Cash holdings and leverage: an empirical investigation into the relation between cash levels and leverage in US listed firms

Cash holdings and leverage: an empirical investigation

into the relation between cash levels and leverage in

US listed firms

Abstract

This research paper examines the relation between corporate cash holdings and leverage, and attempts to formulate the relation if present. I use a sample of 9834 US listed firms extracted from the period of 2005 to 2010. The estimation method used is two-stage least-squares (TSLS) with firm-fixed effects. I find evidence of a negative relation between corporate cash holdings and leverage, supporting the substitution theory, the pecking order theory, the trade-off theory and/or the free cash flow theory. A second hypothesis is in place to investigate the non-linearity of the relation between corporate cash levels and leverage. Using a quadratic model, I similarly find evidence in favor of a theory supporting a negative relation between cash holdings and leverage. Although the squared leverage term is highly significant and one could speak of a non-linear relationship between cash holdings and leverage, a linear model slightly fits the data better. My findings partly corroborate with extant research on this topic.

JEL classification: G32

Keywords: Corporate cash holdings, leverage

2

1

Introduction

In the area of working capital management, the subject of corporate cash holdings is one which has thoroughly been investigated. Early research mainly concerned itself with the motives for holding cash. In the real world, one can assume that markets are imperfect, so firms are forced to make trade-off decisions with regard to the level of cash holdings. According to pertinent literature the main benefit of having cash is often motivated by having the ability to finance valuable growth opportunities when they become available, often at unexpected times, and when other sources of financing, such as those obtained from the equity markets, come expensive. A concern associated with holding cash may nevertheless be that too much cash then becomes "tied up" in the firm. A firm's predisposition to maintaining a certain cash level is motivated by several other major motives. These motives, along with a short history about the developments in relevant research in this field of cash management will be discussed next.

As previously mentioned, a plethora of studies has concerned itself with the determination of a corporate firm's optimal level of cash holdings. One of the first of these dates back to the 30's, when Keynes (1934) describes the motives for holding cash. He refers to the aforementioned motive that concerns transaction costs, which states that firms minimize expenses they would otherwise incur in making transactions such as asset liquidations in order to meet payments. He also mentions a second motive that concerns itself with the convenience of having a cash buffer in place for the financing of activities, which is especially economical when other sources of finance are scarce or expensive to obtain. The latter is referred to as the precautionary theory. Baumol (1952) specifically designed a model almost a decade later. He applies models from the field of inventory theory and uses it in the subject of monetary theory. His model suggests that a trade-off between an interest bearing account on one side and a non-interest bearing account on the other hand is prevalent. The latter account is in place to fulfill payment obligations as well as to deposit cash inflows. Cash transfers between the two accounts will take place at a given rate, determined by both direct and indirect costs. An optimal cash level is then reached when a transfer is made at a certain time interval. Higher values of security sales will thus entail a longer time interval, foregoing more interest earnings on the interest bearing account. On the other side of the spectrum, opportunity costs in the form of interest earnings will be minimized, as they will be diminished by the transactions costs involved in the process. The square root of the volume of transactions is consequently used to derive the minimum costs of cash balance. This simple model has been deemed to be useful in theory, however, as Miller and Orr (1966)

3 more infrequent timing due to (unexpected) events and other variables that are at play. This in turn causes the transfers between the two accounts to be larger, and more "lumpy", as Miller and Orr (1966) describe it. In their paper, they attempt to improve the model by using other assumptions, perhaps the most important one being that the changes in the cash balance are stochastic. Next to the transactions costs being small and fixed, as well as the existence of two accounts, they assume that transfers between the two accounts can almost occur instantly with technology prevalent at that time. A newly introduced assumption involves the required minimum cash balance, which they set at zero. A drift rate and serial independence are important assumptions in a Random Walk process. Miller and Orr (1966) similarly use these assumptions in their model to derive probabilities in order to determine optimum cash levels. This model has been considered by many to be closer to cash management policies in practice. The cash level is bound by an upper and a lower level, and lumpy transfers specified by specific algorithms are made when either boundary is reached. Finally, a minimal cash level can be identified by weighing the cost of the upper bound cash level and interest earnings. Kim, Mauer, and Sherman (1998) develop a theoretical model wherein the optimal level of cash holdings is developed. They find that this level increases depending on factors such as the cost of external financing, future cash flow variance, and the return of investment opportunities, while they state that this level decreases depending on the difference in return between physical assets and highly liquid assets.

Meyers and Majluf (1984) later wrote about the pecking order theory, which posits that that firms prefer to finance investments internally, then with debt and with equity as a last choice. This increase in costs from one source of finance to the next has to do with asymmetric information. Their

research focuses on the agency costs, and amongst other things suggests that firms are better off holding "financial slack", i.e. extra capacity such as cash to be used for opportunities, rather than foregoing potential positive net present value projects, the latter which would be in favor of the shareholders.

4 lower levels of net working capital. They also investigate country- level determinants of corporate cash levels. Their additional result that leverage has a negative relationship with cash holdings is consistent with the pecking order theory as well as with the static tradeoff model (theory explanations are below). In studies in this field in general, size is a commonly used as a control variable. The sign of the variable in most studies is found to be negative. Lower cash holdings are usually expected in larger firms as these firms are expected to obtain financing easier. Bigelli et al. (2007) similarly investigate the variables that determine the size of a firms' cash holdings (albeit private Italian firms), and assert that larger firms are commonly more diversified, which causes these firms to have lower risks in general, lowering their need for cash holdings. Though size is expressed as the natural logarithm in most studies, some studies use the natural logarithm of the total sales. D’Mello, Krishnaswami, and Larkin (2008) study 154 spin-offs in the 1996 - 2000 period to find which variables, used in the trade-off theory, are of importance to the decision about cash decisions in these firms. They find that leverage decreases with firm size, increases with the R&D expenses ratio, decrease with the working capital ratio and decrease with leverage.

Another variable frequently included in research attempting to find the determinants of the levels of corporate cash holdings is risk. This variable is generally expressed as the volatility of cash flows. The common intuition is that riskier firms hold more cash due to unexpected financial downfalls they may face. The sign associated with risk is in most studies therefore is positive. Precautionary motives are usually used for explanation along with this theory. A third variable often included is growth

opportunities. Firms that are sensitive to investment opportunities are more likely to have higher levels of cash holdings as these will, as mentioned above, permit them to finance them when they become due at unexpected times. A final variable worth mentioning is the cash conversion cycle (CCC). The CCC determines at what pace a firm can generate cash. Firms with longer CCC, ceteris paribus, are generally expected to have higher levels of cash holdings, as cash is generated at more infrequent intervals, as opposed to firms that have shorter CCCs (Opler et al. 1999; Bigelli, Sanchez-Vidal, 2007).

5 Mahrt-Smith, and Servaes (2003) examine firms on an international level and use agency theory to explain the level of cash holdings. The use a sample including over 11.000 firms from 45 countries and find that corporations in countries where shareholders are said to be protected poorly, hold up to two times the amount of cash in relation to firms in countries that are said to have stronger shareholder protection. Another important finding they come up with is that the generally used determinants of corporate cash holdings, such as investment opportunities, have much less power in the determination of corporate cash levels in countries where the shareholder protection is deemed to be lower. Finally, perceived excess cash that is not earning any interest and may just be there in the interest of managers, as argued by Jensen (1986). Managers can use this excess cash to invest in projects that have a net present value in the short term or use it to increase dividends. Either method will likely increase the share price, to which the compensation of the manager is frequently directly tied, causing the aforementioned decisions solely to be in the interest of the manager. As has become apparent, much research has been completed on the different facets of managing corporate cash levels and the determination thereof. On the contrary, little research focuses on the leverage level of a firm and its relation to these cash levels. The aforementioned studies find a linear relation between cash holdings and leverage.

I test the availability of a non-linear relation between cash holdings and leverage, that can be defined by a quadratic model. I furthermore test whether that the relationship becomes stronger as the leverage ratio tends to one, as the precautionary motive starts to play a higher role, with higher levered firms having higher chances of financial distress, and therefore marginally higher cash ratios. My contribution to this specific topic lies in the more recent time period in which I have studied the firms, which is from 2005 to 2010. In addition to that, I use a different estimation approach to account for the endogeneity problem than the one used by Guney, Ozkan, Ozkan, (2007). The results of this study show that there is limited evidence for this non-linear relation. Moreover, I find

reasonably statistically strong evidence in support of the negative relation between cash levels and leverage.

6

2

Literature

The main focus of this paper is on the leverage variable. Although commonly involved in other studies, this firm-specific variable is often hardly looked at in detail. Leverage has an influence on the cash levels for several reasons. One important reason, as described by Guney, Ozkan and Ozkan (2007) is that borrowings can be used as a substitution for holding cash. John (1993), considers the fact that debt can be used as a substitute for cash. This has prompted Guney, Ozkan and Ozkan (2007) to argue that there would be a negative relationship between cash holdings and leverage. In other words, firms that have relatively more debt will be more likely to have lower cash holdings, as the debt will partly be used for short-term liquidity purposes in the same way that cash would. Ozkan, Ozkan and Guney (2007) are one of the first researchers to analyze and quantify the relationship between cash holdings and leverage in detail. As mentioned earlier, they hypothesize this relationship to be non-linear, asserting that both the precautionary motive as well as the substitution effect are at play, and that the relationship between the level of cash holdings and leverage is not linear, but rather depending on the position of the firm with regards to its current capital structure. Investigating levels of corporate cash holdings of five major countries in the period of 1996 to 2000, they provide significant evidence for the non-linearity of this relationship. The substitution effect entails a negative relation between levels of cash holdings and leverage, but meets an inflection point, where the relationship becomes positive. At this point, the precautionary effect will play a role and higher levels of leverage will positively influence the level of cash holdings. There are, however, more theories that would suggest a certain relation between cash holdings and leverage. The trade-off theory deals with marginal benefits and marginal costs, from which an optimal cash level can be derived. In general, riskier or small firms, firms with long cash conversion cycles, or firms with high costs of hedging have optimum cash levels that are shifted more to the right. The sign of leverage in this theory is therefore bidirectional. The pecking order theory is, put shortly, about the order of financing. Leverage has a negative sign in this theory. If the investment in a year is larger than the change in retained earnings, the leverage ratio will automatically be

8

Table I A short summary of the theories that are available with regards to the relation between cash

holdings and the leverage variable, with in the first column the name of the theory, the second column the studies in which theory in relation to cash management is discussed, in the third column the sign that the leverage variable takes, and in the fourth column a brief explanation for the relevant sign.

Theory Mentioned in inter alia Sign of leverage Explanation

Substitution theory

Ozkan, Ozkan and Guney

(2007) negative

debt can used as a substitute for cash

Precautionary theory

Ozkan, Ozkan and Guney

(2007), Keynes (1934) positive

an increase in leverage increases chances of financial distress, leading to a need of higher cash buffers

Trade-off theory Baumol (1952), Opler et al.

(1999) bidirectional

the costs and benefits of holding cash are weighed against each other

Pecking order

theory Meyers and Majluf (1984) negative

if change in investment is larger than increase in retained earnings, leverage ratio automatically increases, causing a negative relation

Agency theory

Dittmar, Mahrt-Smith, and Servaes (2003), Mustapha and Huey Chyi (2012), Harris & Raviv (1991)

no concensus -

Free cash flow theory

Mustapha and Huey Chyi

(2012), Shenoy & Koch (1996) negative

increased monitoring in higher levered firms causes there to be less excessive cash

9

3

Hypothesis development

I have reviewed several theories as well as the influence that leverage is predicted to have in those theories, with regards to the level of cash holdings. Putting forward the argument that the

relationship between cash holdings and leverage becomes stronger as the leverage ratio tends to one due to the effects of the precautionary theory, such as previously researched and confirmed by Guney, Ozkan and Ozkan (2007), I formulate the following two hypotheses.

Hypothesis 1 - the influence of leverage: The first hypothesis regards the linearity of the relation

between corporate cash holdings and leverage. The null hypothesis states that, controlling for other firm-level influences, no relation exists between cash holdings and leverage. Whether or not this hypothesis is rejected, I will further investigate the presence of a non-monotonic relation between these two variables.

Hypotehesis 2 - the linearity of the relationship: The second hypothesis deals with the

non-linearity of this relationship. A quadratic equation model will be in place. If the null hypothesis can be rejected, my research results could point out to a non-linear relationship being existent.

4

Model

In the first hypothesis, the significance of the variable leverage is tested. As endogeneity is expected to be a relevant issue in this research (to be explained in more detail later), I will run a two-stage least-squares regression model. The first hypothesis will be tested using the following model:

(I)

(II)

where the cash ratio of firm at time is used in obtaining an estimation for the coefficient and , the latter being the coefficient attached to the leverage ratio of firm at time . Hypothesis one is rejected when the coefficient is non-zero and statistically significant. A disturbance term , per firm and at the time of is finally included, which can be further decomposed in (II) an individual entity effect of firm and , which denotes the remainder disturbance, the latter varying per

10 I similarly test the second hypothesis with a two-stage least-squares estimation technique. In more detail, I estimate the following model:

(III)

(IV)

where the cash ratio of firm at time is used in obtaining an estimation for the coefficient and , the latter being the coefficient attached to the leverage ratio, squared, of firm at time . Similar to the previous model, the error term is composed of a firm specific and remaining disturbance

term, respectively. Coefficients will show the change in the leverage ratio level per one

unit increase of cash ratio. Hypothesis two is rejected when coefficient is non-zero and statistically significant.

4.1 Definition of the variables

The dependent variable, cash ratio, is defined as cash plus cash equivalents, all divided by total assets. This is a definition also used by Opler et al. (1999) and Ozkan and Ozkan (2004). Other

researchers choose to exclude cash equivalents. Bigelli and Sanchez-Vidal (2007) use three measures of cash holdings, including the measure "pure cash", which excludes cash equivalents. I use the measure cash plus cash equivalents as this appears to be the definition of corporate cash holdings most commonly used in this topic. Using a different measure can take away from the comparability of my results to the results of other studies. For the same reason, I proxy for a firm's degree of leverage by dividing the long-term plus short-term debt over total assets (Opler et al.,1999; Guney, Ozkan and Ozkan, 2007). Instead of including certain control variables that have previously been found by some studies to be of influence of the level of cash holdings, I choose to omit these and use firm-fixed effects. These control variables omitted are firm size, usually measured by the natural logarithm of total assets or sales, growth opportunities, cash flow volatility, and others. The reason for exclusion is that not all research in this field is concurring with regards to which definite

determinants of corporate cash holdings exist.

5

Methodology

11 problem, I use three instrumental variables. The selection argument along with their definition are discussed in the following sections. I additionally estimate these two models on a dataset composed for the purpose of sensitivity analysis.

5.1 Endogeneity

The previously discussed, frequently referred to study by Opler et al. (1999) includes a suggestion to control for a possible endogeneity problem. They mention in their paper that the determinants of cash holdings are very much related to the determinants of leverage, and suggest that future research should investigate this more closely from both an empirical as well as a theoretical perspective. Later research, such as that by D’Mello, Krishnaswami, and Larkin (2008) have already adapted their methodologies to account for this possible estimation problem. As the main variable of focus in my research, leverage, may be endogeneous, I will use instrumental variables. This ensures that the variation of the disturbance term on the right hand side of the equation is in no way correlated to the depend variable, as would be the case if endogeneity would be permitted. An alternative method to use in the case of endogeneity would be to use lagged variables, as has been done by Jiminez, Lopez, and Saurina (2009) in their research on the topic of the use of credit lines by corporations.

5.2 The determination of endogeneity - testing the exogeneity of the regressors

In order to test the eventual presence of endogeneity among the regressors, a commonly known (Durbin-Wu-)Hausman test will be applied. The results of this test will yield an F-statistic, with a null hypothesis in place hypothesizing that all coefficients are simultaneously equal to zero. I similarly conduct this test manually. More particularly, I run two regressions: the first regressions has the regressor that is suspected to be endogenous as the dependent variable, and the other regressors and instruments on the right hand side of the equation. The residuals are consequently "freezed" using the EVIews software package. These residuals are used in the second regression, where the original regressions is ran using OLS, this time including the residuals from the previously estimated first-stage regression. The coefficient of the residuals obtained from this first-stage regression are the determinant of the results of this test: they are not significantly different from zero under the null-hypothesis, asserting that the OLS coefficients are consistent and the suspected variable is not endogenous.

5.3 Defining the instrumental variables

12 number of studies. Tangible assets are, in public firms at least, logically easier to assess in value and are expected to have a higher market value than intangible assets. This in turn leads investors to lend money to firms with tangible assets more quickly, which contributes to higher levels of firm leverage, as is commonly argued. Antoniou et al. (2008) study firms in both market-operated economies such as the U.K. and the US, as well as bank-oriented companies, like Germany, Japan and France. The method they apply is a two-step-system GMM. Their period of study reaches from 1987 to 2000, including 4854 firms. They find that, among other variables, asset tangibility positively affects a firm's leverage ratio. De Jong et al. (2008) study the same effect in firms in 47 countries, attempting to find the country-level and firm-level determinants of leverage. They, however, use a larger sample that includes 42 countries, with 12,000 firms with varying types of legal entities, recorded in a period reaching from 1997 to 2000. Using ordinary least square (OLS) estimation techniques, they similarly find that tangibility has a positive effect on leverage. Using only US publicly traded firms and a sample period that is of a completely different magnitude (1950 - 2003), Frank and Goyal (2009), use OLS and similarly find that leverage is positively affected by asset tangibility. Other research that confirms this positive relation comes from Qui and La (2010) and Gungoraydinoglu and Ötzekin (2011). I conclude that asset tangibility would be a well-suited instrumental variable as much research proves a significant relation to be present between asset tangibility and the leverage ratio. No relation has however been found between the level of cash holdings and asset tangibility, the latter being the main requirement for the selection criterion of an instrumental variable. Similar to Gungoraydinoglu and Ötzekin (2011), Kayo and Kimura (2011) and De Jong Kabir and Nguyen (2008), I proxy for asset tangibility by taking the ratio of net fixed assets to total assets.

13 a corporation's cash holdings. Either way, the instruments used will be tested on their validity (to be discussed in the next section). Similar to Byon (2008), I estimate the proxy for distance from

bankruptcy by calculating the following formula: (3.3 x EBIT) + sales + 1.4 x retained earnings + 1.2 x working capital, all over total assets.

Finally, I will use internal cash flows as an instrumental variable for leverage. I argue that the degree of the ability of a firm to generate cash flows is not directly related to the cash level on the debit side of a firm, seen from an accounting perspective, making it an adequate instrumental variable. The relation of this variable to leverage is often hypothesized to be positive or negative. The signaling theory implies a positive relation, where firms that have higher cash flows often have higher leverage ratios to signal aspects of their performance. According to the pecking order theory, there should be a negative relation between internal cash flows and leverage. There are several reasons to motivate this relation. The most common explanation concerns the need for cash, where firms that have a higher cash flow have a lower need for debt (Shenoy & Koch, 1995). Another motive for the aforementioned negative relation has to do with the agency problem. This problem causes a firms' shares to often be undervalued, resulting in a firm having to depend on internal financing more heavily. When the cash flows thus reach very low levels, a firm is better off issuing debt rather than using financing from the more expensive counterpart, the equity market (Mustapha and Huey Chyi , 2012). This negative relationship has been confirmed by several studies, including one by Mustapha and Huey Chyi (2012), Shenoy & Koch (1996), who find evidence consistent with the pecking order theory, and Harris & Raviv (1991). I calculate the internal cash flows by taking the sum of net cash flows, amortization, depreciation and depletion, and dividing it all by total assets.

5.4 The relevance of the instrumental variables

It is commonly said that the TSLS estimation is only as good as the instruments. Just as the regressors on the right hand side of the TSLS equation are tested on the exhibition of endogeneity, so are the instrumental variables that are in place, once it has been suspected that endogeneity indeed applies. The J-Test, also known as Sargan test, is commonly used to test for overidentifying restrictions. Under this test and assuming that at least one of the IV's is exogenous, the test will yield results with regards to the exogeneity of the instruments used. It must be noted that this test can only be executed when there are more instruments available than regressors, which will be the case in this study. This test too has to be calculated manually, when using the EViews statistical software

14 Using this statistic, called the "Sargan", a p-value can be calculated under a Chi-square distributions, using the number of instruments minus the number of regressors as the degrees of freedom. If the test statistics exceeds either of the critical values, using a 10% rejection region, the null hypothesis is rejected, meaning that the instruments are deemed to be invalid.

6

Data description

My data set comprises exclusively out of US publicly traded firms extracted from the time period 2005 to 2010. Unlike the study of Guney , Ozkan and Ozkan and (2007), on which this research is partly based, this research does not have an international scope. The choice as to exclude this international focus has to do with the exclusion of variables controlling for country-specific factors in this study. Constricting this research to the US market thus solves the influence of country-specific factors, such as the type of market a firm is based in (bank economy or capital economy). The time period from which the data has been recorded includes a period of a general bull market as well as a hausse in the equity markets.

I use a systematical method for creating my dataset. I use a top-down approach, with the help of the Orbis software, a program produced by Bureau Van Dijk (BVD). In this method, I start off with viewing the data for the entire population of firms worldwide, then adding certain constraining criteria to narrow down the sample systematically in order to produce one including solely the firms that I wish to be included. These criteria as such lead my sample to include only listed firms from the US, active in the period of 2005 until 2010, and which have a certain standard industrial classification (SIC) code. There is a reason for excluding firms active in a certain industries. Firms with SIC codes 6000 to 6999 and 4910 to 4939 are classified as financial firms and utility firms, respectively. According to Opler et al. (1999), these firms are subject to statutory capital requirements. Including such firms would yield biased estimates, I therefore exclude firms with such SIC codes from my dataset. I start out with 2162 firms in 2005. This number becomes smaller in the consecutive years, as some firms drop out of the sample as they have become inactive. The total amount of

observations adds up to 9835, spread over a six-year period.

15 Table III shows the descriptive statistics of the dataset used for a sensitivity analysis. In this dataset, extreme observations in the instrumental variables have been omitted, using the boxplot principle of erasing observations that are surpassing the critical value past 1,5 times the length of the

interquartile range (the range composing 50% of the observations, between the 25th and 75th percentile)above and below the median observation.

Table II Descriptive statistics of the panel dataset (n=8131) that I have used. The column on the left

side shows the variables included in the regression model. The cash ratio is taken by summing the cash and cash equivalents, and consequently dividing these by total assets. Leverage is calculated by adding short term to long-term debt, dividing both over total assets. Internal cash flows are

calculated as the sum of net cash flows, amortization, depreciation and depletion, all over total assets. The distance from bankruptcy proxy is a variant of the Altman Z-score, calculated by summing the results of the following calculations: the earnings before interest and taxes/total assets ratio multiplied by 3.3, the sales/total assets ratio, the retained earnings/total assets ratio multiplied by 1.4, and the working capital/total assets ratio multiplied by 1.2. Asset tangibility is defined as the ratio of net fixed assets to total assets. In the first row, the kind of descriptive statistic is listed, with from left to right the arithmetic mean, the standard deviation, the minimum observation, 25th percentile observation, the median, the 75th percentile observation and the maximum observation.

16

Table III Descriptive statistics of the panel dataset (n=6374) that I have used. This panel dataset has

been used for a sensitivity analysis, as observations of the instrumental variables outside the critical value of 1.5 times the length of interquartile range (the range composing 50% of the observations, between the 25th and 75th percentile)above and below the median observation have been removed. The values have been derived in the same way as has been done in Table II.

Mean S.D. Minimum 25th percentile Median 75th percentile Maximum Cash ratio 0.30 0.26 0.01 0.08 0.23 0.46 1.00 Leverage 0.17 0.20 0.00 0.00 0.11 0.28 1.00 Internal cash flows 0.03 0.16 -0.58 -0.02 0.07 0.13 0.23 Distance from bankruptcy 1.84 2.32 -7.37 1.04 2.40 3.40 4.57 Asset tangibility 0.19 0.18 0.00 0.06 0.14 0.28 0.73

7

Regression results

Tables IV shows the results of the regression analysis of models 1 and 2. The results of the first hypothesis, where a relation between cash holdings and the variable leverage ratio is tested, is shown in the second column. The results of the second hypothesis, examining the linearity of the relation between these two variables is shown in the third column. Table V similarly reports these results using the dataset in place for the purpose of sensitivity analysis.

7.1 Relation between cash holdings and leverage

The results show that, controlling for firm-fixed effects, the relation between leverage and cash holdings is significantly negative. The amount of observations is 8131. The average value that the cash holdings would take if all of the leverage ratio values were zero, i.e. the intercept, is 0.4718. This value is statistically significant at the 0.01 (one-tail) significance level. There is overwhelming

17 for the leverage ratio becomes smaller in absolute value (0.4246), yet is still significant. This implies a negative relation between cash holdings and leverage, with a increase of a 0.1 unit in the leverage ratio decreasing, on average, the cash ratio with 4.25%. The intercept in the latter regression does not differ much (0.3766) from the first model's intercept and is similarly still significant. The first model estimated on the main dataset has a slightly lower R-squared and adjusted R-squared ratio of 0.68 and 0.55, respectively, versus the 0.73 and 0.61, respectively determined in the dataset used for sensitivity analysis. This means that the model using the sensitivity analysis dataset with outliers omitted fits better. Overall, these slightly mixed results lead me to conclude that the relationship between leverage and cash holdings is negative and both statistically as well as economically significant.

7.2 Testing for non-linearity

The second model applied is used to test the non-linearity of between the cash ratio and leverage ratio. Tables III and IV similarly report the estimations of this model on the main sample as well as the sample used for sensitivity analysis. Here too the results of the same model estimated on the two different datasets vary. To start off with the most important results, it appears that the squared leverage variable is negative and equally significant as the normal (non-squared) leverage variable estimated in the previous model. The coefficient is however larger, taking on the value of -1.1622. This means that for every tenth unit increase in leverage the cash ratio decreases with 11.62%. The results of the estimation of this coefficient in the regression model ran on the sensitivity analysis dataset is less pronounced, where the value taken stands at -0.6517. Here too, the r-squared and adjusted r-squared scores are higher of the second model estimated on the sensitivity analysis dataset (0.72 and 0.60 versus the 0.66 and 0.52 found in the regressions results of the main sample). I compare the model including a squared leverage term with the first model that has a normal

leverage term. As all pertinent coefficients are highly significant, I look at the R-squared and adjusted R-squared score to determine which model first better. Model 2 is inferior with this respect to the first model tested, both in the regression ran on the main sample as well as on the sensitivity analysis sample. Although the relationship between cash holdings and leverage could be a quadratic one, a linear model seems to fit slightly better.

7.3 Instrumental variables

18 main sample and the sensitivity analysis sample respectively, and both having a p-value of 0 to the fourth decimal. This means that there is overwhelming evidence for the presence of endogeneity, and the instruments therefore correctly being in place.

The Sargan test statistic is used to validate the legitimacy of the instrumental variables, i.e. if the instrumental variables are indeed exogenous like they are supposed to be. The p-values of a Chi-Square distribution of the Sargan J-Test statistic, for the main results are both zero to the fourth decimal, meaning that the validity of the instruments are rejected. In other words, I cannot draw conclusions from the results of the estimation on the main sample. For the purpose of sensitivity analysis, I have used the aforementioned sensitivity analysis sample, where observations beyond a certain point (outliers) have been removed. The respective p-value of a Chi-square distribution for the Sargan J-Test statistic in the sample used for sensitivity analysis are 0.0126 (model 1) and 0.0146 (model 2), which are still outside the rejection region of 1%. In other words, the results of the estimation on the sensitivity analysis dataset can be used to draw conclusions as the instruments are valid. This allows me state that I find a negative relationship between the cash ratio and the leverage ratio, with the coefficient value of -0.4246, which means that for every tenth unit of increase in the leverage ratio, the cash ratio decreases, on average, with 4.25%. The coefficient of the squared leverage term in the second model is found to be -0.6517, which means that with every tenth unit of increase in leverage, the cash ratio decreases, on average, with 6.52%. While consistent with the first model and the coefficient being equally significant, the first model does have a slightly better

adjusted R-squared score of 0.61 instead of the adjusted R-squared score of 0.60 in the second model. Although this difference is economically insignificant, I conclude that, per strict definition, the relation between cash holdings and leverage is linear and negative, in congruence with the

19

Table IV TSLS regression results for the main sample. Columns II and III represent the findings of the

estimated models as formulated by hypothesis 1 and 2, respectively. The p-values are shown below the coefficient, in parentheses. The dependent variable, the level of cash holdings, is defined as cash and cash equivalents over total assets. Leverage is calculated by adding short term debt and long term debt together, dividing this by total assets. The probability of the F-test statistic of both regressions takes on a p-value of zero to the second decimal, meaning that the null hypothesis stating that all slope parameters are jointly zero can be rejected. The distributions of the residuals of both regressions have a Jarque-Bera score with a p-value of zero to the fourth decimal, meaning that there is overwhelming evidence to reject the null hypothesis that the distributions are non-normal. The Hausman test has been manually applied on model 1, the coefficient of the first stage residuals is non-zero (-3.37) with a p-value of zero to the fourth decimal, meaning that there is overwhelming evidence of the presence of endogeneity. The Chi-square p-values for the Sargan statistics with 2 degrees of freedom are 0.00003223 and 0.00004031, respectively, meaning that the instruments are invalid.

(I) (II) (III)

20

Table V TSLS regression results obtained from dataset used for the purpose of sensitivity analysis.

Columns II and III represent the findings of the estimated models as formulated by hypothesis 1 and 2, respectively. The p-values are shown below the coefficient, in parentheses. Leverage is defined as short term plus long term debt, divided by total assets. The probability of the F-test statistic of both regressions takes on a p-value of zero to the second decimal, meaning that the null hypothesis stating that all slope parameters are jointly zero can be rejected. The distributions of the residuals of both regressions have a Jarque-Bera score with a p-value of zero to the fourth decimal, meaning that there is overwhelming evidence to reject the null hypothesis that the distributions are non-normal. When manually applying the Hausman test on the model 1, the coefficient of the first stage residuals is non-zero (-4.15) with a p-value taking on 0 to the fourth decimal, meaning that there is

overwhelming evidence of the presence of endogeneity. The Chi-square p-values with two degrees of freedom for the Sargan statistics are 0.0126 and 0.0146, which do not surpass the thresholds of the rejection regions when using a 10% rejection region, meaning that the instruments used are valid.

(I) (II) (III)

n 6374 6374 Intercept 0.3766 0.3486 (0.0000) (0.0000) Leverage -0.4246 - (0.0007) Leverage squared - -0.6517 (0.0007) R-squared 0.73 0.72 Adjusted r-squared 0.61 0.60

8

Summary and conclusions

21 I find evidence that, controlling for firm-fixed effects, leverage is statistically significant and

negatively related to the level of cash holdings. The results pertaining to the first hypothesis are in favor of the substitution theory, the pecking order theory, the trade-off theory and/or free cash flow theory.

My findings do not corroborate with the results of Guney, Ozkan and Ozkan (2007), as I cannot find evidence in favor of the rejection of linearity of the relationship between cash holdings and leverage. As I find no evidence for a positive relation between the two variables, I cannot find support for the existence of the precautionary theory. My evidence thus does not provide support for the

assumption that the relation between cash holdings and leverage is non-monotonic. The results of my study therefore differ from that of Guney, Ozkan and Ozkan (2007) both on the form as well as on the substance of the relationship between corporate cash holdings and leverage.

9

Limitations and further research

22

References

Antoniou, A., Guney, Y., Paudyal, K., 2008. The Determinants of Capital Structure: Capital Market-Oriented versus Bank-Market-Oriented Institutions. Journal of financial and quantitative analysis 43, 59 – 92. Baumol, W.J., 1952. The Transactions Demand for Cash: An Inventory Theoretic Approach. The Quarterly Journal of Economics 66, 545 – 556.

Bigelli, M., Sanchez-Vidal, J., 2012. Cash holdings in private firms. Journal of Banking and Finance 36, 26 – 35.

Byoun, S., 2008. How and When Do Firms Adjust Their Capital Structures toward Targets? The Journal of Finance 63, 3069 – 3096.

Davidson, R., MacKinnon, J.G., 2004. Econometric Theory and Methods. Oxford University Press. D’Mello, R., Kirshnaswami, S., Larkin, P.J., 2008. Determinants of corporate cash holdings: Evidence from spin-offs. Journal of Banking and Finance 32, 1209 – 1220.

Denis, D.J., Sibilkov, V., 1998. Financial Constraints, Investment, and the Value of Cash Holdings. The Review of Financial Studies 23, 247 – 269.

De Jong, A., Kabir, R.B., Nguyen, T.T., 2008. Capital structure around the world: The roles of firm- and country-specific determinants. Journal of Banking & Finance 32, 1954–1969.

Dittmar, A., Mahrt-Smith, J., Servaes, H., 2003. International corporate governance and corporate cash holdings. Journal of Financial and Quantitative Analysis 38, 111 – 133.

Faulkender, M., Wang, R., 2006. Corporate Financial Policy and the Value of Cash. The Journal of Finance 61, 1957 – 1990.

Foley, C.F., Hartzell, J.C., Titman, S., Twite, G., 2007. Why do firms hold so much cash? A tax-based explanation. Journal of Financial Economics 86, 579 – 607.

Frank, M.Z., Goyal, V.K., 2009. Capital Structure Decisions: Which Factors Are Reliably Important? Financial Management 38, 1 – 37.

23 Gungoraydinoglu, A., Öztekin, Ö., 2011. Firm- and country-level determinants of corporate leverage: Some new international evidence. Journal of Corporate Finance 17, 1457 – 1474.

Harford, J., Mansi, S.A., Maxwell, W.F., 2008. Corporate governance and firm cash holdings in the US. Journal of Financial Economics 87, 535 – 555.

Harris, M., Raviv, A., 1991. The Theory of Capital Structure. The Journal of Finance 46, 297 – 355. Jensen, M., 1986. Agency costs of free cash flow, corporate finance and takeovers. American Economic Review 76, 323 – 9.

John, T.A., 1993. Accounting Measures of Corporate Liquidity, Leverage, and Costs of Financial Distress. Financial Management 22, 91 – 100.

Keynes, J.M., 1934. The general theory of employment, interest and money. New York, Harcourt, Brace.

Jiminez, G., Lopez, J.A., Saurina, J., 2009. Empirical Analysis of Corporate credit lines. The Review of Financial Studies 22, 1 – 30.

Kayo, E.K., Kimura, H., 2011. Hierarchical determinants of capital structure. Journal of Banking & Finance 35, 358 – 371.

Kim, C.S., Mauer, D.C., Sherman, A.E., 1998. The determinants of corporate liquidity: theory and evidence. Journal of Financial and Quantitative Analysis 33, 305 – 334.

Miller, M.H., Orr, D., 1966. A model of the demand for money by firms. Quarterly Journal of Economics 80, 413 – 435.

Mustapha, M., Huey Chyi, N., 2012. Firm Leverage-Cash Flow Determinants and Capital Structure Decisions in a Developing Economy. Global Economy and Finance Journal 5, 79 – 96.

Myers, S.C., Majluf, N., 1984. Corporate financing and investment decisions when firms have information that investors do not have. Journal of Financial Economics 13, 187 – 221. Opler, T., Pinkowitz, L., Stulz, R., Williamson, R., 1999. The determinants and implications of corporate cash holdings. Journal of Financial Economics 52, 3 – 46.

24 Qui, M., La, B., 2010. Firm Characteristics as Determinants of Capital Structures in Australia.

International Journal of the Economics of Business 17, 277 – 287.