of the variation in European CDS pricing, where credit ratings were considered as the most influential factor. Chiaramonte and Casu (2013) investigate five-year senior bank CDS spreads' determinants using several key performance indicators (e.g., return on equity (ROE)) and return on assets (ROA)), along with balance sheet indicators (e.g., net loan/deposits, liquid assets/deposits). They find that liquidity variables appear to have the most significant impact on banks' CDS pricing.

Following the 2008 financial crisis, several researchers have begun to explore the interrelations between risk and CDS spreads. Corò et al. (2013) study the role of credit and liquidity factors in explaining corporate CDS price changes during normal and crisis periods.

They find that liquidity plays a critical role in driving the CDS price fluctuations during the 2007-2009 financial crisis. Baum and Wan (2010) examine how US CDS spreads respond to macroeconomic uncertainty measured by GDP growth, and firm-specific characteristics (e.g., stock values, volatility and returns, ROE, dividend pay-outs, credit ratings). Their findings highlight that macroeconomic uncertainty generates tight credit restrictions for firms, by affecting firms' leverage and increasing the default risk of firms, while the effects of dividend pay-outs and ROE on the CDS spread differ greatly according to issuers' credit rating. Kim et al (2017) also show that business macroeconomic conditions and business cycles can determine changes in CDS spreads. Several authors focus on examining the role of uncertainty about the probabilities of future outcomes in addition to risk. For instance, Augustin and Izhakian (2020) explore the impact of the ambiguity on CDS spreads, while they find a negative impact on CDS spreads since major counterparties in CDS markets are net buyers of CDS protections. They also highlight that the effects of ambiguity and risk on CDS spreads are more pronounced in longer horizon contracts.

Another strand of literature focuses on exploring the relative importance of CDS spreads' determinants as the economy evolves over time and across markets. The general conclusion is that after controlling for firm-specific effects, market factors have a greater explanatory power, especially during periods of crises, as well as in post-crisis periods (Galil et al., 2014; Pereira et al., 2018; Fu et al., 2020) . Additionally to business cycles, Fu et al. (2020) show that the relative importance of explanatory variables (e.g., Tobin-Q coefficient, risk-free rates, stock market returns, and stock market volatility) of CDS spreads varies across US, Japan and UK markets. Aman (2019) finds that the long-run factor of the yield curve is a negatively significant determinant of the CDS premia, regardless of the sector and market states. The CDS spreads in the financial sector exhibit strong sensitivity to the yield rate in extreme market states.

In contrast to other natural disasters, COVID-19 is considered as a multi-period event that is almost synchronized within and across countries, with several health, social and economic implications for a prolonged period of time (Ludvigson et al., 2020) . Hence, it is not surprising to see a recent lively debate about the impact of COVID-19 on business survival and credit risk. Most studies document a significant negative impact on firms' liquidity across the globe due to loss of demand and increased uncertainty (Bosio et al., 2020; Didier et al., 2020) . The heterogenous sectoral reaction has been highlighted by Zechner et al. (2020) according to which, certain industries (e.g., high-tech) seem to be more resilient, while others (e.g., travel, tourism, and restaurant), appear to be hit harder by the pandemic. The financial markets reflect the level of resilience through the assets prices channels. Choi et al. (2020) also illustrate the different impact industries across regions (airlines, oil and gas, and consumer-facing industries) are suffering in North America and Europe, automotive in North America, apparel and shoes in Europe. The lack of government support has accelerated the number of business bankruptcies for low income-and lower-middle-income countries (Didier et al., 2020). Banerjee et al. (2020) pinpoint that further bankruptcies and unemployment are expected to increase after the withdrawal of government-aid schemes for business in advanced economies.

The study is also related to the literature concerning the impact of crisis on banking activities. Most firms are facing severe liquidity problems and their growth substantially depends on the extend of the access to bank credit during the COVID-19 pandemic. To be more specific, firms draw funds from their banks because of their current financial disruptions and anticipated declines in future cash flows induced by the pandemic (Li et al., 2020) . In contrast, banks also face important withdrawals of deposits, especially from households and SMEs, as a response to the rise of uncertainty, forcing them to reduce lending activities (Brei et al., 2020) .

The complex implications of COVID-19 pandemic for banking activities are analysed by Barua and Barua (2020), who emphasize that banks face interest rate, liquidity, credit and market risks because of the close linkages among firms, households and governments. It is believed that some firms will face substantial liquidity problems, reflecting onto the banking system later on, implying a major threat to banks' stability, as they are holding riskier assets, calling for imminent actions in support of the banking system. Moreover, in case of the low-rated banks, CDS spreads rise more than in the case of high-rated ones. The same also holds in the case of banks that are more dependent on short-term funding. Andrieş et al. (2020) investigate the impact of the lockdown measures on the European bank's CDS spreads. They provide evidence that both sovereign and bank CDS spreads increase due to lockdown measures, while the number of new confirmed cases seems to be the most important indicator that affects the perceived risk associated with bank investments. Banerjee et al. (2020) predict that in the case of advanced economies, a rise of about 20% of bankruptcies in 2021 vis-a-vis 2019, as well as a rise in unemployment are highly expected. Both negative effects could also be stimulated by the financial system in case that as the result of a low perceived quality credit, the access of firms to bank credit is restricted.

Our model assumes that variations in CDS spreads depend both on its own characteristics and on the environment within which it operates (e.g., Tang and Yan, 2007; Acharya and Johnson, 2007; Pires et al., 2011): (1) , = + 1 DCases + 2 ℎ + ⋅ , + . Du et al., 2013; Eyssell et al., 2013) , interest rates (Longstaff et al., 2004; Ericsson et al., 2009) , and stock market volatility (Gatev et al., 2006; Dotz, 2007) ], the analysis will be making use of the General Method of Moments (GMM) method. As a result, Equation

(1) can be written as:

The disturbance term has two orthogonal components, i.e. the fixed effects, v i and the idiosyncratic shocks, ε it . We assume that ε it are not serially correlated. The lagged dependant variable is still endogenous, since CDS i,t-1 term in CDS i,t-1 -CDS i,t-2 correlates with ε i,t-1 in ε it -ε i,t-1 . Therefore, the analysis needs to use instrumental variables to deal with the problem of endogeneity. From Equation (1)', natural candidates for the dependent variable are higher lags, such as CDS i,t−2 and so on. Z may also contain endogenous variables and, weakly and strictly exogenous variables. In our case, we have the following additional moment conditions, using weak exogenous variables:

From Equation (1)', it is possible to increase efficiency of the Arellano-Bond estimator through a great number of instruments. Arellano and Bover (1995) recommend the idea of a transformation of the system of equations, which favors the use of more information from observations. They develop an approach that transforms the instruments to make them exogenous to the fixed effects. They assume that changes in any instrumenting variable are uncorrelated with the fixed effects in Equation (1)'. From mathematical perspective, we get:

for all p and q.

. Thus, we have the following additional moment conditions:

The above expressions hold because we assume that ε it are not serially correlated. If Z is

This permits to derive an extended 'system' GMM estimator. System GMM estimator uses lagged differences of CDS it as instruments for equations in levels and lagged levels of CDS it as instruments for equations in first differences. The number of such instruments can be easily determined through certain criteria, such as the Akaike criterion, thus generating consistent and efficient parameters estimates.

Arellano and Bond suggest two specification tests to address consistency issues of the GMM estimators. First, the Sargan/Hansen test of over-identifying tests for joint validity of the instruments. The null hypothesis is that the instruments are not correlated with the residuals.

Second, the test for autocorrelation examines the hypothesis that the idiosyncratic disturbance ε it is not serially correlated (the full disturbance µ i + ε it is presumed autocorrelated since it contains fixed effects). To examine for autocorrelation aside from the fixed effects, the test is applied to the residuals in difference. We know that ε i,t -ε i,t-1 is mathematically related to ε i,t-1ε i,t-2 via the shared term ε i,t-1 . Thus, we expected a first-order serial correlation in differences.

To examine first-correlation in levels, the interest goes to the second-order correlation in differences, because we consider that this will detect correlation between the ε i,t-1 in (ε i,t -ε i,t-1 ) and the ε i,t-2 in (ε i,t-2 -ε i,t-3 ).

Finally, given that the moment conditions never hold exactly, the GMM weighting matrix has an impact on the parameter estimates. The estimation analysis uses a pre-specified weighting matrix in the GMM minimization process. The matrix describes a linear combination of the moments, and GMM is trying to minimize this linear combination where the moments are weighted in a flat mode (Newey and West, 1994; Bonnal and Renault, 2001 ).

Despite the literature has shown more firm-specific and macroeconomic factors played an important role in explaining the variations in CDS spreads, our model only embodies with only four variables can be measured daily as the main emphasis of our paper is to explore the impact of the COVID19 pandemic on the CDS spreads from February 2020 to September 2020.

Firstly, we incorporate stock market prices as a proxy to reflect firms' current performance and future prospects. We expect a negative relationship between stock prices and the CDS spreads (Zhang et al., 2005; Abid and Naifar, 2006; Galil et al., 2014) . Recall that CDS is an insurance policy on the default risk of a bond or loan. A higher (lower, respectively) stock market price signals that firms are performing well (worse, respectively), hence, the default risk associated with their debt payments is expected to be lower (higher, respectively 

In the first part of the empirical analysis, we make use of the cross-sectional dependence (CD) statistic by Pesaran (2004) to test the cross dependence of our panel. Under the null hypothesis of cross-sectional independence, the CD test statistic follows asymptotically a twotailed standard normal distribution. The results, reported in Table 3 , uniformly reject the null hypothesis of cross-section independence, providing evidence of cross-sectional dependence in the data, given the statistical significance of the CD statistics. We estimate Equation (1)' using the systems General Method of Moments (GMM) estimator set forth by Arellano and Bover (1995) and Blundell and Bond (1998) . The key merits of the GMM estimator approach is the allowance for firm-specific effects, which may reflect the differences across firms in their riskiness level. We also employ a cluster-robust estimator to allow for arbitrary heteroscedasticity and autocorrelation in the error term. Note that, crosssectional dependence (CD) is a common problem in cross-country panels with various cross sections (firms). To test for CD in the model residuals, we apply the Breusch-Pagan Chisquare, the Pearson LM Normal, and the Pearson CD Normal tests. The GMM results are reported in Table 5 . All three tests generate a p-value test of 0.00; hence, we can reject the null hypothesis of cross-sectional dependence and use the GMM approach to generate efficient estimators. In addition, to test for the validity of the instruments, we compute Hansen's Jstatistic and report the associated p-value in the tables.

The results conclude that there is a positive association between CDS and the metrics of pandemic event. More specifically, the US corporate CDS market seems to have been substantially affected by the presence of pandemic metrics that denote the presence of certain costs that can materialise through both supply and demand effects. For instance, the corporate section suffers from workers who may limit social interactions by reducing both labour supply and consumption. These results confirm the findings reported in the literature. For instance, Arnold et al. (2006) found these detrimental effects. Similarly, Fan et al. (2016) highlighted the importance of mortality costs and the reduction of the labour force because of pandemic events, thus, lowering returns to capital and slowing the pace of capital accumulation and GDP growth. Moreover, aggregate demand side effects could also negatively impact the corporate sector of the economy. Jordà et al (2020) showed that pandemics are usually followed by low natural interest rates for an extended period of time, due to higher precautionary saving and depressed investment opportunities. Therefore, the presence of the COVID-19 pandemic with millions of confirmed cases and deaths informs the corporate CDS markets that the economic loss can be substantially high and persistent, implying negative spillovers in firms' profitability and growth. Additional uncertainty is also transmitted to the sector when the most noteworthy characteristic of COVID-19 is its rapid transmission, resulting in large-scale containment policies, not only in the US, but also around the globe, leading to a global sudden stop in overall economic activity. The combination of a large fraction of global economic activity to a synchronised standstill and the presence of heightened financial market turbulence, amplified the impact of the pandemic shock, as financial markets came to grips with a historical global sudden stop. The presence of bankrupt firms, potentially disrupting supply chains, and unemployed workers, losing their skills, long-term relationships with firms, and lower labour productivity, seem just to have added to higher levels of uncertainty and higher CDS spreads.

Finally, relevant diagnostics are also reported in Table 5 . In particular, for the validity of the instruments, the results need to reject the test for second-order autocorrelation, AR(2), in the error variances. Moreover, they also need to reject the null hypothesis of difference-in-Hansen tests of the exogeneity of instruments. The regression has explicitly used 11 and 24

instruments across the two respective model specifications, where as instruments have been used two lags and one lead from the control variables that ensure the validity of the instrument tests. It is evident that both the test for AR(2) of disturbances and the difference-in-Hansen tests fail to reject the respective nulls. Thus, these tests support the validity of the instruments used, while difference-in-Hansen tests imply the exogeneity of the instruments employed. The table also reports the Hansen test for overidentifying restrictions. In the estimation process, eleven instruments have been used. As the number of instruments was by far lower than the number of observations, it did not create any identification problem, as reflected in the Hansen test. 

In this part of the analysis, we repeat the above estimates by making use of sectoral CDS Table 6 reports the number of firms in each sector. 

The COVID-19 pandemic enters its second consecutive year, businesses have been serious damaged due to the strict government measurements and anxiety from public. There has been a record number of businesses filed for bankruptcies across the world since the beginning of the pandemic. It is difficult to predict the amount and length of continued government support and the depth of the ongoing economic slowdown . It is projected that a rise of 20% of insolvencies are expected after the withdrawal of government-aid schemes in advanced economies (Didier et al., 2020; Banerjee et al., 2020) . For example, a recent survey of businesses found almost 15% of UK businesses are at-risk of permanently closing by the start of April 2021. 7 As expected, we find that during periods of volatility, businesses are more prone to bankruptcies as the variance of expected future returns from an investment would increase and access to external funds would diminish (Caglayan and Xu, 2018) .

This section examines whether the COVID-19 pandemic affects the CDSs through a particular transmission channel. More specifically, the channel is associated with the information transmitted from the COVID-19 metrics to the distressed characteristics of the 386 firms participating in the analysis. In other words, it will provide the first analysis of the capacity of the pandemic on individual firms to act as a signal of the firms' financial condition.

The literature has provided evidence of faster information processing ability in the CDS market associated with the financial condition of firms. Acharya and Johnson (2007) and Berndt and Ostrovnaya (2014) found that CDS markets reveal information in advance of the equity market for corporate entities that experience (or are likely to experience) adverse credit events.

Moreover, Subrahmanyan et al. (2014) , provided evidence that firms with a CDS market on their debt tend to default more. Relatedly, Colonnello et al. (2016) showed that firms with strong shareholders are more likely to default, because their creditors are more likely to purchase credit protection to avoid renegotiation. This part of the analysis also touches the literature that considers various aspects of the CDS-credit market nexus. Saretto and Tookes (2013) documented that CDS referenced firms have higher leverage ratios and borrow on longer debt maturities. Subrahmanyam et al. (2017) found that CDSs lead firms to increase their cash holdings to avoid funding constraints in times of distress.

Furthermore, this part is related to the literature that considers the role of natural disasters and pandemics on corporate default. There are two dominant views that attempt to explain how creditors interpret and respond to information about natural disasters, known chiefly as the debt default risk view and the creative destruction view. According to the former, the occurrence of natural disasters increases a firm's debt default risk due to increased operational risk (Elnahas et al., 2018) , which comes in the form of disruptions to both the supply and demand side of the local market. Furthermore, any disruption of the supply chain may result in the temporary or permanent termination of the buyer-supplier relationship (Wagner and Bode, 2006) . The latter view refers to the predicted positive economic effect of a natural disaster occurrence (Cuaresma et al., 2008) . Natural disasters cause more damage to country's (physical) capital stock, which is most likely made up in large part by outdated equipment, thus motivating the replacement of outdated (physical) destroyed capital stock with capital stock that embodies the latest technology, thereby inducing a positive productivity shock that leads to higher GDP growth rates (Okuyama et al., 2004) . Hence, natural disasters might provide economic opportunities for surviving firms who hope to replace outdated production technologies with new production technologies and/or shift from physical capital investments with a relatively lower return toward more human capital investments with a relatively higher return (Tanaka, 2015) .

Moreover, Queiroz et al. (2020 ), Ivanov (2020 and Sarkis et al. (2020) discuss that pandemic and epidemic events can seriously wreak havoc on supply chains, thus leading to extensive corporate defaults.

The set of the COVID-19 measures are used to investigate whether they can help in explaining firms' financial distress, i.e. likelihood of default. The analysis defines the measure of firm distress and firm failure employed. This measure of financial distress captures the point at which a firm is downgraded by a major rating agency (Fitch, Moody's or Standard and Poor's). The measure defined is binary, taking a value of one when a firm is categorised as failed or downgraded, and a value of zero, otherwise. From a total of 386 US firms, we have 65 firms which failed over the time span under consideration. To explore the explanatory power of the COVID-19 measures for firm failure, the analysis follows Shumway (2001) and Chava and Jarrow (2004) and estimates the probabilities of failure using a logit model. In particular, it assumes that the marginal probability of failure follows a logistic distribution given by:

(2) , ( , + 1 ) =

where is the probability at time that firm will fail, is a dummy variable taking on , , + 1 the value of 1 (0, respectively) if the firm failed (did not fail, respectively) in period , + 1 is the vector of five explanatory variables (i.e., COVID-19 measures and equity returns) at the end of period , and represent the constant and slope parameters characterizing the logistic function, respectively. The models are estimated via maximum likelihood. Note that a higher value of indicates a higher probability of failure. The model is formed such that only + , information available at time t is employed to understand the probability of failure of an individual firm during the subsequent period . Finally, to avoid potential endogeneity + 1 problems, the regression is estimated through Instrumental Variables (IV) Logit method (Chintagunta, 2001; Shugan, 2004) .

The analysis uses two modelling specifications for Equation (2). Specification/column (1) includes only the COVID-19 measures as the explanatory variables, while specification/column (2) includes these four COVID-19 metrics, as well as the control variables from Equation (1)', i.e. US stock market volatility, risk-free interest rates, and stock returns of the underlying firms. To instrument the probability of default, we have considered CDS spreads as potential instruments. The results in Table 8 report a highly significant positive coefficient for all four COVID-19 measures across both specifications. The estimates are significant at the 5% significance level, confirming strong evidence that pandemic measures impound additional information for equity markets. The observed pseudo R-squared values are found to be substantially high, recommending that logit models would have had high explanatory power during the pandemic crisis. Moreover, to test whether the instrument is correlated with the predictor, the analysis conducts first-stage robust F-statistic test. A significant robust F-statistic, which is higher than a rule of thumb value of 10, indicates that the instrument is not weakly correlated with the predictor (Cameron and Trivedi, 2010 is not jointly correlated with the outcome variable, we provide estimates of the Sargan (Sargan, 1958) . The estimates indicate non-significant χ 2 , implying that the instrument is not jointly correlated with the outcome variable.

In addition to logit regression estimations, we also employ a Cox proportional hazards model as an alternative method to analyse the impact of the fitted pandemic CDS spreads for firm failure. The dependent variable is the time to default, with Table 9 containing the estimation results (for the same specifications as in Table 8 ). Both modelling specifications confirm that these pandemic CDS spreads are significantly associated with default, in line with the results obtained from the logit regressions. 

In light of the results reported, this step of the empirical analysis evaluates the (insample) forecasting performance of Equation (1) metrics, than that without the COVID-19 metrics in both panels. It is also imperative to assess whether any reductions in the two measures are statistically significant. Thus, the Diebold and Mariano (1995) test of equal forecast accuracy is presented. The results, also reported in Table   10 , reject the null hypothesis of forecasting accuracy across both cases of the COVID-19 definitions. 

Based on a reviewer's suggestion, a statistically significant parameter does not mean that the independent variable is economically linked to the dependent variable, especially when using time series for short periods. To this end, this part of the empirical analysis does run a panel Granger causality test. In particular, to show causation this part of the analysis makes use of the Dumitrescu and Hurlin (2012) Granger non-causality test, which is suitable for heterogeneous panelized data frameworks which consider individual unit fixed effects and is based on a bootstrap procedure that takes care of cross-sectional dependence challenges. It takes into account the following methodological framework:

with all the variables being defined as in Equation (1) 

where W i,T illustrates the individual Wald statistics in relevance to the ith cross-section unit to match the individual null hypothesis. Table 11 presents the findings of the Granger causality test outcomes generated through a bootstrap application technique. The findings clearly document that across all COVID-19 metrics, the null hypothesis of non-causality is rejected, while the reverse does not hold, indicating the presence of univariate causality running from the COVID-19 metrics to CDS spreads. 

The COVID-19 crisis has created an enormous uncertainty shock with magnitude similar to the Great Depression of 1929-1933. This massive increase in uncertainty has a profound adverse effect on business profitability, mew business formation, R&D, investments, and other factors that affect productivity over the medium and long-term (Baker et al., 2020) . Given CDS spreads act as an important indicator of firm's credit risk, there is a surge on researchers'

intensified effort on gauging the market's ability to correctly price this risk induced by COVID- To be more specific, banking, travel & leisure, transportation, airlines and restaurants were the worst affected sectors, while media, technology, telecommunications, pharmaceutical, information, and data technology firms were spared from the pandemic. Furthermore, the results confirmed three pandemic effects transmission channels that contributed to larger CDS spreads. The analysis also found that COVID-19 measures increased the probability of firms' financial distress that captured the point at which a firm had been downgraded by a major rating agency. In addition, a forecasting exercise clearly indicated the forecasting superiority of the modelling approach that explicitly introduced the COVID-19 metrics, potentially indicating the substantial role of uncertainty associated with the pandemic crisis for correctly pricing CDS spreads.

In sum, the findings provide useful information for macroeconomic policymaking, consumption, capital investments, management risk, and portfolio management. The impact of COVID-19 crisis is large, it requires central banks to carry out aggressive policies to maintain the financial stability and to minimize the number of business failures. By identifying the heterogeneity of industry sensitivities to the pandemic implies that some sectors can provide a channel for diversification during further waves of pandemic cases and fatalities. For example, if there is an anticipated increase cases and deaths, investors may use this information to devise their investment strategies, such as taking short positions in COVID-sensitive sectors (e.g., banking, transportation, airlines, restaurants), and long positions for firms in insensitive sectors (e.g., technology, telecommunications, pharmaceutical, information and data technology). One interesting (future) question will be to examine the impact of COVID-19 on other heavily affected countries.

Barua, B., Barua, S., 2020. COVID-19 implications for banks: the case of an emerging economy with a weak financial system. Available at SSRN:

• The paper shows that US and global COVID-19 metrics positively affect CDS spreads

• There is an heterogenous effects across sectors

• Banking, travel, transportation, airlines, restaurants are the hardest hit sectors

• It also shows that COVID-19 increases corporate CDS spreads via distress mechanisms

(1) , = + 1 DCases + 2 ℎ + ⋅ , + . + + ,

The determinants of credit default swap rates: an explanatory study

Insider trading in credit derivatives

Animal Spirits

Are CDS spreads sensitive to the term structure of the yield curve? A sectorwise analysis under various market conditions

The effects of the COVID-19 pandemic through the lens of the CDS spreads

Another look at the instrumental variable: estimation of errorcomponent models

A potential influenza pandemic: possible macroeconomic effects and policy issues

Ambiguity, volatility, and credit risk

Bankruptcies, unemployment and reallocation from COVID-19

The impact of uncertainty on financial institutions

Macroeconomic uncertainty and credit default swap spreads

Agency costs, collateral, and business cycle fluctuations

Do equity markets favor credit market news over options market news?

Initial conditions and moment restrictions in dynamic panel data models

Minimal chi-square estimation with conditional moment restrictions

It's what you say and what you buy: a holistic evaluation of the corporate credit facilities

Economic policy uncertainty effects on credit and stability of financial institutions

Inflation volatility effects on the allocation of bank loans

Sentiment volatility and bank lending behavior

Microeconometrics Using Stata. College Station

Bankruptcy prediction with industry effects

The determinants of bank CDS spreads: evidence from the financial crisis

Endogeneity and heterogeneity in a probit demand model: estimation using aggregate data

The determinants of credit spread changes

Empty creditors and strong shareholders

Credit and liquidity components of corporate CDS spreads

A simple way to estimate bid-ask spreads from daily high and low prices

Natural disasters as creative destruction? Evidence from developing countries

Are banks too big to fail or too big to save? International evidence from equity prices and CDS spreads

Comparing predictive accuracy

Pricing credit default swaps with observable covariates

Time-varying contributions by the corporate bond and CDS market to credit risk price discovery

Buffering COVID-19 lossesthe role of prudential policy

The effect of credit default swaps on the pricing of audit services

Testing for Granger non-causality in heterogeneous panels

Natural disasters and corporate leverage

The determinants of credit default swap premia

Determinants and price discovery of China sovereign credit default swaps

The inclusive cost of pandemic influenza risk

Credit default swap spreads: market conditions, firm performance, and the impact of the 2007-2009 financial crisis

The determinants of CDS spreads

Pairs trading: performance of a relative-value arbitrage rule

The impact of COVID-19 and government intervention on swaps market liquidity

Predicting the impacts of epidemic outbreaks on global supply chains: a simulation-based analysis on the coronavirus outbreak

World economic outlook

Longer-run economic consequences of pandemics

Evaluating sovereign risk spillovers on domestic banks during the European debt crisis

Macroeconomic conditions and credit default swap spread changes

Debt rollover risk, credit default swap spread and stock returns: evidence from the COVID-19 crisis

Corporate yield spreads: default risk or liquidity? New evidence from the credit default swap market

Automatic lag selection in covariance matrix estimation

Momentum, nonlinearity in cointegration and price discovery: evidence from sovereign CDS and equity markets of emerging countries

Measuring the economic impact of disasters: interregional input-output analysis using sequential interindustry model

Spillovers of COVID-19: the impact on the global economy

What drives corporate CDS spreads? A comparison across US, UK and EU firms

General diagnostic tests for cross section dependence in panels. Cambridge Working Papers in Economics, No. 435 and CESifo Working Paper

A simple panel unit root test in the presence of cross -section dependence

The complete picture of credit default swap spreadsa quantile regression approach

Impacts of epidemic outbreaks on supply chains: mapping a research agenda amid the COVID-19 pandemic through a structured literature review

Corporate leverage, debt maturity, and credit default swaps: the role of credit supply

A brave new world: lessons from the COVID19 pandemic for transitioning to sustainable supply and production. Resources, Conservation and Recycling

The estimation of economic relationships using instrumental variables

Endogeneity in marketing decision models

Forecasting bankruptcy more accurately: a simple hazard model

Credit default swaps, exacting creditors, and corporate liquidity management

Does the tail wag the dog? The effect of credit default swaps on credit risk

The impact of natural disaster on plant's growth: evidence from the Great Hanshin-Awaji (Kobe) earthquake

Liquidity and credit default swap spread

Bankruptcy and the COVID-19 crisis

be informed that in relevance to our submission entitled: CDS spreads and COVID-19 pandemic to be potentially considered for publication by the Journal of International Financial Markets

Nicholas Apergis = Conceptualization; Data curation; Formal analysis

Project administration; Software; Supervision; Validation; Roles/Writing -original draft

Dan Danuletiu = Investigation; Resources; Software; Validation; Roles/Writingoriginal draft

Roles/Writing -original draft; Writing -review & editing