The impact of the financial crisis of 2008 on U.S. firms’ short- and long-term debt

The impact of the financial crisis of 2008 on

U.S. firms’ short- and long-term debt

Janet van Rijs S3537684

j.a.van.rijs@student.rug.nl

University of Groningen Faculty of Economics and Business

MSc Finance

Supervisor: Dr. E. Karmaziene June 12, 2019

Abstract

This study examines the impact of the financial crisis of 2008 on U.S. firms’ short- and long-term debt. Previous literature finds evidence for both an increase and decrease in short- and term debt during periods of financial crisis. However, firms would like to replace long-term debt with short-long-term debt, since short-long-term debt is cheaper during the crisis. I find that firms attract more term debt during the financial crisis. Firms with a high level of short-term debt in the pre-crisis period increase their short-short-term debt level even more during the financial crisis, which is more risky for the company. In addition, I find that firms decrease their long-term debt level during the financial crisis. The decrease is larger for firms with a high level of long-term debt in the pre-crisis period. The results are robust to different estimation methods and alternative specifications.

1 1. Introduction

The global financial crisis of 2008 was the biggest financial crisis since the Great Depression in the 1930s. Many financial institutions collapsed, while others survived because they were nationalized or bailed out by the government. But what about non-financial firms? They did not receive state support, but are affected by the crisis equally well. A well-known effect of the crisis is the reduced amount of debt and equity financing available to non-financial firms. During the crisis, when risk rises and business prospects become more uncertain, firms may be reluctant to issue equity, as stock prices reached their ultimate low. And what about debt financing? As a consequence of the bank failures, many financial institutions reduced their lending to non-financial firms. However, empirical evidence indicates that leverage has increased at the onset of the crisis (Demirgüç-Kunt, Martinez-Peria, and Tressel, 2015; Fosberg, 2012; Iqbal and Kume, 2014), but it is not clear whether this is due to a decrease in equity or to an increase in (short- or long-term) debt. This provides an interesting opportunity to investigate how firms’ short- and long-term debt evolved after the onset of the financial crisis of 2008. The distinction between short- and long-term debt is important for firms’ financial stability, as the crisis has strengthened the vulnerability of firms with a strong maturity mismatch – those who invest long-term and finance themselves short-term. Furthermore, a decline in the maturity of debt shifts rollover risk from the lender to the firm, which may have a negative impact on firm growth and long-term investments.

2 “What is the relationship between the financial crisis of 2008 and non-financial, U.S. firms’

short- and long-term debt?”

To analyze the effect of the financial crisis on U.S. firms’ short- and long-term debt level I use different measures of short- and long-term debt as dependent variable. I regress crisis both on short- and long-term debt, as theory suggests that the financial crisis may have a different impact on short- and long-term debt. The model controls for tangibility, profitability, firm size, growth opportunities, uniqueness, and asset maturity. The main difference with previous literature is that I use lagged values of short- and long-term debt, total debt and total assets in the denominator. The lagged value refers to the value of previous year. This avoids the problem that, for example, an increase in leverage may be due to either a decrease in equity or to an increase in debt. Moreover, this study focuses on firms that use already debt financing before the financial crisis. Firms which do not have any debt on their balance sheet before the crisis may already be constrained in obtaining debt financing, or may not need any debt financing. For robustness, different estimation methods and alternative specifications are used. In addition, I check whether the impact of the financial crisis is different among firms with high and low levels of short- and long-term debt.

I find evidence for a positive relationship between the financial crisis and the use of short-term debt, and a negative relationship between the financial crisis and the use of long-short-term debt. The results indicate that firms have a higher short-term debt level during the financial crisis of 2008, which is more risky for the company, and a lower long-term debt level. Firms, which have already a high level of short-term debt in the pre-crisis period, increase their level of short-term debt even more during the crisis. Firms with a high level of long-term debt in the pre-crisis period decrease their long-term debt borrowing more during the financial crisis compared to firms with a low level of long-term debt in the pre-crisis. The results are robust to different estimation methods and alternative specifications.

3 2. Hypotheses review

The major contribution of this study is to provide an insight about how U.S. firms’ short- and long-term debt evolved after the onset of the financial crisis of 2008. Modigliani and Miller (1958) argue that in perfect capital markets, firm value is independent of how a firm is financed. They state that it does not matter in which market (short-term or long-term) firms obtain financing. However, in the real world, various imperfections can lead to market segmentation, with different debt markets attracting different types of firms, with different amounts and different terms of financing. In such a scenario (during periods of crisis), some firms will change the amount and type of financing, by moving across short- and long-term debt.

In the “maturity rat race” capital structure model of Brunnermeier & Oehmke (2013), periods of crisis increase firms’ incentive to shorten the maturity of debt, despite the costs associated with switching from long-term debt to short-term debt (rollover costs). They argue that when a borrower raises financing for a long-term project from multiple creditors at different maturities, short-term financing is preferred. Shorter maturities provide seniority, which dilutes the pay-offs to long-term lenders. This, in turn, causes all other lenders to shorten their maturity too, resulting in excessively short-term financing. They also show that, if firms value financial flexibility during periods of crisis, they are less likely to enter into long-term contracts with covenants. Covenants impose restrictions on further lending or require monitoring for example, which will decrease the demand for long-term debt.

Milbradt and Oehmke (2014) develop the short-termism spiral, which is especially true during periods of crisis. They argue that periods of crisis increase financing restrictions, such that financing for long-term projects is more expensive. Firms which cannot finance their first-best long-term project, respond by adopting a second-best shorter-term project, for which financing is available. The shortening of maturity is unobservable to lenders, but worsens the pool of borrowers, as they adopt their second-best project. This affects the financing terms of debt contracts offered by lenders, who now face a worse pool of borrowers. This process repeats, leading to a short-termism spiral: Financing terms for firms that could first obtain financing for long-term projects worsen and, a number of these firms are now constrained in obtaining long-term financing, forcing them to adopt their second-best project and shorten maturity too.

4 holders with insufficient returns (Myers, 1977). During periods of crisis, short-term debt becomes more attractive to borrowers, since the value of short-term debt is less sensitive to future investment opportunities than the value of long-term debt.

Empirical evidence about the evolution of non-financial firms’ short-term debt during periods of crisis is scarce. Chen et al. (2018) find that debt maturity drops during periods of crisis and show that short-term debt (measured by the ratio of short-term debt over total debt) increases between 2008 and 2009 from 34 to 44 percent. Their sample covers countries in emerging markets and developing economies (EMDEs) and advanced economies (AEs), and consists mainly of small, medium enterprises (SMEs). Based on this empirical evidence, “the maturity rat race”, the short-termism spiral, and the underinvestment problem, I expect an increase in short-term debt during periods of crisis, and a decrease in long-term debt.

H1: Short-term debt will increase during the financial crisis of 2008. H2: Long-term debt will decrease during the financial crisis of 2008.

Besides the above mentioned theories, another reason why firms might replace long-term debt with short-term debt is because short-term debt is cheaper than long-term debt during periods of crisis. Short-term debt is cheaper because lenders prefer liquid and safe assets, as lenders’ risk aversion increases during periods of crisis. This makes long-term lending less attractive during periods of crisis, as long-term debt is less liquid. So, the shorter the borrowing period, the lower the interest rate. Therefore, because short-term debt decreases firms’ interest expenses, firms may choose to use more short-term debt, although this increases their exposure to rollover risk, and in turn leads to higher expected bankruptcy costs (Chen et al, 2018; Krishnamurthy, 2010).

In addition, Diamond (1991) and Flannery (1986) argue that information asymmetries might increase the use of short-term debt. When firms use short-term debt, they signal that they have favorable information about their business prospects, which can in turn lead to lower interest expenses. This may also force other firms with more unfavorable information about their business prospects to use short-term debt, in order to avoid the stigma associated with long-term debt.

5 investment opportunities during the crisis. Unfortunately, both papers do not make a distinction between short- and long-term investments, so this is not an useful explanation for the replacement of long-term debt by short-term debt.

As stated above, an increase in short-term debt leads to an increase in rollover risk. Therefore, Diamond and He (2014) argue that firms will increase long-term debt during periods of crisis, because the rollover costs of short-term debt increase. A decline in the maturity of debt shifts rollover risk from the lender to the firm. Hence, too much short-term debt is very risky, as firms can experience higher financial constraints during periods of crisis. Long-term debt on the other hand reduces rollover risk during the crisis, and allows firms to finance large (long-term) investments.

Where few studies consider the impact of the financial crisis on a firms’ short- and long-term debt level with the balance sheet data, more empirical studies consider the impact of the financial crisis on leverage with balance sheet data. All these studies have in common that they use three periods to investigate the effect of the crisis, namely pre-crisis, crisis, and post-crisis. This study simply focuses on two periods to investigate the effect of the crisis: pre-crisis and crisis. Fosberg (2012) and Iqbal and Kume (2014) find that in the United States, United Kingdom, and Germany, leverage for publicly listed firms increases in the crisis period, but decreases in the post-crisis period back to the pre-crisis level. The increase in leverage is not significant for France (Iqbal and Kume, 2014). They measure leverage as long-term debt to total assets and total debt to total assets, respectively. Demirgüç-Kunt, Martinez-Peria, and Tressel (2015) note that leverage ratios for firms listed on a stock exchange increase at the onset of the crisis, while leverage for private firms and SMEs decreases. Their sample includes firms across 79 countries. They measure leverage as total debt to total assets and long-term debt to total assets.

6 United Kingdom during 2007 and 2008, but decreases after 2008 (Kahle and Stulz, 2013; Pattani, Vera, and Wackett, 2011). This is contradictory to what “the maturity rat race” and the short-termism spiral predict, as long-term debt increases.

Based on the existing literature, it is clear that the crisis has an effect on firms’ short- and long-term debt level. There are theories for both an increase and decrease in short- and long-long-term debt during periods of crisis. However, empirical evidence indicates that both short- and long-term debt increase during periods of crisis (Adrian, Colla, and Shin, 2013; Becker and Ivashina, 2014; Chen et al., 2018; Cortina and Didier, 2018). The results of this study help to reconcile the mixed findings about the interaction between short- and long-term debt during periods of financial crisis.

3. Methodology

To assess how U.S. firms’ short- and long-term debt evolved after the onset of the financial crisis of 2008, I will use the following model:

𝐷𝑒𝑏𝑡_𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑖,𝑡 = α + 𝛽1𝐶𝑟𝑖𝑠𝑖𝑠 + 𝛽2𝐹𝑖𝑟𝑚𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡−1+ 𝑓𝑖+ 𝜀𝑖,𝑡 (1)

Where 𝐷𝑒𝑏𝑡_𝑀𝑒𝑎𝑠𝑢𝑟𝑒_𝑖,𝑡 refers to different measures of short- and long-term debt by firm 𝑖 during year 𝑡. The variable of interest is the dummy variable 𝐶𝑟𝑖𝑠𝑖𝑠 and 𝐹𝑖𝑟𝑚_𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_{𝑖,𝑡−1} represent a set of time-varying firm characteristics by firm 𝑖 lagged with one year (e.g., 𝑇𝑎𝑛𝑔𝑖𝑏𝑖𝑙𝑖𝑡𝑦, 𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦, 𝑆𝑖𝑧𝑒, 𝑀𝑎𝑟𝑘𝑒𝑡_𝑡𝑜_𝐵𝑜𝑜𝑘, 𝑈𝑛𝑖𝑞𝑢𝑒𝑛𝑒𝑠𝑠, and 𝐴𝑠𝑠𝑒𝑡_𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦). The error term is denoted as 𝜀𝑖,𝑡. The term 𝑓𝑖 relates to firm fixed effects (see model

specification), and is only relevant when using a fixed effects model. When using OLS, 𝑓𝑖 is not

relevant.

A. Measuring the dependent variable

Previous literature does not provide one universal definition of the maturity of long-term debt. The maturity of long-term debt is typically defined as debt due after 1 year (Scherr and Hulbert, 2001), 3 years (Barclay and Smith, 1995) or 5 years (Ozkan, 2002). Short-term debt is typically defined as debt due within 1 year (Chen et al., 2018). As the maturity of short-term debt is clear, and it is not desirable to have a gap in between the maturity of short- and long-term debt, I follow Scherr and Hulbert (2001) and use debt due after 1 year as the definition of long-term debt.

I want to evaluate whether the level of short- and long-term debt for US’ companies is different during the crisis. To construct 𝐷𝑒𝑏𝑡_𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑖,𝑡, I use different measures of short-

7 Martinez-Peria, and Tressel, 2015; Fosberg, 2012; Iqbal and Kume, 2014). However, an increase in leverage does not provide a clear picture of how a firms’ capital structure changed. An increase in leverage can be either due to a decrease in equity or to an increase in debt. To avoid this problem, I use the lagged value of short- or long-term debt, total debt or total assets in the denominator. The lagged value refers to the value of previous year. In the numerator I use the current level of short- or long-term debt, so the value of this year. In total, I use six different dependent variables, three for short-term debt and three for long-term debt. Short-term debt is calculated as the ratio of short-Short-term debt over lagged short-Short-term debt (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), and short-term debt over lagged total assets (STD/L.TA). Long-term debt is calculated as the ratio of long-term debt over lagged long-term debt (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Specifically, STD/L.STD and LTD/L.LTD measure whether short- or long-term debt is higher or lower than last year. When this measure is above 1, debt is higher than last year. When it is below 1, debt is lower than last year. STD/L.TD and LTD/L.TD measure the maturity composition of debt. STD/L.TA and LTD/L.TA represent the extent to which firms finance their assets with short- and long-term debt, respectively. By using three different measures for both short- and long-term debt, I can see if the results hold under all the three specifications, and if the crisis really has an impact on a firms’ short- and long-term debt level. For robustness, I also check whether the impact of the financial crisis is different among firms with high and low levels of short- and long-term debt. The sample is split according to the median value of the relevant dependent variable in the pre-crisis period. This implies that half of the companies has low levels of short- or long-term debt, and the other half has high levels of short- or long-long-term debt. The crisis may have a different effect on companies with high or low levels of debt in their capital structure. For the construction of the dependent variables, I use the book value of debt and assets rather than the market value, because debt is more related to the assets of a company rather than its growth potential (Myers, 1977). Book values are also preferred because financial markets fluctuate a lot, and managers do not rebalance their capital structure in response to these market fluctuations (Graham and Harvey, 2001). Moreover, Compustat does not provide market values of debt.

8 during periods of crisis, I follow Rajan and Zingales (1995) and define short-term debt as the total amount of short-term borrowings and the current portion of long-term debt that is due within 1 year. Short-term debt typically includes loans payable, notes payable, and bank acceptances and overdrafts. I define long-term debt as the total amount of long-term debt that is due after 1 year. Long-term debt typically includes bonds, mortgages, obligations that require interest payments, and capitalized lease obligations. It might be that these measures overstate short- and long-term debt, resulting in a measurement error problem, as the definition may contain some non-financing items. For example, short-term debt also includes sinking fund payments, which are not used for financing. This introduces a potential bias if the use of this non-financing item is correlated with the independent variables. For example, larger firms might have more sinking fund provisions on their balance sheet. However, data limitations force me to use this broader definition of short- and long-term debt, as Compustat does not provide detailed information about a firms’ short- and long-term debt, but only the total amount of debt at the balance sheet. Furthermore, an explanation for an increase in short-term debt and a decrease in term debt could be that over the crisis period long-term debt turned into short and no additional long-long-term debt was given. Compustat provides the current portion of long-term debt that falls due within one year. However, Compustat does not provide details about debt issuance. Therefore, it is difficult to examine whether no additional debt is given.

In presenting my results, I focus on firms that use already debt financing before the financial crisis. Firms that do not have any debt on their balance sheet before the crisis may already be constrained in obtaining debt financing, or may not need any debt financing. However, for robustness, I also check whether the results hold when I include those firms.

B. Measuring crisis

9 period contains the years 2004 till 2007, and the crisis period contains the years 2008 till 2010. As I use lagged values in the denominator of the dependent variables, the pre-crisis period and the crisis period contain each three years. The pre-crisis period contains short- or long-term debt in 2005 relative to 2004, 2006 relative to 2005, and 2007 relative to 2006. The crisis period contains short- or long-term debt in 2008 relative to 2007, 2009 relative to 2008, and 2010 relative to 2009. This means that the start of the financial crisis in 2007 (relative to 2006) is in the pre-crisis period, as 2006 is still in the pre-crisis period. Another reason to include 2007 (relative to 2006) in the pre-crisis period is based on the fall of the U.S. stock market. The Dow Jones Industrial Average (DJIA) peaked in October 2007, fell rapidly during 2008, and reached its ultimate low in March 2009. Based on the DJIA, the biggest part of 2007 cannot be counted as crisis, so I include it in the pre-crisis period. I define 𝐶𝑅𝐼𝑆𝐼𝑆 as a dummy variable that equals 1 if the firm is in a crisis period and 0 if the firm is in the pre-crisis period. For robustness, I also check if the results hold if the crisis period is defined as 𝑡 = (+1, +2), which is in line with the studies of Demirgüç-Kunt, Martinez-Peria, and Tressel (2015) and Iqbal and Kume (2014).

C. Measuring the control variables

Regardless of the effect of the crisis, a firms’ debt level is influenced by a lot of other (firm-specific) factors. Where the determinants of capital structure are intensively researched, fewer studies consider the determinants of corporate debt maturity. Moreover, most studies differ in which factors they find significant, and in which direction they move. Because the results differ so much between the different studies, I focus on the most recurring ones, namely: 𝑇𝑎𝑛𝑔𝑖𝑏𝑖𝑙𝑖𝑡𝑦, 𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦, 𝑆𝑖𝑧𝑒, 𝑀𝑎𝑟𝑘𝑒𝑡_𝑡𝑜_𝐵𝑜𝑜𝑘, 𝑈𝑛𝑖𝑞𝑢𝑒𝑛𝑒𝑠𝑠, and 𝐴𝑠𝑠𝑒𝑡_𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦.

10 to have a negative impact on a firm’s short-term debt level, since smaller firms tend to use more short-term debt (Titman and Wessels, 1988). 𝑆𝑖𝑧𝑒 is measured as the natural logarithm of total assets (Frank and Goyal, 2009).

The market-to-book ratio typically serves as a proxy for the firms’ growth opportunities. Myers’ (1977) underinvestment problem predicts that firms with more growth opportunities (higher market-to-book ratio) use less long-term debt. The underinvestment problem could be mitigated by using more short-term debt, as this reduces the potential for underinvestment. 𝑀𝑎𝑟𝑘𝑒𝑡_𝑡𝑜_𝐵𝑜𝑜𝑘 is typically defined as the ratio of the market value of assets over the book value of assets (Barclay and Smith, 1995; Frank and Goyal, 2009; Rajan and Zingales, 1995; Stohs and Mauer, 1996). The market value of assets equals book value of assets plus market value of equity minus book value of equity, where market value of equity equals the stock price at the end of year 𝑡 multiplied by the total number of shares outstanding.

𝑈𝑛𝑖𝑞𝑢𝑒𝑛𝑒𝑠𝑠 is measured as the ratio of R&D expenses over total assets (Titman and Wessels, 1988). Debt levels are negatively related to the uniqueness of a firm, since customers, employees, and suppliers of firms which produce unique products suffer higher costs in the event that these firms go bankrupt (Titman and Wessels, 1988).

A common prescription in the finance literature is that a firm should match the maturity of its assets to the maturity of its liabilities. When debt maturity is shorter than asset maturity, firms may not have sufficient cash to repay the debt when it is due. On the other hand, when debt maturity is longer than asset maturity, firms may have to repay the debt while the cash flow from assets stop (Stohs and Mauer, 1996). 𝐴𝑠𝑠𝑒𝑡_𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦 is measured as a book-value weighted sum of the maturity of long-term assets and the maturity of current assets(Alcock, Finn, Keng Tan, 2012; Ozkan, 2002; Stohs and Mauer, 1996). The maturity of long-term assets is measured as net PP&E divided by depreciation expense, and the maturity of current assets is measured as current assets divided by operating expenses. Total asset maturity is the weighted sum of these measures where net PP&E over total assets is the weight for long-term assets, and current assets over total assets is the weight for current assets.

11 debt. However, it is unlikely to assume that the lagged value of tangibility is subject to the same problem, and therefore it serves to mitigate simultaneity concerns.

D. Model specification

To determine the impact of the financial crisis on the level of short- and long-term debt, I make use of panel data. Panel data contain observations on multiple firms where each firm is observed over time. As a starting point, the study employs t-tests for differences in means (assuming unequal variances) to identify if the means of the different short- and long-term debt measures are significantly different from each other during the pre-crisis and the crisis period.

To test the impact of the financial crisis, previous studies use a pooled ordinary least squares (OLS) model (Fosberg, 2012) or a fixed effects model (Iqbal and Kume, 2014). To test the two hypotheses, I use pooled OLS and a fixed effects model, to see if the results hold under both specifications. For robustness checks, I only use a fixed effects model. Pooled OLS and fixed effects regressions exploit both the cross-sectional and time-series variation in the data. The latter in particular is important for my regression, as I want to study if short- and long-term debt levels vary during the crisis. However, the OLS assumption of uncorrelated errors with the independent variables is unlikely to be satisfied in this regression. The fixed effects model serves as a method to deal with the problem of correlated errors by demeaning the variables using a within transformation. The within transformation subtracts the time-mean of each firm away from the values of the variables, obtaining a regression which contains demeaned values only (Brooks, 2014). The within transformation is equivalent to adding a dummy variable for each firm in the sample. Thus, although the fixed effects model preserves the time-series dispersion in the sample, it ignores most of the information from differences across firms, so variables which do not change over time will cancel out. This mitigates the omitted variable bias in the analysis.

Furthermore, the Hausman test in Stata can be used to determine if the fixed effects model is preferred. There are two classes of panel data models: the fixed effects model and the random effects model. The Hausman test is performed to decide which model is most appropriate. The null hypothesis of the Hausman test states that the error term is uncorrelated with the independent variables, which implies the use of the random effects model. If the null hypothesis is rejected a fixed effects model is preferred. After performing the Hausman test, the results show that, with p-values between 0.00 and 0.01, the null hypothesis is rejected, and the use of the fixed effects model is confirmed. Appendix A reports the results of the Hausman test for STD/L.STD, since this model has the highest p-value.

12 A White test is performed to test if heteroskedasticity exists in the data. When heteroskedasticity exists, the error terms do not have a constant variance. The null hypothesis of the White test states that the error terms have a constant variance. For all models, the probability of the White test is 0.00, which means that heteroskedasticity is a problem. To correct this problem, White’s heteroskedasticity consistent standard error is used.

In addition, a Durbin-Watson test is performed to test for autocorrelation. When autocorrelation exists, there is a pattern in the error terms. The null hypothesis of the Durbin-Watson test states that there is no first order autocorrelation. The Durbin-Durbin-Watson test reports a test statistic, where 2 means no autocorrelation, 0 to <2 means positive autocorrelation, and >2 to 4 means negative autocorrelation. However, the Durbin-Watson test does not follow a standard statistical distribution. Durbin and Watson established two critical values, the lower limit and the upper limit, which depend on the sample size and the number of independent variables. If the test-statistic lies between the two critical values, in my case, 1.70 and 1.84, the test is inconclusive, and the null hypothesis can neither be rejected nor not rejected (Brooks, 2014). However, for all models, the test statistic lies between 0.56 and 1.58, which means that there is positive autocorrelation. Positive autocorrelation means that OLS and fixed effects still give unbiased, consistent coefficients, but the standard errors are biased downwards (too small), which means that any inferences can be misleading. I am aware that this is an severe limitation of my research.

13

Table 1

Correlation Coefficients between the Independent Variables

(1) (2) (3) (4) (5) (6) (7) (1) Crisis 1.00 (2) Tangibility 0.02 1.00 (3) Profitability -0.02 0.02 1.00 (4) Size 0.06 0.24 0.48 1.00 (5) Market-to-book -0.04 -0.02 -0.68 -0.36 1.00 (6) Uniqueness 0.03 -0.13 -0.66 -0.42 0.41 1.00 (7) Asset maturity 0.03 0.61 0.09 0.16 -0.06 -0.15 1.00 4. Data

The data for this study is obtained from Compustat, acquired from the Wharton Research Data Service database. The sample covers the years from 2004 through 2010. For this period, annual firm-specific data is gathered for publicly listed, U.S. firms. Financial firms with SIC codes between 6000 and 6999 are excluded, because these firms have a different capital structure in comparison to non-financial firms. In addition, financial firms are often subject to regulatory capital requirements, which makes these companies not comparable to non-financial firms (Rajan and Zingales, 1995). Furthermore, firms that do not have any short- or long-term debt on their balance sheet in the pre-crisis period are excluded from the sample, as these firms may already be constrained in obtaining debt financing, or may not need any debt financing. All the ratio variables are winsorized at the 1% level in both tails of the distribution to minimize some of the extreme values. Winsorizing is the procedure of replacing extreme values with the value of the observation at the cutoff, and is better than dropping observations. The final sample, after exclusions, consists of 8,775 firms and has 43,582 observations in total. Appendix B lists the number of observations per industry and their corresponding SIC code.

Descriptive statistics

14

Table 2

Descriptive Statistics for Publicly Listed, Non-financial Firms in the U.S., 2004 – 2010

Panel A reports the descriptive statistics for the dependent variables for all observations, panel B reports the descriptive statistics for the sample firms with pre-crisis debt>0, and panel C reports the descriptive statistics for the control variables, respectively. The dependent variable is measured as short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), and short-term debt over lagged total assets (STD/L.TA), or as long-term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Tangibility is the ratio of fixed assets over total assets. Profitability is EBIT divided by total assets. Size is the natural logarithm of total assets. Market-to-book is the ratio of the market value of assets divided by the book value of assets. Uniqueness is the ratio of R&D expenses divided by total assets. Asset maturity is a book-value weighted measure of asset maturity of long-term assets and current assets. All variables are winsorized at the 1% level in both tails of the distribution.

Distribution

Variable Obs. Mean St. Dev. 10th _{Median 90}th

Panel A: All Observations

STD/L.STD 24,245 4.42 20.01 0.02 1.00 4.68 STD/L.TD 26,927 0.52 1.53 0.00 0.13 1.04 STD/L.TA 33,057 0.21 1.05 0.00 0.01 0.22 LTD/L.LTD 23,813 2.46 9.94 0.14 0.98 2.17 LTD/L.TD 26,903 1.14 2.95 0.00 0.78 1.49 LTD/L.TA 32,972 0.23 0.37 0.00 0.11 0.57

Panel B: Sample Firms with Pre-Crisis Debt>0

STD/L.STD 22,335 4.80 20.80 0.21 1.02 5.18 STD/L.TD 23,167 0.60 1.63 0.01 0.18 1.13 STD/L.TA 24,046 0.29 1.22 0.00 0.03 0.32 LTD/L.LTD 19,814 2.68 10.44 0.46 0.99 2.35 LTD/L.TD 20,434 1.30 3.13 0.16 0.83 1.59 LTD/L.TA 20,970 0.31 0.39 0.01 0.21 0.66

Panel C: Control Variables

15 In most analyzes, however, I focus on firms that use already short- or long-term debt in the pre-crisis period. With this exclusion, I expect higher descriptive statistics (except the standard deviation) for the dependent variables, because firms with zero debt are excluded from the sample. Table 2, panel B, shows that, among firms that use already short- or long-term debt before the crisis, the descriptive statistics for the variables are indeed higher. The median value of STD/L.STD and LTD/L.LTD remains close to 1. Since firms with zero debt are excluded from the sample, the number of observations in Panel B are lower than in Panel A. In both Panel A and Panel B, the number of observations for short-term debt are slightly higher than the number of observations for long-term debt. This indicates that more companies make use of short-term debt compared to long-term debt.

Panel C presents the descriptive statistics for the control variables. The mean value of profitability is negative, which indicates that the average EBIT for the sample period is negative. This could be a consequence of the crisis, but this is not certain. The lower number of observations for the market-to-book ratio is due to the fact that a lot of firms have missing values for either the stock price or shares outstanding. The mean value of the market-to-book ratio is also quite high in comparison to previous literature, and characterized by a high standard deviation. Furthermore, a lot of firms do not have R&D expenses, which indicates the lower number of observations for uniqueness. The descriptive statistics for tangibility, size, uniqueness, and asset maturity are in line with previous literature (Frank and Goyal, 2009; Rajan and Zingales, 1995; Stohs and Mauer, 1996).

16

Figure 1

STD/L.TA and LTD/L.TA over the Sample Period

The graph plots the mean values of STD/L.TA and LTD/L.TA over the sample period, excluding firms without any debt in the pre-crisis period. Both variables are winsorized at the 1% level in both tails of the distribution. The x-axis reports the year, and the y-axis the level of short- or long-term debt over lagged total assets.

5. Results

Table 3 reports the mean values and t-test results for different measures of short- and long-term debt between two periods: pre-crisis and crisis. The table only includes firms with positive short- and long-term debt before the crisis. Table 3 shows that the mean values between the pre-crisis and crisis period for almost all variables are significantly different at the 1% level. For STD/L.STD, the mean value in the crisis period is higher than the mean value in the pre-crisis period, suggesting that a firms’ short-term debt level relative to last year’s increases during the crisis. However, the mean value in the pre-crisis period is also above 1, indicating that short-term debt is higher relative to last year. But since the mean value is higher in the crisis period, the amount of short-term debt seems to increase faster during the crisis. The mean value of STD/L.TD is lower in the crisis period than the mean value in the pre-crisis period. This indicates that firms, during the financial crisis, have less short-term debt relative to last year’s total debt. The mean value of STD/L.TA does not deviate between the pre-crisis and crisis period. In addition, the natural logarithm of the amount of short-term debt is taken, as this makes the right-skewed distribution more symmetric. Ln(STD) shows an increase in the amount of short-term debt in the crisis period, being statistically significant at the 1% level. Therefore, the amount of short-term debt, and thus the level of short-term debt over last year’s short-term debt, seems to increase during the financial crisis. However, the level of short-term debt relative to last year’s total debt seems to decrease. For long-term debt, the mean values of the three dependent variables are lower in the crisis period than the mean

17

Table 3

T-test for Differences in Mean Values Between Pre-Crisis and Crisis Period if Debt>0

The table reports the means and t-test results for differences in means for different variables across two periods, assuming unequal variances. The table only includes firms which have positive debt in the pre-crisis period. The years 2005, 2006 and 2007 are defined as the pre-crisis period, and 2008, 2009 and 2010 as the crisis period. The variables reported here are: short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), short-term debt over lagged total assets (STD/L.TA), the natural logarithm of the amount of short-term debt (Ln(STD)), long-term debt over their own lagged value (LTD/L.LTD), term debt over lagged total debt (LTD/L.TD), long-term debt over lagged total assets (LTD/L.TA), and the natural logarithm of the amount of long-long-term debt (Ln(LTD)). All the variables are winsorized at the 1% level in both tails of the distribution.

Short-term Debt Long-term Debt

Pre-crisis Crisis Pre-crisis Crisis

STD/L.STD LTD/L.LTD Mean 4.40 5.24 Mean 3.04 2.28 t-statistic -2.93*** t-statistic 5.20*** STD/L.TD LTD/L.TD Mean 0.64 0.57 Mean 1.51 1.06 t-statistic 3.27*** t-statistic 10.54*** STD/L.TA LTD/L.TA Mean 0.29 0.29 Mean 0.32 0.29 t-statistic 0.12 t-statistic 6.61*** Ln(STD) Ln(LTD) Mean 2.00 2.38 Mean 3.70 4.16 t-statistic -10.32*** t-statistic -10.48*** ***, **, and *, significant at the 1, 5, and 10 percent level, respectively.

18 A. Regression

This paper investigates the impact of the financial crisis on U.S. firms’ short- and long-term debt, and therefore it is mainly interested in the sign, value, and significance of the crisis dummy. A positive (negative) sign of the crisis dummy indicates that a firm has more (less) short- or long-term debt in its capital structure.

Table 4

Regression Results of Crisis on Short-term Debt if STD>0

The table shows the results for firms with positive short-term debt in the pre-crisis period. The dependent variable is measured as term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), and short-short-term debt over lagged total assets (STD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. Tangibility is the ratio of fixed assets over total assets. Profitability is EBIT divided by total assets. Size is the natural logarithm of total assets. Market-to-book is the ratio of the market value of assets divided by the book value of assets. Uniqueness is the ratio of R&D expenses divided by total assets. Asset maturity is a book-value weighted measure of asset maturity of long-term assets and current assets. All the control variables are lagged by one year to mitigate simultaneity concerns. The regression is estimated using ordinary least squares (OLS) and fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

STD/L.STD STD/L.TD STD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES OLS FE OLS FE OLS FE

Crisis 0.86* 1.96*** -0.04 0.03 0.02* 0.10*** (0.45) (0.57) (0.03) (0.03) (0.01) (0.02) Tangibility -5.54*** -8.17** -0.66*** -0.94*** 0.16*** 0.40* (1.58) (4.12) (0.11) (0.36) (0.05) (0.24) Profitability 0.32 0.84** -0.02 0.08* -0.37*** -0.31*** (0.22) (0.40) (0.02) (0.04) (0.03) (0.05) Size 0.11 -1.68** -0.10*** -0.41*** -0.02*** -0.27*** (0.07) (0.77) (0.01) (0.08) (0.00) (0.05) Market-to-book 0.05* 0.10** 0.00 0.00 0.02*** 0.02*** (0.03) (0.05) (0.00) (0.00) (0.00) (0.00) Uniqueness 1.25 0.58 -0.03 -0.03 -0.19** -0.29* (0.78) (1.63) (0.09) (0.14) (0.10) (0.16) Asset maturity 0.21** 0.20 0.01* 0.02 -0.01*** -0.01 (0.08) (0.14) (0.01) (0.01) (0.00) (0.00) Constant 4.62*** 14.72*** 1.32*** 3.09*** 0.12*** 1.47*** (0.50) (4.29) (0.05) (0.43) (0.03) (0.26)

Firm effects No Yes No Yes No Yes

Observations 10,816 10,816 11,240 11,240 11,768 11,768

R-squared 0.00 0.00 0.04 0.01 0.55 0.36

19 Table 4 presents the results of the relationship between the financial crisis and the firms’ short-term debt level for pooled OLS and fixed effects. The table only includes firms with positive short-term debt before the crisis. Regression results for STD/L.STD and STD/L.TA show that the coefficients of the crisis dummy are positive and statistically significant among both specifications, but show a greater effect among the fixed effects specification. A positive coefficient of the crisis dummy indicates that firms increase their short-term borrowing in periods of crisis. For example, the dummy variable crisis in column 6 of Table 4 indicates that short-term debt relative to last year’s total assets increases with 10% during the crisis. This finding is consistent with the theories of Brunnermeier and Oehmke (2013) and Milbradt and Oehmke (2015) and the empirical evidence of Chen et al. (2018). Regression results for STD/L.TD show that the coefficients of the crisis dummy are statistically insignificant, while the t-test showed a statistically significant difference between the two periods. The values of the crisis dummy vary over the three dependent variables, but this is in line with their mean value. The mean value of STD/L.STD is higher than the mean value of STD/L.TA, and therefore the coefficient estimate of the crisis dummy is higher for STD/L.STD relative to STD/L.TA. The findings hold after controlling for the possibility that a firms’ short-term debt may have changed simply because firms experience changes in profitability during the crisis, which may affect their debt level. The coefficients of the firm-specific control variables are broadly consistent with the findings of existing literature, especially for STD/L.TA (Frank and Goyal, 2009; Rajan and Zingales, 1995; Titman and Wessels, 1988). However, some of the signs of the coefficients change among the different dependent variables. For example, tangibility is negative for STD/L.STD and STD/L.TD, but positive for STD/L.TA. The positive coefficient indicates that firms with more tangible assets have more short-term debt in their capital structure. This is not in line with the prediction of Rajan and Zingales (1995).

20

Table 5

Regression Results of Crisis on Long-term Debt if LTD>0

The table shows the results for firms with positive long-term debt in the pre-crisis period. The dependent variable is measured as long-term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The control variables are defined in Table 4 and lagged by one year to mitigate simultaneity concerns. The regression is estimated using ordinary least squares (OLS) and fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

LTD/L.LTD LTD/L.TD LTD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES OLS FE OLS FE OLS FE

Crisis -0.59*** -0.08 -0.39*** -0.21*** -0.01** 0.01** (0.22) (0.26) (0.06) (0.07) (0.01) (0.01) Tangibility -3.53*** -8.38** -1.06*** -2.21*** 0.17*** 0.18** (0.94) (3.45) (0.23) (0.79) (0.03) (0.09) Profitability 0.17 0.63* 0.05 0.35*** -0.07*** -0.02 (0.22) (0.33) (0.05) (0.11) (0.01) (0.01) Size -0.21*** -3.28*** -0.03** -1.21*** -0.01*** -0.15*** (0.04) (0.53) (0.01) (0.15) (0.00) (0.01) Market-to-book 0.02 -0.01 0.01 -0.00 0.01*** 0.00* (0.02) (0.04) (0.01) (0.01) (0.00) (0.00) Uniqueness 0.59 0.44 0.19 0.32 0.01 0.04 (0.61) (1.18) (0.19) (0.39) (0.04) (0.06) Asset maturity 0.09 0.07 0.04*** 0.05 0.01*** -0.00 (0.07) (0.23) (0.01) (0.05) (0.00) (0.00) Constant 4.97*** 24.40*** 1.76*** 8.96*** 0.20*** 1.09*** (0.35) (3.37) (0.10) (0.89) (0.01) (0.09)

Firm effects No Yes No Yes No Yes

Observations 10,533 10,533 10,929 10,929 11,408 11,408

R-squared 0.01 0.02 0.01 0.04 0.15 0.15

***, **, and *, significant at the 1, 5, and 10 percent level, respectively.

debt are also in line with the findings of previous literature. However, it is predicted that larger firms have more long-term debt in their capital structure (Rajan and Zingales, 1995), but the coefficients of size are negative. This means that in my sample larger firms have less long-term debt in their capital structure.

21 there might be other factors, not incorporated in the regression, that influence the level of short- and long-term debt. Furthermore, some models have a very high constant (intercept), mainly among the fixed effects specification. In addition, the number of observations in the regressions is lower than expected. This is because the control variables market-to-book and uniqueness have many missing values, and OLS and FE only include observations in the regression of which no (control) variable is missing. To overcome the problem of losing too much observations, Appendix D presents the univariate results of Table 4 and Table 5. The main difference with Table 4 is that the coefficient of the crisis dummy for STD/L.TD is negative and statistically significant at the 1% level. For long-term debt (Appendix D, Table 2), the coefficients of the crisis dummy for all dependent variables are negative and statistically significant at the 1% level.

B. Robustness checks

For robustness, I check whether the impact of the financial crisis is different among firms with high and low levels of short- and long-term debt. Table 6 presents the regression results for firms with high and low levels of short-term debt. The sample is split according to the median value of the relevant dependent variable in the pre-crisis period. However, the number of observations for firms with high and low levels of short-term debt is not equal. This is because the sample is split according to the median value of the dependent variable. However, the number of observations of the dependent variable are not equal to the number of observations in the regression. This is because fixed effects only includes observations in the regression of which no (control) variable is missing. Table E1 of Appendix E presents the univariate regression of Table 6, in which the number of observations for firms with high and low levels of short-term debt is equal.

22

Table 6

Robustness Check for Firms with High and Low Levels of Short-term Debt

The table shows the results for firms with positive short-term debt in the pre-crisis period and splits the sample according to the median value of the relevant dependent variable. The dependent variables reported here are: short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), and short-term debt over lagged total assets (STD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The control variables are defined in Table 4 and lagged by one year to mitigate simultaneity concerns. The regression is estimated using fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

STD/L.STD STD/L.TD STD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES High Low High Low High Low

Crisis 2.21** -0.05*** -0.01 -0.00 0.08*** -0.00* (1.05) (0.01) (0.06) (0.00) (0.02) (0.00) Tangibility -16.79 0.12 -2.49*** 0.03 -0.17 0.01** (13.46) (0.13) (0.74) (0.02) (0.24) (0.00) Profitability 1.01 -0.01 0.12* 0.00 -0.20** 0.00 (1.08) (0.01) (0.07) (0.00) (0.08) (0.00) Size -5.10** 0.07*** -0.53*** 0.00 -0.30*** -0.00** (2.20) (0.02) (0.12) (0.00) (0.07) (0.00) Market-to-book 0.05 -0.00 0.01 0.00 0.02*** 0.00 (0.09) (0.00) (0.01) (0.00) (0.01) (0.00) Uniqueness -4.44 0.09** 0.13 -0.02* -0.11 0.00* (5.95) (0.04) (0.25) (0.01) (0.28) (0.00) Asset maturity -0.13 -0.01 0.06** -0.00 0.01 0.00 (0.51) (0.01) (0.02) (0.00) (0.00) (0.00) Constant 43.64*** 0.13 4.23*** 0.04** 1.81*** 0.01*** (13.08) (0.13) (0.66) (0.02) (0.41) (0.00)

Firm effects Yes Yes Yes Yes Yes Yes

Observations 5,381 5,435 5,538 5,702 5,808 5,960

R-squared 0.01 0.01 0.02 0.00 0.32 0.02

***, **, and *, significant at the 1, 5, and 10 percent level, respectively.

relative to their total assets decrease their level of short-term debt during the crisis, although the effect is small.

23 negative and statistically significant among firms with high and low levels of long-term debt. This implies that firms decrease their long-term debt level during the crisis, independent of they have high or low levels of long-term debt. However, the effect is greater for firms with a high level of long-term debt in the pre-crisis period. The impact of the financial crisis does not differ significantly between firms with high and low levels of LTD/L.TA. Table E2 of Appendix E gives the univariate regression of Table 7.

Table 7

Robustness Check for Firms with High and Low Levels of Long-term Debt

The table shows the results for firms with positive long-term debt in the pre-crisis period and splits the sample according to the median value of the relevant dependent variable. The dependent variables reported here are: long-term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The control variables are defined in Table 4 and lagged by one year to mitigate simultaneity concerns. The regression is estimated using fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

LTD/L.LTD LTD/L.TD LTD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES High Low High Low High Low

Crisis -1.15 -0.03*** -0.31* -0.03*** 0.01 -0.00 (0.75) (0.01) (0.17) (0.01) (0.01) (0.00) Tangibility -22.03*** 0.16 -5.54** 0.15* 0.05 0.05*** (7.83) (0.11) (2.24) (0.09) (0.18) (0.02) Profitability 1.50 -0.01 1.20*** -0.00 0.01 -0.01*** (0.96) (0.02) (0.44) (0.01) (0.02) (0.00) Size -4.49*** 0.03* -2.21*** 0.04*** -0.31*** 0.00 (1.39) (0.02) (0.41) (0.01) (0.03) (0.00) Market-to-book -0.01 -0.00 0.06 0.00 0.00 0.00 (0.11) (0.00) (0.04) (0.00) (0.00) (0.00) Uniqueness 4.75 -0.06 1.98 0.01 0.04 -0.01* (3.31) (0.06) (1.74) (0.04) (0.09) (0.01) Asset maturity -0.01 -0.00 0.12 -0.00 -0.00 0.00 (0.27) (0.00) (0.08) (0.00) (0.00) (0.00) Constant 39.88*** 0.45*** 17.55*** 0.21*** 2.39*** 0.04* (9.42) (0.12) (2.84) (0.08) (0.21) (0.02)

Firm effects Yes Yes Yes Yes Yes Yes

Observations 5,184 5,349 5,565 5,364 5,625 5,783

R-squared 0.02 0.01 0.06 0.01 0.23 0.01

***, **, and *, significant at the 1, 5, and 10 percent level, respectively.

24 coefficient of the crisis dummy is no longer significant. The coefficient of the crisis dummy for STD/L.TD is negative and statistically significant at the 10% level, but was insignificant in the basic specification (Table 4). For the three measures of long-term debt, the coefficients of the crisis dummies are negative and statistically significant at the 5% level. Therefore, the results of Table 5 are robust to the inclusion of firms without long-term debt in the pre-crisis period. However, the results change a bit when including firms without short-term debt. This could also be a consequence of the inclusion of industry effects. The R-squared remains low for the most models. Appendix F gives the univariate regression of Table 8.

Table 8

Robustness Check Including Firms Without any Debt in the Pre-Crisis Period

The table includes firms without any short- or long-term debt in the pre-crisis period and controls for industry effects. The dependent variables reported here are: short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), short-term debt over lagged total assets (STD/L.TA), long-term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The control variables are defined in Table 4 and lagged by one year to mitigate simultaneity concerns. The regression is estimated using fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

Short-term Debt Long-term Debt

(1) (2) (3) (4) (5) (6)

VARIABLES /L.STD /L.TD /L.TA /L.LTD /L.TD /L.TA Crisis 0.78*** -0.04* 0.02 -0.60** -0.37*** -0.02*** (0.28) (0.02) (0.01) (0.25) (0.06) (0.00) Tangibility -4.63*** -0.33*** 0.25*** -2.99*** -0.61** 0.28*** (1.39) (0.12) (0.07) (0.93) (0.26) (0.04) Profitability 0.17 -0.04** -0.37*** 0.21 0.07*** -0.01 (0.15) (0.02) (0.04) (0.14) (0.02) (0.01) Size 0.20** -0.08*** -0.01* -0.13** 0.03** 0.00 (0.09) (0.01) (0.01) (0.05) (0.01) (0.00) Market-to-book 0.04** 0.00 0.01** 0.02 0.00 0.00* (0.02) (0.00) (0.01) (0.02) (0.00) (0.00) Uniqueness 0.86 -0.07 -0.29*** 0.58* 0.19 0.03 (0.79) (0.06) (0.09) (0.34) (0.11) (0.03) Asset maturity 0.12* 0.01* -0.01*** 0.09 0.03** 0.00 (0.07) (0.01) (0.00) (0.07) (0.01) (0.00) Constant 3.80*** 1.08*** 0.07* 4.05*** 1.15*** 0.08*** (0.67) (0.05) (0.04) (0.31) (0.08) (0.02)

Industry effects Yes Yes Yes Yes Yes Yes

Observations 11,843 13,124 17,160 11,508 13,103 17,125

R-squared 0.00 0.03 0.45 0.00 0.01 0.04

25 Table 9 shows the regression results when using an alternative time period for crisis. The crisis dummy in this table represents the years 2008 and 2009, in line with the studies of Demirgüç-Kunt, Martinez-Peria, and Tressel (2015) and Iqbal and Kume (2014). For STD/L.STD and STD/L.TA, the coefficients and the significance of the crisis dummy are in line with Table 4, which means that firms increase short-term debt borrowing in periods of crisis. For the measures of long-term debt, the coefficient of LTD/L.LTD is no longer significant. Appendix G presents the univariate regression of Table 9.

Table 9

Robustness Check for Alternative Time Period for Crisis

The table shows the results for firms with positive short-term debt in the pre-crisis period and uses an alternative time period for the crisis. The dependent variables reported here are: short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), short-term debt over lagged total assets (STD/L.TA), term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), and long-long-term debt over lagged total assets (LTD/L.TA). Crisis is a dummy variable for the years 2008 and 2009. The control variables are defined in Table 4 and lagged by one year to mitigate simultaneity concerns. The regression is estimated using fixed effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

Short-term Debt Long-term Debt

(1) (2) (3) (4) (5) (6)

VARIABLES /L.STD /L.TD /L.TA /L.LTD /L.TD /L.TA

Crisis 2.30*** 0.06 0.05*** -0.23 -0.14** 0.00 (0.61) (0.03) (0.01) (0.24) (0.06) (0.00) Tangibility -7.43* -0.92** 0.43* -8.45** -2.29*** 0.18** (4.09) (0.36) (0.24) (3.45) (0.79) (0.09) Profitability 0.78** 0.08** -0.32*** 0.63* 0.37*** -0.02 (0.39) (0.04) (0.05) (0.33) (0.11) (0.01) Size -1.55** -0.41*** -0.25*** -3.24*** -1.27*** -0.15*** (0.73) (0.07) (0.04) (0.52) (0.14) (0.01) Market-to-book 0.11** 0.00 0.02*** -0.01 -0.00 0.00* (0.05) (0.00) (0.00) (0.04) (0.01) (0.00) Uniqueness 0.24 -0.04 -0.29* 0.51 0.31 0.04 (1.62) (0.14) (0.16) (1.18) (0.39) (0.06) Asset maturity 0.17 0.02 -0.01* 0.07 0.05 -0.00 (0.14) (0.01) (0.00) (0.18) (0.04) (0.00) Constant 14.11*** 3.12*** 1.37*** 24.19*** 9.25*** 1.07*** (4.17) (0.43) (0.25) (3.31) (0.90) (0.08)

Firm effects Yes Yes Yes Yes Yes Yes

Observations 10,816 11,240 11,768 10,533 10,929 11,408

R-squared 0.00 0.01 0.35 0.01 0.02 0.12

26 6. Conclusion

This paper investigates the impact of the financial crisis of 2008 on a firms’ short- and long-term debt level. The distinction between short- and long-long-term debt is important for firms’ financial stability, as more short-term debt is more risky for the company (has to be paid back within one year). To analyze the impact of the financial crisis on a firms’ debt level, I use different measures of short- and long-term debtas dependent variable, and regress crisis both on short- and long-term debt. The dataset used includes all publicly listed, non-financial firms in the U.S., and consists of yearly data over a 7-year period from 2004 to 2010. The study focuses on firms that use already debt financing before the financial crisis. Firms that do not have any debt on their balance sheet before the financial crisis may already be constrained in obtaining debt financing, or may not need any debt financing.

I find evidence for a positive relationship between the financial crisis and the use of short-term debt, which indicates that firms attract more short-short-term debt during the financial crisis of 2008. Firms, which have already a high level of short-term debt in the pre-crisis period, increase their level of short-term debt even more during the crisis, which is more risky for the company. Firms with a low level of short-term debt in the pre-crisis decrease their level of short-term debt during the crisis, although the effect is very small. In addition, I find evidence for a negative relationship between the financial crisis and the use of long-term debt, which indicates that firms decrease their level of long-term debt during the financial crisis. Firms with a high level of long-term debt before the crisis decrease their long-term debt more during the financial crisis compared to firms with a low level of long-term debt before the crisis. However, the impact of the financial crisis on a firms’ long-term debt level is a little bit ambiguous. On one hand, the t-tests show that the amount of long-term debt increases. On the other hand, in the regressions, the coefficients of the crisis dummy are negative and statistically significant, indicating that firms attract less long-term debt during the financial crisis. Therefore, the level of long-term debt increases during the financial crisis, but slower compared to the period before the crisis. A firms’ short-term debt level increases faster during the financial crisis. The results are robust to different measures of long-term debt. This is also true for short-term debt, although short-term debt relative to last year’s total debt decreases in some specifications. Moreover, the results are robust to different estimation methods and alternative specifications.

27 of crisis. However, the regression results do not support the existing empirical evidence regarding an increase in leverage during periods of crisis (Demirgüç-Kunt, Martinez-Peria, and Tressel, 2015; Fosberg, 2012; Iqbal and Kume, 2014). A possible explanation for this would be that I use the lagged values of long-term debt, total debt and total assets.

A limitation of this research is that my definition of short- and long-term debt include some items which are used for non-financing purposes. This introduces a potential bias if the use of this non-financing item is correlated with the independent variables. In addition, an explanation for an increase in short-term debt and a decrease in long-term debt could be that over the crisis long-term debt turned into short and no additional long-term debt was given. Compustat provides the current portion of long-term debt that falls due within one year. However, Compustat does not provide details about debt issuance. Therefore, it is difficult to examine whether no additional debt is given. Moreover, the R-squared of the regressions is limited, which suggest that there might be other factors, not incorporated in the regression, that influence the level of short- and long-term debt. Furthermore, the standard errors in the regressions are not corrected for positive autocorrelation and are biased downwards. This means that the test statistics of the coefficients in the regressions are probably too high, and therefore have been wrongly identified as statistically significant.

This research only used data of publicly listed U.S. firms. It would be interesting to see whether the financial crisis has the same impact on European or Asian firms’ short- and long-term debt. The impact of the financial crisis on short- and long-term debt could differ among market based economies (UK) and bank based economies (Japan, Germany). Another possible extension of this research could be to investigate whether the impact of the crisis on firms’ short- and long-term debt is different between stable and suffering industries, or between industries which rely highly on external finance or not. Furthermore, it would be interesting to know what the increase in short-term debt and the decrease in long-term debt caused. For example, an increase (decrease) in short-term notes, commercial paper, bond issuance, or bank loans. Finally, it would be interesting to see whether short-term debt still increases when all current liabilities are included.

7. References

Adrian, T., Colla, P., Shin, H.S., 2012. Which financial frictions? Parsing evidence from the financial crisis of 2007-09. NBER Macroeconomics Working Paper 18335.

28 Barclay, M., Smith, C.W., 1995. The maturity structure of corporate debt. Journal of Finance 50(2), 609-631.

Becker, B., Ivashina, V., 2014. Cyclicality of credit supply: firm level evidence. Journal of Monetary Economics 62, 76-93.

Brooks, C., 2014. Introductory econometrics for finance. Cambridge University Press, Cambridge.

Brunnermeier, M.K., Oehmke, M., 2013. The maturity rat race. Journal of Finance 68(2), 483-521.

Campello, M., Graham, J.R., Harvey, C.R., 2010. The real effects of financial constraints: evidence from a financial crisis. Journal of Financial Economics 97(3), 470-487.

Chen, S., Ganum, P., Liu, L., Martinez, L., Martinez-Peria, M.S., 2018. Debt maturity and the use of short-term debt: evidence from sovereigns and firms. International Monetary Fund Working Paper.

Cortina J.J., Didier, T., Schmukler, S.L., 2016. Corporate borrowing and debt maturity: the effects of market access and crises. World Bank Policy Research Working Paper 7815.

Demirgüç-Kunt, A., Martinez-Peria, M.S., Tressel, T., 2015. The impact of the global financial crisis on firms’ capital structure. World Bank Policy Research Working Paper 7522.

Diamond, D.W., 1991. Debt maturity structure and liquidity risk. Quarterly Journal of Economics 106(3), 709-737.

Diamond, D.W., He, Z., 2014. A theory of debt maturity: the long and short of debt overhang. Journal of Finance 69(2), 719-762.

Duchin, R., Ozbas, O., Sensoy, B.A., 2010. Costly external finance, corporate investment and the subprime mortgage credit crisis. Journal of Financial Economics 97(3), 418-435.

Fiore, F., de, Uhlig, H., 2014. Corporate debt structure and the financial crisis. NBER Macroeconomics Working Paper 20730.

Flannery, M.J., 1986. Asymmetric information and risky debt maturity choice. Journal of Finance 41(1), 19-37.

Fosberg, R.H., 2012. Capital structure and financial crisis. Journal of Finance and Accountancy 11, 46-52.

29 Graham, J.R., Harvey, C.R., 2001. The theory and practice of corporate finance: evidence from the field. Journal of Financial Economics 60(2/3), 187-243.

Iqbal, A., Kume, O., 2014. Impact of financial crisis on firms’ capital structure in UK, France, and Germany. Multinational Finance Journal 18(3/4), 249-280.

Jong, A., de, Kabir, R., Nguyen, T.T., 2008. Capital structure around the world: the roles of firm- and country-specific determinants. Journal of Banking & Finance 32(9), 1954-1969.

Kahle, K.M., Stulz, R.M., 2013. Access to capital, investment, and the financial crisis. Journal of Financial Economics 110(2), 280-299.

Krishnamurthy, A., 2010. How debt markets have malfunctioned in the crisis. Journal of Economic Perspectives 24(1), 3-28.

Levine, R., Lin, C., Xie, W., 2016. Spare tire? Stock markets, banking crises, and economic recoveries. Journal of Financial Economics 120(1), 81-101.

Milbradt, K., Oehmke, M., 2015. Maturity rationing and collective short-termism. Journal of Financial Economics 118(3), 553-570.

Modigliani, F., Miller, M.H., 1958. The cost of capital, corporate finance and theory of investment. American Economic Review 48(3), 261-297.

Myers, S.C., 1977. Determinants of corporate borrowing. Journal of Financial Economics 77(5), 147-175.

Ozkan, A., 2002. The determinants of corporate debt maturity: evidence from UK firms. Applied Financial Economics 12, 19-24.

Pattani, A., Vera, G., Wackett, J., 2011. Going public: UK companies’ use of capital markets. Bank of England Quarterly Bulletin 51(4), 319-334.

Rajan, R.G., Zingales, L., 1995. What do we know about capital structure? Some evidence from international data. Journal of Finance 50(5), 1421-1460.

Scherr, F.C., Hulburt, H.M., 2001. The debt maturity structure of small firms. Financial Management 30(1), 85-111.

Stohs, M.H., Mauer, D.C., 1996. The determinants of corporate debt maturity structure. Journal of Business 69(3), 279-312.

30 8. Appendices

Appendix A

Table A Hausman test STD/L.STD

Variable Fixed (b) Random (B) Difference (b-B) Square root standard error Crisis 1.50 0.81 0.69 0.19 Tangibility -30.9 -3.39 0.30 2.35 Profitability 0.33 0.09 0.25 0.21 Size -1.10 0.04 -1.15 0.44 Market-to-book 0.05 0.03 0.02 0.02 Asset maturity 0.10 0.13 -0.03 0.06 Chi-square 17.63 Probability 0.01 Appendix B

Table B Number of Observations by Industry

Industry SIC Codes No. of Observations

Agriculture, Forestry and Fishing 0100-0999 193

Mining, Oil and Gas Extraction 1000-1499 2,764

Construction 1500-1799 457

Manufacturing 2000-3999 20,031

Transportation, Communications, Electric, Gas and Sanitary Services

31 Appendix C

Table C T-test for Differences in Mean Values Between Pre-Crisis and Crisis Period

The table includes firms without any short- or long-term debt in the pre-crisis period.The table reports the means and t-test results for differences in means for different variables across two periods, assuming unequal variances. The years 2005, 2006 and 2007 are defined as pre-crisis, and 2008, 2009 and 2010 as crisis. The variables reported here are: short-term debt over their own lagged value (STD/L.STD), short-term debt over lagged total debt (STD/L.TD), short-term debt over lagged total assets (STD/L.TA), the natural log of short-term debt (Ln(STD)), long-term debt over their own lagged value (LTD/L.LTD), long-term debt over lagged total debt (LTD/L.TD), long-term debt over lagged total assets (LTD/L.TA), and the natural log of long-term debt. The variables are winsorized at the 1% level in both tails of the distribution.

Short-term Debt Long-term Debt

Pre-crisis Crisis Pre-crisis Crisis

32 Appendix D

Table D1 Univariate Regression Results of Crisis on Short-term Debt if STD>0

The table shows the results for firms with positive short-term debt in the pre-crisis period. The dependent variable is measured as term debt over their own lagged value (STD/L.STD), short-term debt over total debt (STD/L.TD), and short-short-term debt over lagged total assets (STD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The regression is estimated using ordinary least squares (OLS) and fixed-effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

STD/L.STD STD/L.TD STD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES OLS FE OLS FE OLS FE

Crisis 0.84*** 1.44*** -0.07*** -0.07*** -0.00 0.02* (0.29) (0.34) (0.02) (0.02) (0.02) (0.01) Constant 4.40*** 4.12*** 0.64*** 0.64*** 0.29*** 0.28***

(0.14) (0.16) (0.02) (0.01) (0.01) (0.01)

Firm effects No Yes No Yes No Yes

Observations 22,335 22,335 23,167 23,167 24,046 24,046

R-squared 0.00 0.00 0.00 0.00 0.00 0.00

***, **, and *, significant at the 1, 5, and 10 percent level, respectively. Table D2 Univariate Regression Results of Crisis on Long-term Debt if LTD>0

The table shows the results for firms with positive long-term debt in the pre-crisis period. The dependent variable is measured as long-term debt over their own lagged value (LTD/L.LTD), long-term debt over total debt (LTD/L.TD), and long-term debt over lagged total assets (LTD/L.TA). Crisis is a dummy variable for the years 2008, 2009 and 2010. The regression is estimated using ordinary least squares (OLS) and fixed-effects (FE). Standard errors adjusted for heteroscedasticity are reported in parentheses below the coefficient estimates. All variables are winsorized at the 1% level in both tails of the distribution.

LTD/L.LTD LTD/L.TD LTD/L.TA

(1) (2) (3) (4) (5) (6)

VARIABLES OLS FE OLS FE OLS FE

Crisis -0.76*** -0.86*** -0.44*** -0.49*** -0.03*** -0.04*** (0.15) (0.16) (0.04) (0.05) (0.01) (0.00) Constant 3.04*** 3.09*** 1.51*** 1.53*** 0.32*** 0.33***

(0.11) (0.07) (0.04) (0.02) (0.00) (0.00)

Firm effects No Yes No Yes No Yes

Observations 19,814 19,814 20,434 20,434 20,970 20,970

R-squared 0.00 0.00 0.01 0.01 0.00 0.01