Master Thesis Finance
Master International Financial Management
M.P.J.A. Oud BSc.
1
Cash Holdings and Financial Distress: A
Focus on European Firms
Faculty of Economics And Business
University of Groningen
Abstract
Higher cash holdings lead to financial distress. Using both the tradeoff theory and the precautionary motive for holding cash, this research finds that when the benefits of holdings cash, which is a reduced probability of financial distress, outweigh its opportunity cost, firms hold more cash. When modeling financial distress, ordered logit models allow for multiple categories of financial distress to be estimated and the individual importance of the independent variables to be measured. In doing so, the marginal effects of the independent variables on each category of financial distress can be calculated separately, while the slope coefficients of the independent variables on all categories together can be estimated in addition.
Name student: M.P.J.A. Oud Student number: 1890158
Email address: m.p.j.a.oud@student.rug.nl Date: 16-01-2015
2
1. Introduction
This research finds that more cash holdings lead to financial distress. In an ordered logit model where the interest coverage ratio is regressed on cash holdings, two additional liquidity metrics, and seven control variables, this research finds that high cash holdings are an important predictor of financial distress. The dependent variable in the ordered logit model includes four categories of financially healthy firms and one category of financially distressed firms. These categories are composed of the interest coverage ratio ranges, where for the last category a firm is said to be in financial distress if its interest coverage ratio falls below one. The results indicate that firms having high cash holdings end up in the lowest category, which indicates financial distress.
The relevance of this paper is twofold, since it contributes to both the literature on cash holdings and financial distress. For the literature on cash holdings, my paper both validates and complements the results found by Acharya, Davydeko, and Strebulaev (2012). First of all, my paper validates the results found by these latter authors using a different dataset based on the European stock market. Additionally, my paper complements the research by Acharya, Davydeko, and Strebulaev (2012), by proposing how the results contribute to the debate on the purpose of cash holdings, using both the tradeoff theory and the pecking order theory. Secondly, for the literature on financial distress, my paper presents a novel model to predict financial distress. Using an ordered logit model, the importance of cash holdings in predicting financial distress is shown.
3 This research is conducted using a sample of 297 firms that are listed on the STOXX Europe 600 index and the data covers a period of 24 years, from 1990 until 2013. In total 6828 firm-years of data on non-financial companies is collected from Worldscope Datastream. To analyze the results, both the model coefficients and the marginal effects are analyzed. Firstly, the effects of the independent variables on all categories together is estimated using model coefficients. Secondly, the marginal effects of the independent variables on each separate category of the interest coverage ratio is calculated using limitpoints. The ordered logit based on EBIT is validated using several robustness checks. These robustness checks include using a binary logit model, estimating the coefficients using ordinary least squares, and including an ordered logit model in which the dependent variable is calculated using EBITDA.
Showing that firms with higher cash holdings are more likely to be in financial distress has several interesting implications. First of all and most strikingly, firms that have more cash holdings are not necessarily in good financial health. By showing that a cash hoarding policy is more likely to be followed by firms close to financial distress, a counter intuitive finding is presented. Secondly, in modeling financial distress, cash holdings are found to play an important role. Thirdly, this research shows the tradeoff faced by managers when considering saving cash. On the one hand, saving cash lowers the probability of financial distress, because enough funds are available to repay debt obligations. On the other hand, saving cash has a negative effect on future cash flows, because holding cash creates an opportunity cost.
2. Literature review
2.1. Literature on cash holdings
4 anticipate future cash flow shortages. Secondly, holding large amounts of cash decreases future cash flows, because firms postpone their investments. My paper continues on the results in the Acharya, Davydeko, and Strebulaev (2012) paper by shifting the focus from credit spreads to the interest coverage ratio. In doing so, I expect, using the precautionary motive for holding cash, that an increase in cash holdings leads to financial distress.
Following Cambell and Taksler (2003) the interest coverage ratio is defined as an accounting based variable used to measure a firm’s financial health. Low interest coverage ratios thereby show if firms have difficulties in repaying their interest expenses, making low interest coverage ratios an indicator of credit risk. While both my paper and the research of Acharya, Davydeko, and Strebulaev (2012) use a measure of credit risk, using interest coverage ratios instead of credit spreads provides several benefits. Firstly, a credit spread contains information that is not specific to the risks inherent to the firm. Three studies in particular find that macro-economic variables such as: investor sentiment, demand and supply shocks, and monetary policies influence credit spreads (see, e.g., Tan and Yan, 2010; Collin-Dufresne, Goldstein, and Martin, 2001; Bhamra, Fisher, and Kuehn, 2011, respectively). The interest coverage ratio, on the other hand, uses firm specific information only. Secondly and relatedly, the peculiarity of the interest coverage ratio as a measure of credit risk is that it is a financial ratio that uses accounting data only. There exists a distinction between accounting variables and market based variables when assessing a firm’s credit risk. However, studies such as Das, Hanouna, and Sarin (2009), Cardone-Riportella, Samaniego-Medina, and Trujillo-Poncea (2014), and Agarwal and Taffler (2008) find that accounting based variables predict credit risk comparably or even better than market based variables. Interestingly, these authors opt for a hybrid model using both market and accounting based variables in order to determine a firm’s credit risk. Thirdly, a model using the interest coverage ratio can also be adopted when obtaining a firm’s credit spread is not possible. As an illustration, Moerland (1995) argues that Europe is classified as having a network oriented system where capital is often provided by banks, insurance companies, families or nations.
5 balance sheet. These authors find that for the market in the US the average cash-to-asset ratio has grown from 10.5% in 1980 to 23.2% in 2006. The debate on the purpose of cash holdings leads to two opposing views. On the one hand, cash is regarded as a hedging device that supports the timely payment of interest. On the other hand, cash is merely seen as negative debt and a consequence of the firm’s operating activities.
This latter view, which implies that cash holdings do not reduce the probability of financial distress, is adopted by the pecking order theory. The theoretical explanation of the pecking order theory explains why firms hold cash using information asymmetries. As explained in the research of Myers (1984) firms prefer cash over debt and debt over equity, because those financing options have information asymmetries embedded in them in a decreasing order. Subsequently, the pecking order theory states that cash is valuable to firms primarily because it allows firms to invest when investment opportunities occur. Or as proposed by Myers and Maljuf (1984), when using cash instead of external finance, firms will not be subject to the costs of asymmetric information that prevents them from investing. In terms of financial distress, however, there is no benefits from holdings cash under the pecking order theory. The pecking order theory states that the amount of cash holdings are subjected to the changes in the fortunes of the firm. Opler, Pinkowitz, Stulz, and Williamson (1999) note that the pecking order theory states that there is no optimal level of cash and the amount of liquid assets merely adjust as a consequence of the firm’s operating activities. For this to happen, the pecking order theory states that firms are indifferent between either holding cash or repaying debt. Cash holdings are thereby used to finance investments or repay debt and only when it is not used for either, firms accumulate cash. Thus, based on the pecking order theory, an increase in cash holdings is not an indicator of financial distress.
6 transaction cost theory from Keynes (1936), that the benefits of holding cash is its ability to surpass the additional costs when transferring liquid assets into cash. The second benefit argues that when holding cash, firms are able to finance its investment opportunities regardless of the availability and costs of external funds. The tradeoff theory therefore differs from the pecking order theory in terms of cash on two aspects. Firstly, when considering the benefits and costs to holding cash, cash holdings are no longer seen as negative debt as under the pecking order theory. Secondly, only the tradeoff theory assumes that there is an optimal level of cash holdings. Thus, when considering that the tradeoff theory balances the benefits and costs leading to an optimal level of cash, this theory is able to explain how increases in cash holdings lead financial distress. In order to see how the level of cash holdings increase by the benefits of holding cash in terms of financial distress, the precautionary motive for holding cash is introduced below.
7 Lastly, the precautionary motive for holding cash is assumed not to hold for financially healthy firms. It can be argued using the concept of financial constraints that only financially constrained firms benefit from holding large amounts of cash. For example, in terms of cash flow sensitivity, the research of Han and Qui (2007) finds that only the cash holdings of constrained firms are sensitive to cash flow volatility. Similarly, Faulkender and Wong (2006) find that financially constrained firms value cash holdings more, have more cash holdings, and rely more on cash holdings than firms that do not have difficulties in accessing the capital market. Importantly, the financial constraints in the studies above were proxied using a firm’s credit risk. Therefore, I assume that after a specific point, when firms are sufficiently financially healthy, the precautionary motive does not hold. Those firms have sufficient alternatives to financing and do not rely primarily on cash holdings to repay their debt obligations. Similarly, when using the tradeoff theory, it can be argued that for financially healthy firms the cost of saving cash outweighs the benefits, because there are plenty other ways to finance investments. For example, Acharya, Davydeko, and Strebulaev (2012) describe in their framework that firms which do not anticipate cash flow shortages, use their future cash flows as collateral to invest in profitable projects.
2.2. Literature on financial distress
8 focusing on accounting variables. As an example, financial distressed can be measured using negative earnings, cash flows below long term debt maturities, or negative interest coverage ratios (see, e.g., John, Lang, and Netter, 1992; Whittaker, 1999; Asquith, Gertner, Scharfstein, 1994, respectively). These three latter studies focus their definition of financial distress on the point where firms are unable to repay their debt obligations. Thirdly, a handful of studies use financial distress indicators based on information provided in the press. Two examples include the research of Brown, James, and Mooradian (1994) which measures financial distress using information on debt restructurings and the research of Gleason and Simpson (1999) that uses safety ratings for banking firms. These latter two definitions of financial distress are highly dependent on the methodology that is used and are therefore narrow in their nature. My definition of financial distress is similar to the second set of studies, since I use the point where a firm’s earnings before interest and taxes is insufficient to pay interest expenses
9 dividend payments until bankruptcy. Similarly, Johnsen and Melicher (1994) used three classifications of financial health: non-bankrupt, financially weak, and healthy firms. The drawback of this latter model is, however, that the dependent variable has multiple categories, but they are not ordered in a logical way. To solve this last problem an ordered logit or ordered probit model should be used. With ordered models, multiple stages of financial distress, ordered from low to high can be regressed on a set of explanatory variables using cumulative probabilities. One example of an ordered model that predicts financial distress as defined by a safety rating for banking firms is the research by Gleason and Simpson (1999). Thus, financial distress in my research is modeled using an ordered logit model, because multiple, ordered stages of financial distress can be estimated and the importance of the individual independent variables can be determined.
3. Methodology
3.1. The ordered logit model
My analysis shows that firms having higher cash holdings, have lower interest coverage ratios, making cash hoarding behavior a predictor of financial distress. An ordered logit model is used that includes three categories for financially healthy firms, one category for financially healthy firms that suffer from a high credit risk, and one category for financially distressed firms. Financial distress is defined as having an interest coverage ratio below one. The ordered logit model tests if higher cash holdings causes the interest coverage ratios to fall below one. In doing so, the ordered logit model regresses the ranges of the interest coverage ratio as a function of cash holdings, two additional liquidity metrics and seven control variables.
10 can’t be simply said that firms covering their interest expenses 20 times are twice as risky compared to firms covering their interest expense 40 times. Similarly it cannot be said that, when comparing interest coverage ratios, the difference between 1 and 5 is the same as the difference between 5 and 10. The dependent variable should therefore not be treated as interval or ratio data, but should instead be used as ordinal, discrete data. Although, Ordered models do not have a continuous variable, it does allow the dependent to be ordered from financially healthy to financially distressed firms. In doing so, the dependent variable is an integer valued value that can either be regressed using an ordered logit or ordered probit model. Using either of those models produces similar results, however for my research the ordered logit model produced a higher Pseudo R².
3.2. Model explanation
Ordered logit models relate the categories of the interest coverage ratio to the observed independent variables through an unobserved continuous linking variable. The ordered logit model therefore contains two parts. The first part relates the observed independent variables to the unobserved linking variable through a linear equation. The second part transforms the values of the unobserved linking variable to the categories of the interest coverage ratios.
The first part of the ordered logit model relates the liquidity metrics and control variables to the unobserved, continuous linking variable . This linear function of the ordered logit model is displayed in equation (1). Let be the values on the linking variable of firm at time , β the vectors of slope coefficients, a vector of liquidity variables including cash holdings, seven control variables that influence financial distress, and the error term. The model is estimated for time T, ranging from 1990 until 2013, for every firm using the yearly observations . The error term follows a logistic distribution and captures the effects for either wrongly measured variables or missing relevant variables. Note that there is no intercept, instead the effects for the independent variable for each category of the interest coverage ratio will be calculated separately using limitpoints. The limitpoints will be explained using equation (6).
11 The importance of the linking variable is that it determines the categories of the interest coverage ratio for firm at time . In doing so, the continuous values on are transformed into the different categories of , by the thresholds specified in equation (2). Equation (2) displays this transformation which is the second part of the ordered logit model. The variable is assigned the value one if the interest coverage ratio for firm at time is between hundred and twenty, the value two for interest coverage ratios below twenty until ten, the value three for interest coverage ratios below ten until five, the value two for interest coverage ratio below five until one, and the value five if the interest coverage ratio is below one.
(2)
Equation (3) shows the ordered logit model that includes both equation (1) and equation (2).
( + + + ) ( ) ( ) (3) Where, ( = ), ( = ), ( = ), ( = ), ( = ),
= are assigned values for the categories of financial distress according to the ordering showed in equation (2),
= a vector of liquidity variables,
= seven control variables that influence financial distress, β (1-7) = are parameters to be estimated,
12 Ordered logit models are based on the notion of cumulative probabilities, where an individual observation falls into = 1 , … , J categories. The cumulative probability represents the
probability that the th firm ends up in the th or higher category. Cumulative probabilities in line with equation (3) are defined as:
( )
(4)
When calculating the cumulative logits for the th or higher category, equation (5) can be used. The cumulative logits calculate the probability of ending up in a certain category versus the probability of ending up in a higher category. For example, when calculating the cumulative probability of ending up in category two, the log of the probability of ending up in categories two plus one divided by the probability of ending up in categories three until five, is taken. As can be seen in equation (5), the model calculates all cumulative logits except for one. Similar to equation (3), the cumulative logit for the first category of the interest coverage ratio is missing due to the cumulative nature of the logit.
( ) ( ( )) ( )
( )
(5)
When generalizing equation (5) and including the covariates, the logit function can be defined as in equation (6). This equation includes on the right hand side two vectors of slope coefficients and one limit point. Importantly, the equation shows that, in contrast to the two slope coefficients, only the limitpoint α is indexed by . The difference is that the β coefficients measure the slope of the independent variables on all categories taken together, while the limitpoint α is used to calculate the marginal effects for ending up in each category of the interest coverage ratio separately. Firstly, the workings of the limitpoint α is explained followed by an explanation of the coefficients.
l ( ) ( ( ))
13 The limitpoint α is the intercept for each category of the interest coverage ratio. The limitpoint α plays no role in the linear function as can be seen in equation (1) and for that reason it will be referred to as limitpoint instead of intercept. The limitpoints are indexed by and indicate the cumulative odds for ending up in the or lower category. Each limitpoint therefore shows the cumulative odds for ending up in a specific category of the interest coverage ratio for a one unit increase in the independent variables. In doing so, the cumulative odds for ending up in or lower category can be calculated for each specific independent variable separately when regressing the categories of the interest coverage ratio on the independent variables one by one. For example, when the regressing the categories of the interest coverage ratio on cash holdings, the limitpoint α indicates the cumulative odds for ending up in the or lower category for a one unit increase in cash holdings. To translate the limitpoints into their marginal effects, several transformations need to be made. Firstly, the limit points need to be exponentiated to calculate the cumulative odds. These cumulative odds then needs to be divided by one plus the cumulative odds, in order to obtain the cumulative probabilities. From these cumulative probabilities, the marginal effects can be calculated.
In contrast, the β coefficients are not indexed by and do not change for each categories of the interest coverage ratio separately. The cumulative probabilities of the β coefficients for the variables and are therefore the same for all categories. Hence, the β coefficients indicate the slope for the independent variables with reference to the categories of the interest coverage ratio. This slope indicates the log odds of moving up in the categories of the interest coverage ratio by a one unit increase of the independent variables. For example, the β coefficient for cash holdings indicates the log odds of moving from category to category +1 for a one unit increase in cash holdings. The result section includes a discussion on the outcomes of both the β coefficients and the intercepts α. Further, Table 5 displays the results for the β coefficients, while Table 7 shows the results for the marginal effects based on α.
3.2. The ordering of the dependent variable
14 Firstly, the ordering of the interest coverage ratio is inspired by the ordering used in the research of Blume, Lim, and Mackinlay (1998). This ordering of the interest coverage ratio is interesting, because it accounts for the non-linear relationship between interest coverage ratios and credit risk. In predicting credit ratings, the study of Blume, Lim, and Mackinlay (1998) find that an interest coverage ratio up till five is significant, positive, and has a large coefficient, whereas from this level onwards the interest coverage ratio lowers its coefficient until the ratio exceeds twenty, and stops being statistically significant. Consequently, low interest coverage ratios accurately predict credit risk, while the relative importance of having an high interest coverage ratio diminishes when interest coverage ratios are sufficiently high. For my own model, I have split the category with the lowest category of interest coverage ratios from the Blume, Lim and Mackinlay (1998) to include firms having financial distress. The ordering of the interest coverage ratio in Blume, Lim and Mackinlay (1998) is presented in Table 8 in Appendix A, while the ordering for my paper is presented in Table 1
Table 1
The Blume, Lim, and Mackinlay (1998) ordering with financial distress
This ordering shows the ordering that is used for the of ordered logit model as estimated in equation (3). The ordering is inspired on the ordering based on the Blume, Lim, and Mackinlay (1998) paper which is shows in Table 8 in Appendix A. The difference between this ordering and the ordering shown in Table 8 is that the fourth category is split into two additional categories to include a category for financial distress. The first category includes interest coverage ratios that ranges from hundred until twenty, the second category has interest coverage ratios below twenty until ten, the third category has interest coverage ratios below ten until five, the fourth category has interest coverage ratios below five until one, and the fifth category has interest coverage ratios below one until zero. Category one until four represent the categories for financially healthy firms, where category four includes financially healthy firms that suffer from substantial credit risk. Category five includes all financially distressed firms.
[20,100]
(20,10]
(10,5]
(5,1]
15 Based on the literature of financial distress and credit risk, the splitting of the last category in the ordering of Blume, Lim and Mackinlay (1998) can be reasoned. The literature on financial distress measures financial distress on the basis of having an interest coverage ratio that is below one (see, e.g., Desai and Jain, 1999; Asquith, Gertner, and Scharfstein, 1994; Nixon, Roenfeldt, and Sicherman, 2000). In the ordering for my paper interest coverage ratios below one are therefore used for the last category which indicates financial distress. Similarly, the literature on credit ratings, shows that a category of interest coverage ratios that ranges from below five until one can reasoned. The study of Amato and Furfine (2004), for example, finds that companies with credit scores ranging from CCC to C, on average have an interest coverage ratio of two and in the lowest 25th percentile even an coverage ratio of just above one. Another article by Shin and Moore (2003) finds that BBB rated companies have on average an interest coverage ratio of 4.9 when measured by S&P and 5.1 for scores by Moody’s. Further, Asbaugh-Skaife, Collins, and LaFond (2006) describe companies rated C as being in the lowest speculative class and companies rated BBB as in the lowest investment grade group. In sum, the literature on financial distress and credit risk shows that the fourth category in the Blume, Lim and Mackinlay (1998) is too broad. Instead the ordering should differentiate between healthy firms that suffer from high credit risk and firms which are in financial distress.
16 from category two to three. Secondly, Figure1 shows that the categories of financially healthy firms are sufficiently different from the category of financial distress. In terms of cash holdings, category five has the largest increase compared to category one until four.
Figure 1
Distribution of cash holdings per interest coverage ratio
Plot of the cash holdings with its corresponding interest coverage ratio. The panel data for cash holdings and the interest coverage ratio includes observation for 297 firms from the period 1990 until 2013.Cash is displayed on the horizontal axis and includes cash plus short term investments divided by total assets. On the vertical axis
INCOV represents the interest coverage ratio calculated as earnings before interest and taxes divided by interest
expense on debt. The figure shows nihil differences in cash holdings for interest coverage ratios ranging from hundred until ten. Cash holdings show the largest increases for firms having interest coverage ratios between five and zero. This figure validates that categories four and five are different from categories three until one in terms of cash holdings. Similarly, the figure shows that category four and five are different from each other.
3.3. The variables
17 include depreciation which is a non-cash expense. Therefore, I consider EBIT a superior measure, since non-cash expenses include non- actual cash payments.
The independent variables are based on both the research of Acharya, Davydeko, and Strebulaev (2012) and on the literature of financial distress. The main variable of interest, cash holdings, are given by a firm’s cash and cash equivalents scaled by firm size through total assets. In addition, two additional liquidity variables are included which are working capital scaled by total assets and total current assets divided by total current liabilities. The results in the Acharya, Davydeko, and Strebulaev (2012) model show for all three liquidity measures a positive relationship between liquidity and credit spreads. Subsequently, the prediction for my model is that an increase in liquidity measured either as cash holdings, working capital, or the current ratio causes firms to have an interest coverage ratio that falls below one.
Inspired by Acharya, Davydeko, and Strebulaev (2012) are three control variables: the return on the STOXX Europe 600, asset volatility, and size. The returns for the STOXX Europe 600 includes annual observations for the entire STOXX Europe 600 stock index. The asset volatility is measured by the annualized standard deviation of assets returns calculated from monthly observations. Size is calculated as the logarithm of total assets, where total assets are measured as the sum of total liabilities plus total shareholders’ equity. Since the inclusion of the above control variables is inspired on the research of Acharya, Davydeko, and Strebulaev (2012), the relationship between those control variables and financial distress is expected to be similar. Consequently, asset volatility is expected to have a positive relationship with financial distress, while both the returns on the STOXX Europe 600 and size are expected to have negative relationship with financial distress.
18 Secondly, the study by Campbell, Hilscher, and Szilagyi (2008) about distressed stocks indicate that distressed firms have a higher market beta and a lower return on equity.
Leverage is defined as total debt divided by total debt plus shareholders ‘equity and the operating profit margin is defined as operating income divided by net sales or revenues times hundred. To explain why leverage and the operating profit margin are assumed to have an effect on financial distress the logic of Asquith, Gertner, and Sscharfstein (1994) is used. These latter authors argue that changes in two particular factors related to the performance of management lead to financial distress: an increase in interest payments and insufficient operating performance. Firstly, for leverage it is expected that a higher leverage leads to financial distress. Intuitively, a higher leverage leads to higher interest payments and when assuming that higher interest payments increase the difficulties in repaying debt, higher leverage will lead to financial distress. Therefore, it assumed that leverage has a negative effect upon financial distress. Similarly, for the second control variable concerning profitability a negative relationship with financial distress is assumed. Again, following the logic of Asquith, Gertner, and Sscharfstein (1994), a lower profitability leads to a deterioration of operating performance. Therefore, when assuming that a lower firm performance causes financial distress, a low operating profit margin will increase the probability of financial distress.
19
4. Data description
4.1. Data sources and sample selection
To test equation (3) information on the interest coverage ratio, cash holdings, two additional liquidity measures, and seven control variables predicting financial distress is needed. This information is collected for 297 firms that are listed on the STOXX Europe 600 index during the period of 1990-2013. In total 6828 firm-years are collected from Worldscope Datastream. The variables include data as reported for the beginning of the year. Picking firms from the STOXX Europe 600 index reduces sample bias, because this index is representative for the whole European market. The Europe STOXX 600 index includes large, mid, and small cap companies from eighteen countries in the European region. All firms in the data sample were active during the sample period from 1990 until 2013. In 2014, after the testing period, two out of the 297 firms in the sample were removed from the STOXX Europe 600 index. Table 10 in appendix A shows which two firms were removed.
The total population of 600 firms on the STOXX Europe 600 index was reduced to a sample of 297 firms, because of missing data and the clear focus on non-financial companies. Firstly, firms are deleted if they have missing data for cash holdings or EBIT for more than 5 years. Clearly, it is not desirable to measure the relationship of cash holdings on financial distress using only on fractions of firm-year information. Within at least of 19 years of data, firms have sufficient time to change their position in terms of financial distress and adapt their cash holdings accordingly. Also, including firms that report seldom EBIT might introduce bias into the sample, because firms might purposely decide not to report EBIT when having financial distress. Secondly, only non-financial firms are included in the sample. The sample includes only non-financial firms for two reasons. Non-financial firms seldom report cash holdings and most of them were deleted for that reason. In addition, the study of Acharya, Davydeko, and Strebulaev (2012) excludes all non-financial firms. If financial firms had been included in my study, the results would not be comparable to the findings in the Acharya, Davydeko, and Strebulaev (2012) paper. To delete all financial firms, DataStream’s industry classification is used. In doing so, Insurance, banks and other financial institutions are deleted from the sample. Consequently, the sample includes information on industrial, utility and transportation firms only.
20 involves replacing the missing variable on cash holdings with the mean value of cash holdings dependent on a second variable. This second variable is the interest coverage ratio. This means if cash holdings are not available for a firm with an interest coverage ratio of above 7 and less than 8, the mean cash holdings of all firms with an interest coverage ratio between 7 and 8 will be used. The imputation only entails a handful of observations and the results do not differ significantly with or without cash imputation. Since this difference is nihil, the results as measured without cash imputation are left out of this study. Secondly, those companies that include observations with missing data on EBIT are removed. Importantly, these observations include nonrandom data. Mostly, when a firm does not report sufficient data on EBIT, the independent variables suffer also from missing data. This might indicate that the firm was not operating at that time or did not have sufficiently strict reporting requirements. Further, the interest coverage ratio is not easily replaced by a substitution variable. This is because values on the interest coverage ratio can be very volatile. Since, values can double or triple between years, replacing missing variables with an average or last year’s value is not a valid measure.
Lastly, it should be noted that the values of the interest coverage ratio are limited to ranges from zero up till hundred. This approach is similar to the ordering in Blume, Lim, and Mackinlay (1998). These authors find that interest coverage ratios below zero and above hundred do not reveal any relative importance in terms of information on credit risk. Importantly, values below zero are not considered outliers and should not be deleted from the sample. Extent research finds that firms with a negative EBITDA or EBIT have a higher probability of ending up in financial distress (see, e.g., Maskara and Mullineaux, 2011; John, Lang, and Netter, 1992). Values for the interest coverage ratio below zero are therefore set to equal zero. Similarly, interest coverage ratios that exceed hundred are set to equal hundred.
4.2. Time trend
21 finds for the US market. Secondly, average values for all three liquidity metrics are not influenced by time specific events such as financial turmoil. Table 9 in appendix A shows the amount of firms in each category of the interest coverage ratio per year. The amount of firms that are in either category one, three and five are stable across years, while the amount of firms in categories two more than doubles. Similarly, the amount of firms in category four decreases substantially. However, these changes do not impact the amount of firms having financial distress relative to the amount of firms that are financially healthy.
Figure 2
Cash holdings, working capital, and the current ratio per year
Plot of the average values of cash holdings, working capital, and the current ratio per year. Cash is defined as the average ratio of cash plus short term investments divided by total assets for all firms in the sample per year.
Worca represents working capital and is defined as the average ratio of working capital divided by total assets
for all firms per year. Current represents the current ratio and is defined as the average ratio of total current assets divided by total current liabilities for all firms per year. Firstly, the figure shows no increasing trend for either of the three liquidity metrics. Secondly, no time specific events such as financial turmoil influences the amount of liquidity.
4.3. Cash holdings and the interest coverage ratio
The cash holdings for all financially healthy firms are displayed using a scatterplot in Figure 3A. On average, the cash to asset ratio for financially healthy firms are less than two, only a few cash holdings exceed this ratio. When looking at the scatter plots per interest coverage ratio category in Figure 4A until 4D in appendix A, it can be seen that only firms in the third
22 and fourth category of the interest coverage ratio exceeds the cash ratio of two. While there are only three firms in category three that exceed the cash ratio of two, category four shows a substantial amount of firms that exceed this ratio. For the financially distressed firms, Figure 3B shows the ratio of cash holdings to total assets. Two things should be noted when analyzing this figure. Firstly, substantially more firms are having cash holdings that exceed the cash ratio of two. This shows that financially distressed firms have often higher cash holdings compared to financially healthy firms. Secondly, a large amount of financially distressed firms have cash holdings that does not exceed the cash ratio of two, while some firms even have a cash ratio equal to zero. This latter result might impede the predictability of cash holdings on financial distress. It can be explained, however, when considering that these firms have alternatives to cash holdings that might protect them from financial distress.
Figure 3
Cash holdings per interest coverage ratio for financially healthy firms and financially distressed firms
Figure 3A shows the amount of cash holdings per interest coverage ratio for financially healthy firms. Figure 3B shows the amount of cash holdings per interest coverage ratio for financially distressed firms. The panel data for cash holdings and the interest coverage ratio include observation for 297 firms from the period 1990 until 2013. Cash is defined as cash holdings plus short term investments divided by total assets. INTCOV represents the interest coverage ratio and is defined as earnings before interest and taxes divided by the interest expense on debt. For the financially healthy firms the cash to asset ratio is on average less than two and only a few firms have a cash to asset ratio that exceeds two. For the financially distressed firms substantially more firms have a cash to asset ratio that exceeds two. The cash to asset ratio is, however, for some financially distressed firms still around zero.
3A 3B
23 To compared if the means of the financially distressed firms are statistically different from the means of the financially healthy firms, a Welch t-test is performed. The Welch-t test tests the two samples for the hypothesis that the means of the samples are equal. As can be seen in Table 2, the P-value is for this test is significant which rejects the hypothesis that the means of the financially healthy firms are equal to the means of the financially distressed firms. The statistics in the lower part of Table 2 shows that the average cash holdings of the financially distressed firms are 1.61 higher than the average cash holdings of the financially healthy firms. Lastly, the Welch t-test is used instead of the normal t-test, because the Welch t-test does not assume equal variances.
Table 2
Comparing the mean cash holdings for financially healthy and financially distressed firms using a Welch t-test.
This table shows the results of the Welch t-test that tests the hypothesis that the mean cash holdings of the financially distressed firms equals the mean cash holdings of the financially distressed firms. The mean cash holdings is defined as the average of the ratio of cash plus short term investments divided by total assets for each firms, calculated separately for financially healthy and financially distressed firms. The hypothesis is rejected at the 1% level and the category statistics below shows that the mean cash holdings of the financially distressed firms is 1.61 higher than the mean cash holdings of the financially healthy firms.
Based on the results above it can be concluded that financially distressed firms differ significantly in terms of average cash holdings from financially healthy firms. It is still, however, interesting to see how the mean values of cash holdings change per category of the interest coverage ratio. Table 11 in appendix A shows the descriptive statistics of the cash holdings per category of the interest coverage ratio. This table shows two interesting things. Firstly, the majority of categories have nearly similar amounts of observations, only category four and two are either over -or underrepresented. The deviations are, however, not substantial and therefore comparisons are still valid. Secondly, the variable of interest, the mean cash holdings per interest coverage ratio category, contains values that change in magnitude according to the predicted pattern. The average cash holdings in the first and the
Method
Welch t-test df Value P-value
1213,36 12,062 0,000
Category statistics
Variable Count Mean Std. Dev. Std. Err. of mean Distressed 1205 1,76 4,644 0,134
24 second category are nearly identical. These latter categories both have a low average cash holdings which differs only one percent. The average cash holdings for category three shows a subtle increase compared to categories two and one. When comparing categories four and five with categories one until three, a significant increase in average cash holdings is visible. Category five, which represents financially distressed firms, has the most cash holdings on average. It should be noted, however, that categories four and five also have the highest levels of standard deviation. Unfortunately, this finding might impede the predictive power of cash holdings on financial distress. Still, overall it should be noted that by looking at the descriptive statistics, it seems that there is a strong relationship between high average cash holdings and low interest coverage ratios.
4.4. Control variables
Table 3 displays the summary statistics of the dependent variable and the independent variables. Strikingly, the average leverage ratio for financially healthy firms are two percent higher compared to the ratio for financially distressed firms. This result is counter to what can be expected by looking at the literature on financial distress. All other results are, however, in line with previous expectations on the independent variables.
In line with the research of Acharya, Davydeko, and Strebulaev (2012) are the mean values of all three liquidity metrics and the mean values of the control variables used in the latter study. Firstly, cash holdings, workings capital and the current ratio all have higher means for financially distressed firms compared to financially healthy firms. These changes of mean values correspond to the positive relationship found in the above research. Similarly, the mean values of the control variables also correspond to the findings in Acharya, Davydeko, and Strebulaev (2012). This latter research shows for assets volatility a positive relationship, while for both size and return on the stock index it shows a negative relationship. Comparably, Table 3 shows for the annualized standard deviation of asset returns substantially higher means for financially distressed firms compared to financially healthy firms, while both size and the returns on the STOXX Europe 600 index show lower means for financially distressed firms.
25 and the return on equity are substantially lower for financially distressed firms as compared to financially healthy firms. Financially healthy firms on the other hand are far more profitable as financially distressed firms and have a substantially higher return on equity. In sum, all observations are in line with the literature on financially distress except for leverage.
Table 3
Summary statistics for the interest coverage ratio and all independent variables
This table shows the summary statistics for the interest coverage ratio and all the independent variables for both financially healthy and financially distressed firms. Firms are financially distressed if their interest coverage ratio drops below one, which corresponds to firms being in category five. INTCOV measures the categories of the interest coverage ratio and is calculated as earnings before interest and taxes divided by interest expense on debt.
CASH is a variable that measures the amount of cash holdings defined as cash plus short term investments
divided by total assets. WORKCA is working capital and is calculated by dividing working capital by total assets.
CURRENT represents the current ratio and is the ratio of total current assets divided by total current liabilities. LEVERAGE is defined as total debt divided by total debt plus total shareholders ‘equity. BETA is the firm’s beta
with respect to the market returns. OPM is the operating profit margin and is calculated as operating income divided by net sales or revenues times hundred. ROE stands for return on equity and is calculated as fund from operations divided by last year’s common equity times hundred. ROA measures asset volatility and is defined as the annualized standard deviation of the asset returns and is calculated by annualizing the monthly observations on the return on assets. SIZE is calculated as the logarithm of total assets, where total assets are calculated as the sum of total liabilities plus total shareholders’ equity. RSTOXX represents the annual returns on the STOXX Europe 600. The values of CASH, WORKCA, CURRENT, BETA, and ROA are in line with expectations higher for financially distressed firms than for financially healthy firms. The values of INTCOV, OPM, ROE, SIZE, and
RSTOXX are in line with expectation higher for financially healthy firms. Contrary to expectations is the higher
value for LEVERAGE for financially healthy firms.
Table 3 shows some puzzling findings for financially distressed firms for the minimum value of cash, the current ratio, leverage, and the return on assets. When looking at the values for all
Financially Healthy
INTCOV CASH WORKCA CURRENT LEVERAGE BETA OPM ROE ROA SIZE RSTOXX Mean 20,65 0,15 0,11 1,46 0,44 0,94 11,46 39,96 0,78 15,68 329,37 Median 7,66 0,08 0,09 1,29 0,39 0,93 9,56 28,63 0,00 15,73 355,45 Maximum 100,00 25,44 0,86 65,26 175,49 2,71 1048,30 15761,29 114,66 20,59 547,62 Minimum 1,00 0,00 -1,24 0,02 -26,94 -1,48 -3922,73 -160,33 0,00 9,20 86,35 Std. Dev. 29,23 0,62 0,17 1,15 2,42 0,44 68,59 225,07 3,58 1,71 147,46 Observations 5623 5623 5308 5308 5582 5551 5550 5478 5095 5582 5623 Financially Distressed
26 observations for these variables, it shows that for cash, leverage, and the current ratio this involves outliers. Only a view observations of these variables have a value of zero. When looking at the return on assets, however, more than three hundred observations report a return on assets that is zero. Additionally, a lot of observations are missing for this variable. Therefore, my interpretation is that the minimum value of zero for the return on assets can be caused by two reasons. Firstly, the minimum value can be explained by outliers. Secondly, the minimum value can be explained by measurement error. Possibly, the data in Datastream for this variable is not as accurate.
4.5. Assumptions of the statistical model
Logistic regressions have, in contrast to linear models such as Ordinary Least Squares, not as stringent assumptions. Firstly, logistic regressions do not assume linearity between the dependent and independent variables, since the dependent variable is not of continuous nature. Secondly, logistic regressions do not assume that either the independent variables or the error terms follows a normal distribution. Lastly, logistic regressions do not require its variances to be of heteroscedastistic nature. Logistic regressions do, however, have one stringent assumption that needs to be met. This latter assumption requires that the independent variables are independent predictors of the dependent variable by minimizing the collinearity between the independent variables.
27
Table 4
Correlation summary including the dependent variable and all independent variables
This table shows the correlation between the dependent variable with the independent variables and the correlation between independent variables. The dependent variable is named CATEGORY which measures in which of the five categories of the interest coverage ratio the firm is located. CASH is a variable that measures the amount of cash holdings defined as cash plus short term investments divided by total assets. WORKCA is working capital and is calculated by dividing working capital by total assets. CURRENT represents the current ratio and is ratio of total current assets divided by total current liabilities. LEVERAGE is defined as total debt divided by total debt plus total shareholders ‘equity. BETA is the firm’s beta with respect to the market returns.
OPM is the operating profit margin and is calculated as operating income divided by net sales or revenues times
hundred. ROE stands for return on equity and is calculated as fund from operations divided by last year’s common equity times hundred. ROA measures asset volatility and is defined as the annualized standard deviation of the asset returns and is calculated by annualizing the monthly observations on the return on assets. SIZE is calculated as the logarithm of total assets, where total assets are calculated as the sum of total liabilities plus total shareholders’ equity. RSTOXX represents the annual returns on the STOXX Europe 600. Even the highest correlation, which is between working capital and the current ratio, is not considered too high.
Correlation Matrix
CATEGORY CASH WORKCA CURRENT BETA LEVERAGE OPM ROE ROA SIZE RSTOXX CATEGORY 1 CASH 0,206 1 WORKCA -0,095 0,121 1 CURRENT -0,065 0,078 0,611 1 BETA 0,021 0,003 0,027 0,003 1 LEVERAGE 0,217 -0,074 -0,439 -0,280 0,047 1 OPM -0,198 0,008 -0,021 0,068 -0,059 -0,040 1 ROE -0,003 0,000 -0,034 -0,023 -0,021 0,100 0,004 1 ROA 0,068 0,058 0,027 0,020 0,002 0,019 -0,021 0,007 1 SIZE -0,130 -0,210 -0,276 -0,159 0,123 0,253 0,017 -0,028 -0,087 1 RSTOXX -0,090 -0,027 -0,140 -0,081 -0,012 0,051 0,080 0,008 0,006 0,305 1 5. Empirical results
5.1. Coefficient and significance analysis
28
Table 5
Estimation result for the ordered logit model using five categories based on EBIT as the dependent variable
This table shows the estimation results for 5639 observation from the sample of financially distressed and financially healthy firms. The ordered logit model is estimated using GLM robust variances and the results show that all coefficients are significant at least at the 5% level. The coefficients for the return on equity and beta are significant at the 5% level, while the current ratio is only marginally significant at the 10% level. The Pseudo R² has a value of 0,105. The coefficients show the log odds of moving up one category of the interest coverage ratio by a one unit increase in the independent variables. The dependent variable is composed of the five categories of the interest coverage ratio, where four categories include financially healthy firms and one category contains financially distressed firms. CASH is a variable that measures the amount of cash holdings defined as cash plus short term investments divided by total assets. WORKCA represents working capital and is calculated by dividing working capital by total assets. CURRENT represents the current ratio and is ratio of total current assets divided by total current liabilities. LEVERAGE is defined as total debt divided by total debt plus total shareholders ‘equity. BETA is the firm’s beta with respect to the market returns. OPM is the operating profit margin and is calculated as operating income divided by net sales or revenues times hundred. ROE stands for return on equity and is calculated as fund from operations divided by last year’s common equity times hundred.
ROA measures asset volatility and is defined as the annualized standard deviation of the asset returns and is
calculated by annualizing the monthly observations on the return on assets. SIZE is calculated as the logarithm of total assets, where total assets are calculated as the sum of total liabilities plus total shareholders’ equity.
RSTOXX represents the annual returns on the STOXX Europe 600.
Coefficient Std. Error z-Statistic P-value
CASH 1,674 0,098 17,161 0,000 WORKCA -1,506 0,200 -7,521 0,000 CURRENT 0,060 0,031 1,941 0,052 BETA 0,115 0,053 2,162 0,031 LEVERAGE 2,467 0,146 16,872 0,000 OPM -0,060 0,003 -19,966 0,000 ROE -0,001 0,000 -2,224 0,026 ROA 0,023 0,007 3,281 0,001 SIZE -0,127 0,016 -7,771 0,000 RSTOXX -0,001 0,000 -3,961 0,000 N 5639 Pseudo R² 0,105
29 ratio variable. This contrary results, however, is not due to the correlation with the current ratio, because even when regressed separate, this variable still shows a negative sign on its coefficient. The other variables show similar signs and are almost similar in magnitude as in the Acharya, Davydeko, and Strebulaev (2012) paper. Both cash and the current ratio are positive, while the coefficient on cash is much bigger in magnitude compared to the current ratio. Also, both size and the return on the STOXX Europe 600 index show, in line with the above paper, a negative sign which is very small in magnitude. While, both leverage and the asset volatility show a positive relationship similar to the findings in the paper mentioned above, the magnitude of the assets volatility variable differs from the results found in the latter paper. Lastly, Table 5 shows that asset volatility is small in magnitude which runs contrary to the findings in Acharya, Davydeko, and Strebulaev (2012).
30 The lower part of Table 6 shows how much predictions the model has performed correctly. Unfortunately, the model only predicts less than 40 % of the interest coverage ratio categories. The model’s predictive power could be improved by two adjustment. Firstly, more independent variables can be added on the condition that they significantly contribute to the predictability of the model. Secondly, the amount of categories in the dependent variable can be reduced. In that way, the differences between the categories become bigger leading to more accurate predictions. For now, it can be concluded that the model is sufficiently accurate to answer the question if more cash holdings lead to financial distress.
Table 6
Model fit and relevance of the ordered logit model using five categories based on EBIT as the dependent variable
This table shows the results for the Wald test, the log likelihood ratio test, and the prediction evaluation. The Wald test which tests the hypothesis that all coefficients of the independent variables equal zero is rejected at the 1% level. Similarly, the log likelihood ratio test which tests that the restricted model fits the data comparable to the unrestricted model is rejected at the 1% level. The bottom of the table shows that the model places less than 40% of the firms in the correct category of the interest coverage ratio.
Model Fit Test
Wald test [1159,746] Chi-square (0,000)
Model Fit Test
Log likelihood ratio
Model -2 log likelihood Chi-Square df prob. Intercept only 17890,31 1883,528 10 0,000 Full model 16006,78 Predictability Prediction evaluation % Correct 38,96% % Incorrect 61,04%
5.2. Analysis of the marginal effects
31 likely end up by a one unit change in the independent variables. Table 7 shows that nearly all marginal effects and coefficients are significant at the 1% level. The marginal effects for the full model are calculated followed by the marginal effects for all three liquidity metrics. Firstly, the full model as shown on the upper part of Table 7, shows that the marginal effects for category one and three are rather small. This small marginal effect shows that for a one unit increase in the independent variables, the probability of ending up in category two or three is small. Hence, in terms of the marginal effects for the full sample, it would make sense to combine category two and three. Further, Table 7 shows that changes in the independent variables are most likely to cause firms to end up in category five, since this marginal effect is the highest.
32
Table 7
The marginal effects for the full model, cash holdings, working capital, and the current ratio
The marginal effects show how a one unit increase in the independent variables relates to the probability of ending up in a specific category of the interest coverage ratio. Firms are hereby most likely to end up in the category with the highest marginal effect. The marginal effects are calculated from the limitpoints and are displayed on the right side under the heading individual probability. For the full model it is shown that changes in the independent variables will most likely cause firms to end up in the fifth category which represents financial distress. Similarly, changes in cash holdings, workings capital and the current ratio will most likely cause firms to end up in the fifth category indicating financial distress. All limitpoints are significant except for the fourth category of the current ratio.
5.3. Robustness checks
Table 16 and 17 in appendix B show the results from the regression as displayed in Table 5, but for different model specifications, using both a binary logit and an ordinary least squares model. Firstly, it is interestingly to compare the results from the ordinary least squares model in Table 17 to the results from the ordered logit model in Table 5. Previously in the methodology section it was argued that regressing the relationship between the interest coverage ratio and cash holdings should not be performed by linear regressions. It was argued
Limitpoints Prob. Cumulative Probability Individual Probability
33 that the information content within the dependent variable is not continuous. In contrast, it was argued that the link between the interest coverage ratio and financial distress could best be estimated while assuming that the dependent variable contains ordinal data. When looking at Table 17, it can be seen that, when using a linear relationship that regresses the interest coverage ratio as a continuous variable on the set of independent variables similar to Table 5, this does not produce valid results. All variables in Table 17 have the opposite sign compared to the variables in Table 5. The results are not only in contrast to the results specified in Table 5, they are also in contrast to the results found by Acharya, Davydeko, and Strebulaev (2012) as well as the literature on financial distress. Furthermore, three variables are not even significant at the 10% level and one variable is not significant at the 5% level, while they are significant in the model specified in Table 5. Overall, it can be concluded that regressing the interest coverage ratio as a continuous, dependent variable on the set independent variables provides invalid results. Thus, the relationship between the interest coverage ratio and cash holdings can best be estimated by treating the dependent variable as ordinal data.
The binary logit model is calculated with a dichotomous dependent variable that separates financially healthy from financially distressed firms. In doing so, the categories ranging from one until four are used as financially healthy, whereas category five represents the financially distressed firms. When comparing Table 16 and Table 5, the models are similar in terms of probabilities, signs, and magnitudes. The most notable differences are that beta and leverage change in magnitude, the return on the STOXX Europe 600 has a reversed sign and that the current ratio is now only marginally significant, while the return on the STOXX Europe 600 index is now significant at the 1% level. Overall, it can be concluded that that the ordered logit model has given valid results, since these results are supported by the results from the binary logit model.
5.4. Discussion of the results
34 holdings are a positive predictor of credit risk. Two other liquidity metrics; working capital and the current ratio, on the other hand, show mixed results when compared to the above study. While, the current ratio shows a significant, positive relationship with financial distress, working capital is found to have a significant, negative impact upon financial distress. Therefore, for the other two liquidity metrics, my research only partially confirms the results found in the Acharya, Davydeko, and Strebulaev (2012) study.
The findings in this research supports both the tradeoff theory and the precautionary motive for holding cash. The research by Acharya, Davydeko, and Strebulaev (2012), concludes that higher cash holdings leads to higher credit risk using the precautionary motive. While, my paper not only highlights the importance of the precautionary motive, the tradeoff theory is argued to contribute to the understanding of the predictive role played by cash holdings as well. Firstly, the precautionary motive for cash holdings predicts that cash hoarding firms postpone investment and instead hold cash when they anticipate financial distress. The precautionary motive therefore states that there is an hedging attribute associated with cash holdings that protects firms from possible cash flow shortfalls in the future. Secondly, the tradeoff theory complements the precautionary motive with the notion that there is an optimal level of cash holdings. Thereby should the benefits of reducing the probability of financial distress be outweighed against the opportunity cost of holding cash. The tradeoff theory therefore complements the precautionary motive, by using the argument of an optimal cash level to explain why only firms anticipating financial distress increase their cash holdings.
6. Conclusion
The results show that managers should make a tradeoff when deciding the optimal level of cash holdings in their firm. On the one hand, cash holdings function as a precaution towards financial distress by lowering the probability of having financial distress, and on the other hand saving cash decreases future profitability due to a decrease in investment. Managers should therefore outweigh the benefits of holding cash to its opportunity cost.
35 While there are good theoretical reasons and empirical results supporting the causal relationship of cash holdings on firm risk, a debate on causality is still inevitable.
36
7. References
Acharya, V., Davydenko, S.A., Strebulaev, I.A., 2012. Cash holdings and credit risk. The Review of Financial Studies 25, 3572-3609.
Agarwal, V., Taffler, R., 2008. Comparing the performance of market-based and accounting-based bankruptcy prediction models. Journal of Banking & Finance 32, 1541-1551.
Aharony, J., Jones, C.P., Swary, I., 1980. An analysis of risk and return characteristics of corporate bankruptcy using capital market data. The Journal of Finance 35, 1001-1016.
Altman, E.I., Frydman, H., Kao, D., 1985. Recursive partitioning for financial classification: the case of financial distress. The Journal of Finance 40, 269-291.
Amato, J. D., Furfine, G.H., 2004. Are credit ratings procyclical? Journal of Banking & Finance 28, 2641-2677.
Ashbaugh-Skaifea, H., Collins, D.W., LaFond, R., 2006. The effects corporate governance on firms' credit ratings. Journal of Accounting and Economics 42, 203-243.
Asquith, P., Gertner, R., Scharfstein, D., 1994. Anatomy of financial distress: an examination of junk-bond issuers. The Quarterly Journal of Economics 109, 625-658.
Bates, T.W., Kahle, K.M., Stulz, R.M., 2009. Why do U.S. firms hold so much more cash than they used to? The Journal of Finance 64, 1985-2021.
Bhamra, H.S., Fisher, A.J., Kuehn, L., 2011. Monetary policy and corporate default. Journal of Monetary Economics 58, 480-494.
Blume, M.E., Lim, F., Mackinlay, A.G., 1998. The declining credit quality of U.S. corporate debt: myth or reality? The Journal of Finance 53, 1389-1413.
Bolton, P., Chen, H., Wang, N., 2011. A unified theory of Tobin’s q, corporate investment, financing, and risk management. The Journal of Finance 66, 1545-1578.
Brown, D.T., James, C.M., Mooradian, R.M., 1994. Asset sales by financially distressed firms. Journal of Corporate Finance 1, 233-257.
Campbell, J.Y., Hilscher, J., Szilagyi, J., 2008. In search of distress risk. The Journal of Finance 63, 2899-2939.
Campbell, J.Y., Taksler, G.B., 2003. Equity volatility and corporate bond yields. The Journal of Finance 58, 2321-2350.
Cardone-Riportella, C., Samaniego-Medinaa, R., Trujillo-Poncea, A., 2014. Examining what best explains corporate credit risk: accounting based versus market-based models. Journal of Business Economics and Management 15, 253-276.
