Tax system differences and their impact on capital structure

(1)

Tax system differences and their impact on

capital structure

by

Frederike Veenstra

1

University of Groningen

Faculty of Economics and Business

MSc. Finance and MSc. Fiscale Economie

June 2014

Supervisor: dr. P. P. M. Smid

1_{Email address: frederikeveenstra@gmail.com, studentnumber: 1907468.}

(2)

Abstract

Building on theoretical work initiated by Modigliani and Miller (1963), we try to measure the impact of tax related variables on the leverage within a firm. We use data on 2.367 companies across 71 countries for the period 2003 to 2013 and examine the effect of differences in the statutory corporate tax rate and the effective tax rate on debt ratios. The novel feature of this contribution is the quantification of carry back and carry forward periods and the fiscal freedom of a country on corporate debt ratios. We find that a rise of the statutory corporate tax rate as well as the carry back period will increase the firm’s debt ratio, whereas a higher degree of fiscal freedom lowers the ratio. We also report that when a change in one of the tax determinants occurs, most likely companies in the non-financial industries will react more to that change than firms in the financial sector. Also, the tax variables have a stronger effect on the long term debt ratio (measured as the long term debt over total assets). Tax variables seem to have less impact in the post-crisis period (2008 – 2013) compared to the pre-crisis period (2003 – 2007). Lastly, the capital structure of companies situated in Europe is more influenced by tax variables than their counterparts in Asia and North America.

Keywords

Capital structure, Leverage, Tax, Carry back period, Carry forward period, Fiscal freedom

JEL classification

(3)

1. Introduction

As of 2008 a financial crisis roamed the financial market. This crisis has prompted governments to take measures to assure companies could survive the economic downturn. Often these measures were introduced in the tax field. For example, the Netherlands introduced a temporary lengthening of the carry back period whilst shortening the carry forward period so companies could compensate their current losses with past profits to reduce taxable income2_{. However a potential negative side} effect of these measures is that firms may increase their debt ratio as higher interest payments reduce taxable income even further, allowing them to benefit more from the tax advantages of debt financing. Hence these policy measures could increase the risk borne by companies. More specifically with respect to the financial sector, given the desired policy objective to decrease leverage ratios of financial institutions (in line with Basel I, II and III), it is important to know if and how these changes in the tax field impact the capital structure of firms. Increased debt ratios in the financial sector may threaten financial stability and hamper economic recovery. This study investigates to what extent taxes and tax related variables influence firms’ decisions regarding their capital structure.

To completely fathom the choices companies make regarding their capital structure when changes in legislation occur, we have to start at the core. Optimality is something companies have always strived for. From the little bakery around the corner trying to maintain its course of business to the large multinationals trying to maximize their shareholder’s profits. There are a lot of ways reaching the goal of profit maximization, such as optimizing the capital structure of a firm. The first to recognize that firms could enhance their firm value with the inclusion of leverage were Nobel laureates Modigliani and Miller (1963). This is due to the debt tax shield, the advantage of interest payments over dividends, because these are deductible for tax purposes. Their research was the underlying of many more studies in the theoretical field as well as the empirical field, as economists are yearning for the answer as to why many firms

(6)

Page | 6 do not optimally gain from the benefits of debt. As Ju et al. (2005) put it nicely: “A central issue in corporate finance research is the question of why firms have fairly low leverage ratios, despite the large tax advantage enjoyed by debt.” (p. 259). We aim to contribute in this area and gain further insight regarding this capital structure puzzle and in particular what role there is for tax and tax related variables in this matter.

The question of Ju et al. (2005) gives rise to two important theoretical streams trying to identify the reason(s) why firms deviate from an optimal leverage ratio. The so called trade-off theory and the pecking order theory. The trade-off theory, as the name suggests, weighs the benefits of newly attracted external financing against the increased bankruptcy cost associated with the new leverage (Kraus and Litzenberger, 1973). An adjustment of the firm’s debt-equity mix is motivated by this trade-off. Another somewhat contradicting theory regarding the capital structure is the pecking order theory. This theory suggest that the capital structure is being determined by a manager’s preferential order of funding opportunities (Myers, 1984). When an investment opportunity arises, managers prefer to use retained earnings. If retained earnings are not available, external funding is attracted. As a last resort, a firm will issue new equity to satisfy the need for capital.

With these theories, a lot of researchers try to identify the determinants of the capital structure. Firm size, asset tangibility, profitability and growth opportunities are just a few of the determinants that are considered to be of importance in empirical research [see for example Titman and Wessels (1988) and Fan, Titman and Twite (2012)]. In line with theory, also tax and tax related variables have proven to (partly) explain the debt level in a firm. For example, a higher statutory tax rate (Faccio and Xu, forthcoming) leads to relatively higher debt ratios. The same is true for the effective tax rate (Huang and Song, 2006). An increase of non-debt tax shields (NDTS) will lead to lower leverage within a firm (Kim and Sorensen, 1986).

(7)

Page | 7 changes in (country specific) tax variables on the debt level of firms, so governments as well as companies can anticipate the effects that a change in tax legislation may induce. In this paper we study, among other things, the effects of the tax loss compensation periods on the capital structure of a firm. Also the effect of Fiscal Freedom, the governmental tax burden on companies, on the debt level is explored. This is the first study we are aware of that analyzes the impact of these tax related variables.

This study contains a data sample of 2.367 companies within 71 countries for the years 2003 to 2013. Using the ordinary least squares regression model in an unbalanced panel setting, we find that the debt ratio in a firm will be higher when the statutory tax rate rises. Also an increase in the newly introduced variable carry back period leads to greater debt levels. The new determinant Fiscal Freedom proofs to be inversely related to the leverage within a firm indicating that the increase of governmental tax burden in a country increases the leverage of firms situated in that country.

Concerning the control variables used in our analysis and in line with the studies of Booth et al. (2001) and Fama and French (2002), we see that an increase of the size of a firm increases the firm’s leverage. The same relationship is seen with tangibility of assets and the capital structure. These results are both in line with the trade-off theory. A decrease of profitability reduces the company’s debt level, as was also suggested by the pecking order theory.

(8)

post-Page | 8 crisis period than in the pre-crisis period. Lastly, we report a stronger effect of the tax variables on the capital structure of firms situated in the European continent.

The remainder of this paper will have the following structure. First, section 2 discusses the relevant literature and outlines the theoretical framework. Thereafter an elaboration of the determinants of capital structure will be presented with the empirical evidence ending every subsection. In section 3 the methodology regarding this paper will be treated. The regression equation and its components will be discussed. The 4th_section_{contains all the relevant information regarding the data}

sample employed throughout this paper. Section 5 reports the results and interprets the outcomes. In this section also the robustness of the results is being investigated. In

(9)

Page | 9

2. Literature review

This section contains an overview of the relevant literature regarding the relationship between taxes and the capital structure. First, an overview of the theoretical literature regarding capital structure will be given. Thereafter the determinants of capital structure will be conveyed via the applicable theoretical and empirical research conducted on this subject.

2.1. Theoretical framework

When discussing the theory of capital structure, one must first mention the founding fathers of this topic, Modigliani and Miller. In 1958 they published a groundbreaking paper regarding the capital structure of a firm, which became known as the irrelevance theory. In this paper they suggest that the leveraging within a firm does not contribute to the value of the firm in a perfect market. In their complementary paper a few years later, Modigliani and Miller (1963) came to a different insight regarding the value of firms. They state that the value of the firm will be higher when the firm’s leverage is increased, all other things being equal. This is due to the advantage of interest deductions for tax purposes, the so called debt tax shield. The Modigliani and Miller theorem (MM theorem) was the underlying for many other research papers regarding capital structures. Although there has been an ongoing investigation of the determination of an optimal capital structure for a firm, there is still no univocal theory amongst researchers which perfectly explains the debt-equity choice of companies (Adeyemi and Oboh, 2011). Several other (contradicting) capital structure theories have been brought forward, such as the trade-off theory and the pecking order theory, which will be discussed in turn below.

2.1.1. Trade-off theory

(10)

Page | 10 effect the optimal capital structure in two different ways, as is argued by Fama and French (2002). First, when the personal tax rate on debt is higher, relative to the personal tax rate on equity, this pushes firms to employ less debt in their enterprise. Secondly, Miller and Scholes (1978) state that the difference between the constant marginal corporate tax saving and the constant marginal personal tax cost (negative or positive) causes a firm to alter their optimal leverage regarding the outcome to maximum leverage, no leverage or somewhere in between.

However, some of the assumptions regarding the trade-off theory, as brought forward by Myers (1984), are disputable (Frank and Goyal, 2007). For example, the target debt level has to be extracted from a structure which is different for every empirical research done on the subject. Also, the theory may be oversimplifying the tax code for research reasons instead of using the actual features of the tax code. Third, it is possible that the bankruptcy costs are transferable from one creditor to another rather than being a deadweight cost to the company as the theory now assumes. Lastly, transaction costs can only work in this theory if they take on a special form, hence not all kinds of transaction costs can be modeled. Therefore the theory is far from ideally explaining the capital structure within a firm. Nowadays, there are broadly three type of scholars interpreting the trade-off theory differently. These are the static trade-off theory, dynamic trade-off theory and the agency cost theory. They will be introduced in section 2.1.1.1 to 2.1.1.3.

2.1.1.1. Static trade-off theory

(11)

Page | 11

Frank and Goyal (2007) state: “The model has a solution for leverage, but there is no room in the model for the firm ever to be anywhere but at the solution.” (p. 145). They also point out the difficulty of merging the theory with dynamic features of tax codes, due to the one period underlying the static trade-off theory.

2.1.1.2. Dynamic trade-off theory

The foregoing problems have led to the development of the dynamic trade-off theory (Stiglitz, 1973; Kane et al., 1984; Brennan and Schwartz, 1984). This extension of the trade-off theory has some beneficial features which can give extra insight in the debt level decision process of companies. For example, one of the underlying assumptions of the model is that firms last for more than one period, hence dynamic features of the tax code can be incorporated, such as (changes in) the period of tax loss deductions and (changes in) the statutory tax rate (see for example Shevlin, 1987 or 1990). In addition the before mentioned mean reversion and retained earnings problems are being tackled with this multi-period model. According to the dynamic trade-off theory, the financing need that the company foresees in the next period will affect the firm’s financing decision of the current period (Frank and Goyal, 2007).

2.1.1.3. Agency cost theory

Another derivation of the trade-off theory is the agency cost theory (Jensen and Meckling, 1976; Myers, 1977). This theory is centered on the assumption that the minimization of the cost of conflicts between the principal and agent will result in an optimal capital structure. The principal and agent in this case are the stockholders of the company and its management, the so-called separation of ownership and control. This theory, as well as the static trade-off theory, supports that an optimal capital structure can be achieved with the debt-equity choice of the firm.

2.1.2. Pecking order theory

(12)

Page | 12 of financing for investment opportunities. The theory predicts that retained earnings will be first used, thereafter additional debt will be attracted and lastly equity will be issued to finance new investment opportunities. Different explanations as to why the pecking order theory holds exist. First of all, Stiglitz (1973) identifies an asymmetry in the taxation of money coming into the firm and money going out of the firm. This gives an incentive to finance investments as much as possible with retained earnings and otherwise with debt. Also, adverse selection can be a (partial) explanation of the pecking order theory. Adverse selection (Myers and Majluf, 1984; Myers, 1984) describes the phenomenon of managers of a firm knowing the true value of the firm and hence act upon this value. They would be willing to issue equity if they view their company as being currently overvalued by the market and would rather use retained earnings or issue debt if they perceive their firm as being undervalued by the market.

Frank and Goyal (2007) identify two characteristics common for the pecking order theory. First, the theory is in essence a simplistic model which contributes to the ease of understanding and interpretation of the effects. Also this theory can be used to interpret other theories or extend them. For example an agency theory can also contain a financing hierarchy due to differences in the level of agency cost per financing decision (Myers, 2003).

According to Myers (1984), the difference between the trade-off theory and the pecking order theory largely depends on the underlying assumption whether the amount of investment opportunities of a firm directly influences the company’s capital structure.

2.2. Capital structure and its determinants

(13)

Page | 13 2.2.1. Capital structure measure

In the academic literature regarding the capital structure puzzle, there is no consensus regarding the measure of the debt-equity mix of a company. Various proxies are used throughout different studies. Modigliani and Miller (1963) started by modelling the entire (leveraged) value of the firm as the capital structure proxy. This was also often done in other research (such as Bradley, Jarrell and Kim (1984) and Myers (1984)). Other capital structure measures were put forward to study what the capital determinants underlying the capital structure are. Titman and Wessels (1988), for example, regress the debt-to-equity ratio on their perceived determinants. They use various measures to see which one is the best proxy for a firm’s capital structure, which include short term debt, long term debt and convertible debt in the numerator and the book value of equity and the market value of equity in the denominator. However, the most often used capital structure measure is the debt ratio. Rajan and Zingales (1995) use the total debt ratio as a measure for the capital structure of a firm. This measure is the same as the before mentioned ratio of Titman and Wessles (1988), however total debt is now divided by total assets. The total debt ratio is the most common capital structure measure in the literature of the last two decades (see for example Booth et al. (2001); De Jong, Kabir and Nguyen (2008); Fan, Titman and Twite (2012)). Some use the book value of assets (Hovakimian, Opler and Titman, 2001) or the market value of the firm (Fan, Titman and Twite, 2012) or both (Fama and French, 2002).

(14)

Page | 14 needed with every issuance. Therefore a dynamic structure of the research is not necessary, which could cause extra difficulties as mentioned in the dynamic trade-off theory section. However, obtaining such information is very difficult.

2.2.2. Determinants of capital structure

The numerous empirical studies have led to a wide variety of determinants proven to be of significance in determining the capital structure of a firm. In this paper, the focus is on tax and tax related variables influencing the debt-equity choice of a firm. Hereafter, the effect of taxes on the capital structure will be further elucidated as well as other important control variables influencing the debt-equity mix of a company.

2.2.2.1. Taxes

The first predominant link between capital structure and tax was, as mentioned before, brought forward by Modigliani and Miller (1963). This led to further research in the taxation area where for example Miller (1977) shows that the advantage of deductibility of interest can be lost due to taxation at the personal level. Hence one should take all taxes into consideration when determining the optimal debt level.

DeAngelo and Masulis (1980) extend Miller’s framework by focusing on the marginal tax benefit of debt, because unlike the statutory tax rate, this marginal tax rate can change over time due to non-debt tax shields (such as depreciation and tax loss compensation). These non-debt tax shields (NDTSs), ceteris paribus, crowd out the incentive to use debt, because the NDTSs lower the taxable income of the firm, therefore the creation of extra interest deductions is not necessary. Kim (1989) also points out that firms cannot always fully benefit from all incremental interest deductions due to the fact that negative income is often not taxed and hence the interest deduction cannot be used to press taxation at the company level. In the same way, due to the limitation of tax loss compensation periods firms cannot benefit fully from loss deductions.

(15)

Page | 15 firms have a tax incentive to lease from high tax rate lessors, though this implication is only true for some combinations of tax rules and leasing arrangements” (p. 1083). This results in depreciation and interest deductions (if the purchase of the asset by the lessor was financed with debt) at the high tax rate lessor. Though, the leasing decision does not have to be solely based on taxation motives (Smith and Wakeman, 1985).

(16)

Page | 16 lower debt levels. This has been confirmed frequently in empirical literature (see Kim and Sorensen, 1986; MacKie-Mason, 1990; Huang and Song, 2006). However, also an increase in debt levels after an increase in NDTS has been encountered (Bradley, Jarrel and Kim, 1984; Titman and Wessels, 1988). Bradley, Jarrel and Kim (1984) therefore question the suitability of NDTS as a proxy for the tax shield of debt. They argue that the direct relation between NDTS and the leverage ratio could also reflect the ability to collateralize assets, the so called tangibility (see section 2.2.2.4). Graham (2003) argues that these different outcomes can occur due to the possibility of correlation of NDTS with profitability and investment. In earlier papers Graham (1996a and 1996b)

simulates the marginal tax rate to capture the dynamic effects of a country’s tax code (see also Shevlin, 1987 and 1990 for earlier simulations). This simulated measure also captures the influence of profitability, which is not captured by the “standard” NDTS as mentioned before. This method is also in line with the dynamic trade-off theory.

Graham (1996b) reports that an increase in these simulated marginal tax rates induce a rise in debt ratios.

Another method to test what the effects of taxes are on the capital structure of firms is to check in which tax legal system a company is situated. Fan, Titman and Twite (2012) test the hypothesis that the differences in tax legal systems have different effects on the magnitude of the debt ratio of a company. They divide the countries into three different legal systems (classical tax system, dividend relief tax system and dividend imputation tax system) and find that companies incorporated in a country where the tax gain from leverage is positive have, on average, more debt.

2.2.2.2. Profitability

(17)

Page | 17 advantage that arises when extra debt is attracted by a firm who is relatively more profitable. The pecking order theory, however, stipulates a contradicting connection. With the pecking order theory, internal funds are preferred over external funds. Hence, more profitable firms, ceteris paribus, should become less levered, because they are less inclined to attract new external funds (Frank and Goyal, 2007). Most often empirical evidence is delivered advocating the pecking order theory. For example Bradley et al. (1984), De Jong, Kabir and Nguyen (2008) and Fan, Titman and Twite (2012) find that the firm’s debt level decreases when a rise in profitability occurs. This evidence is contradicting the trade-off theory, as mentioned above. However, one should also take into account that the profitability of a firm also conveys information about its growth potential. The trade-off theory suggests that an increase in growth reduces the debt level. Therefore, one should be cautious when interpreting the predictions of the trade-off theory regarding profitability (Frank and Goyal, 2007). In combination with taxes,

Fama and French (2002) state that governments often levy taxable income more than they favor tax losses (due to limitations of tax loss compensation periods). This will lead to the higher earnings firms increasing their debt levels, because the expected payoff from deductible interest payments is higher for more profitable firms.

There are a lot of different proxies for profitability, for example operating income or net income over book value of total assets (De Jong, Kabir and Nguyen, 2008; Fan, Titman and Twite, 2012), a three year average of return on assets (Hovakimian, Opler and Titman, 2001) or operating income over sales (Downs, 1993). Notwithstanding the different proxies, they all present an inverse significant relationship between profitability and leverage, advocating the pecking order theory.

2.2.2.3. Size

(18)

Page | 18 diversified and accordingly have a lower chance of going into default, so external financing is easier attracted. Also, from an agency cost theory perspective, often large firms are more mature firms who have the advantage of having built up a (good) reputation which will lead to lower agency cost of debt (Frank and Goyal, 2007). The pecking order theory however, predicts an opposite effect. They state that an increase of firm size induces a decrease of leverage within a firm. This is due to lower adverse selection, because firms are more mature and are well known. However, large firms typically also have more assets which will induce more adverse selection. Therefore, the inverse relationship supported by the pecking order theory is somewhat equivocal. When looking at the empirical literature, the majority of researchers found an increase in leverage when an increase of firm size occurs (see for example: Booth et al., 2001;

Fama and French, 2002; Fan, Titman and Twite, 2012).

Several kind of measures have been brought forward to reflect the size of a specific firm. For example the natural logarithm of sales (Faccio and Xu, forthcoming) or the natural logarithm of total assets (Fan, Titman and Twite, 2012). They both report that an increase in the proxy for size induces an increase in the external capital within a firm. However, Fan, Titman and Twite (2012) also present one significant inverse relationship between size and the short term debt ratio. They attribute this to the high leverage ratios of one specific country (South Korea). Nonetheless the aforementioned result, they report that overall an increase in firm size also increases the leverage of a company.

2.2.2.4 Tangibility

(19)

Page | 19 This leads to lower agency costs for companies with a high level of tangible assets (Frank and Goyal, 2007). They also state that the agency theory as well as the static trade-off theory both expect that a rise of tangibility increases the debt level of a company. In the empirical literature these conclusions are endorsed by statistically positive coefficients for tangibility proxies (see for example: Rajan and Zingales (1995);

Booth et al. (2001) and more recent de Jong, Kabir and Nguyen (2008); Fan, Titman and Twite (2012)). From a pecking order theory perspective Harris and Raviv (1991) argue that an increase in tangibility of assets should reduce a firm’s debt level due to more asymmetric information when the level of tangible assets is low. This, as is argued, would lead to a higher debt level over time. They also state that the asymmetric information could lead to underinvestment which also contributes to, ceteris paribus, higher debt levels. However, this is not a very commonly shared opinion.

The most often used proxy to test the effect of tangibility of assets on the debt-equity mix in a firm is the net property, plant and equipment divided by the book value of assets (for example Rajan and Zingales, 1995). As mentioned above, they report an increase of leverage when the tangibility proxy rises. However, Frank and Goyal (2007) point out that the inclusion or exclusion of inventory in the measure of tangibility can explain deviations in levels of long and short term debt within an enterprise.

2.2.2.5. Growth

(20)

Page | 20 will increase the debt level when profitability is fixed, due to more investment opportunities, which need to be financed with debt (Frank and Goyal, 2007).

To check whether growth opportunities have an effect on the level of debt, different proxies are used throughout the literature. Most often the market-to-book ratio of a firm is applied (Barclay and Smith, 1995; Booth et al., 2001). They both detect a highly significant inverse relationship between the proxy and a firm’s leverage. Another example is from Titman and Wessels (1988), they used capital expenditure over total assets, the percentage change in total assets and research and development over sales to extract information regarding the effect of growth opportunities on leverage. They find both negative and positive coefficients for all the proxies when trying to determine the leverage ratio. However, they point out that their results do not provide solid evidence for an effect on a firm’s leverage. Overall, an inverse relationship between the growth opportunities of a firm and its level of debt is presented.

2.2.2.6. Firm age

A relatively new research conducted by Pfaffermayr, Stöckl and Winner (2013) found a new variable influencing the capital structure of a firm. An increase in firm age, measured as the date of incorporation or significant reorganization, is believed to reduce the level of debt in a company. This effect is assigned to the fact that older firms tend to have more internal funds (see for example Berger and Udell, 1998) which, according to the pecking order theory, are preferred over external financing. However,

(21)

Page | 21 2.2.2.7. Bankruptcy

As early as 1968, Altman showed that the probability of bankruptcy is influencing the capital structure of a firm. Bankruptcy is one of the key features underlying the trade-off theory, as this theory balances the deadweight cost of the probability of default with the tax shield savings from extra leverage to obtain the optimal capital structure of a firm (Myers, 1984). Therefore, a decrease in the debt level is expected when the probability of bankruptcy increases. This is also confirmed in the literature quite often (see for example Bradley, Jarrell and Kim (1984) and Graham (1996)).

Several proxies have been brought forward to capture the effect of bankruptcy probability on the debt-equity choice of a firm. Altman (1968) introduces a combined measure taking several (scaled) factors into account such as total assets, EBIT, sales, retained earnings and working capital. This is the so-called ZPROB, the Z-value for the firm’s probability of bankruptcy. Several other researchers have put together similar formulas measuring ZPROB (see for example the MacKie-Mason (1990) version in

section 3.1.3). He confirms the before mentioned inverse relationship between leverage and the probability of bankruptcy with this measure. Other proxies to capture the effect of bankruptcy probability are for example whether or not a specific bankruptcy code is enforced in a country, see e.g. Fan, Titman and Twite (2012), who report that the existence of a bankruptcy code in a country increases a firm’s leverage. Also the variability of the return on assets is used as a proxy for the bankruptcy probability (Booth et al., 2001). However, they report mixed effects of this measure.

2.3. Hypotheses

(22)

Page | 22 which the company is subject will be investigated as well as the company’s effective tax rate.

2.3.1. Independent variables

We examine whether differences in the statutory corporate tax rate have an effect on a firm’s debt ratio. The expected effect of an increase in statutory tax rate is that the debt level will increase, because the interest paid is deductible for tax purposes. In other words, the interest tax shield is more valuable. This is in line with the empirical findings mentioned in section 2.2.2.1.

The deductibility of net operating losses have been a subject in the capital structure literature for quite some time (see for example DeAngelo and Masulis, 1980). However, the research conducted often used the quantified amount which firms could deduct, the so called NDTS. In this paper, we study whether the difference in the duration of loss carry back and loss carry forward periods have an impact on the debt ratio of a firm. This measure captures one of the dynamic effects of a country’s tax system as is also in line with the before mentioned dynamic trade-off theory (section 2.1.1.2). The extension of the period of tax loss deduction (carry back or carry forward) will lead to higher debt levels, other things being equal, due to the possibility of settling losses with more periods of positive taxable income.

(23)

Page | 23 The effective tax rate, the actual tax rate the company perceives on their taxable income, can have an effect on the capital structure of a firm, since a high effective tax rate could be reduced by adding extra debt, and therefore create extra interest deductions, to lower the taxable income. Hence, an increase of the debt ratio is anticipated when the effective tax rate increases.

Lastly, also an increase of the level of debt is expected when the joint impact of the statutory tax rate and the tax loss carry back and carry forward rises. This is due to the fact that a high tax rate will incentivize firms to, other things being equal, act faster upon a change in tax loss deduction periods. This indicates that a change in the loss carry back and carry forward period has a greater effect when the statutory tax rate of country is high.

2.3.2. Control variables

To get a clear view on how the before mentioned tax and tax related determinants influence the capital structure of a firm, one has to also take other relevant proxies into account to prevent an omitted variable bias in the model. Therefore the above discussed control variables and the empirical results related to the variables (section 2.2.2.2. to 2.2.2.7) will also be taken into account when testing for the effects of tax and tax related proxies on the capital structure of firms. A short recapitulation of the expected effect of the determinants will be given hereafter.

(24)

Page | 24 is no complete consensus about the growth opportunities amongst researchers. However, the majority of the empirical literature finds an inverse relationship between growth and the debt ratio. Fifthly, an increase of age of a company is expected to reduce the leverage within a firm, because older firms tend to have more internal funds. Therefore external financing is not or less needed. Lastly, the probability of bankruptcy is inversely related to the firm’s debt level, because when the bankruptcy probability is low, new creditors can be found willingly to finance new debt.

The before mentioned hypothesized effects of the determinants of capital structure are summarized in table 2.1.

Table 2.1 Expected coefficient signs

This table conveys information regarding the expected sign of the effects of the variables on the debt level of a company when an increase in the independent variables occurs. Column one contains a description of the variables. The second column gives the expected sign of the change in debt ratio when an increase occurs in the variable of column one.

Variable Expected sign

Independent variables

Statutory corporate tax rate +

Carry back period +

Carry forward period +

Fiscal Freedom -

Effective tax rate +

Interaction of the statutory tax rate with

tax loss carry back/forward period +

Control variables

Profitability -

(25)

Page | 25

Tangibility +

Growth -

Firm age -

(26)

Page | 26

3. Methodology

In this section, we will discuss the methodology used throughout this research. In the first section the procedure regarding the regression equation will be discussed. This is subdivided in the defining of the dependent variable, independent variables and the control variables. To get a good understanding of the definition of the variables and how they were retrieved, we refer to Appendix A. Thereafter, the model used to test the hypotheses is being considered.

3.1. Regression equation

To see what the effects of tax and tax related variables are on the capital structure within a firm, we conduct an econometric analysis using equation (1). Hereafter, we will go further into detail regarding the variables from which the equation was constructed.

Our regression equation is of the following form:

(1)

Where denotes the debt ratio of firm i in industry j at time t. Furthermore, is a (row) vector of independent (tax-related) variables that include the statutory corporate tax rate (CIT), carry back period (CBP), carry forward period (CFP), Fiscal Freedom (FF), effective tax rate (ETR), interaction term of the statutory tax rate with the carry back period (CIT·CBP) and the interaction term of the statutory tax rate with the carry forward period (CIT·CBP). is the (column) vector of corresponding parameters. Similarly, consists of the firm-specific control variables, including the lagged debt ratio (LAG DEBT) as well as profitability (PROFIT), firm size (SIZE), tangibility of assets (TANG), growth opportunities (GROWTH), age of the firm (AGE) and the probability of bankruptcy (ZPROB) again with being the vector of corresponding parameters. In addition, and denote the industry and time fixed effects, whereas

(27)

Page | 27 3.1.1. Dependent variable

The dependent variable, , represents the debt ratio of firm i, in industry j at time t. This ratio is defined as the total debt within a firm divided by the book value of assets. The first independent variable from equation (1) is the lag of the dependent variable,

- or LAG DEBT. This determinant is included in the regression to capture the

dynamic effect of the capital structure, whilst also reducing the autocorrelation in the model. For robustness, also two regressions will be run with long-term debt and short-term debt in the numerator, as also was done by Titman and Wessels (1988).

3.1.2. Independent variables

To test the hypotheses mentioned before, we included the following independent variables: , , , and . They represent the statutory corporate tax rate, the carryback period, the carry forward period, Fiscal Freedom and the effective tax rate. Also, two interaction terms are included: and .

The statutory corporate tax rate, , is defined as the highest standard corporate tax rate imposed on companies incorporated within the relevant country.

The carry back and carry forward period, and are defined as the maximum period that it is possible for companies to deduct their net operating losses from past or future taxable income.

Fiscal freedom, , is a ranking between zero and 100 that communicates the fiscal burden in a country. The Heritage Foundation calculates the fiscal freedom per country based on three macro-economic pillars, namely the top marginal tax rate on individual income, the top marginal tax rate on corporate income and the total tax burden as a percentage of GDP. A high score on Fiscal Freedom represents a low governmental tax burden on its country’s inhabitants.

(28)

Page | 28 Also, two interaction terms are entered into the equation to capture the joint effect of the statutory corporate tax rate ( ) and the carry back and carry forward period ( and ). The interaction term captures the conditional effect that the length of the carry back and carry forward period have on the effect of the statutory tax rate on the leverage within a firm. The and measures are the mere product of with and with .

3.1.3. Control variables

The control variables in this model constitute several regressors put forward by the empirical literature on the capital structure subject. They have proven to be explaining the debt-equity mix of a firm and are therefore taken into account to prevent that the independent variables are wrongly interpreted as being statistically significant (omitted variable bias or type 1 error). Also, a few dummy variables are comprised in the regression equation to account for time and industry effects.

The variable conveys information regarding the profitability of firm i, in industry j at time t. This measure is defined as the net income scaled by the book value of total assets.

The size of the company can be an important determinant, as it can influence the relative ease of attracting funds to the firm. The variable is represented in the equation by and is defined as the natural logarithm of total assets. The natural logarithm is used, because of its useful underlying properties. The logarithm reduces the effect of outliers on the regression results and eases the interpretability of the coefficient.

Tangibility, represented by , is defined as the net property, plant and equipment of a firm divided by the firm’s book value of total assets. This determinant captures the effect of the level of collateral within a firm on its debt ratio.

(29)

Page | 29 ambiguous when investigating the (empirical) literature. The variable is defined as the market-to-book value.

Also firm age enters the equation. Pfaffermayr, Stöckl and Winner (2013)

showed that the increase of the age of a firm reduces the debt ratio of that firm. is measured as the difference between year t and the year that the firm was founded. When this date is not available, the date of incorporation is used.

Lastly, as was already mentioned in section 2.2.2.7, Altman (1968) showed the importance of bankruptcy cost on the level of debt of a firm. This was in later years also confirmed by MacKie-Mason (1990) and Graham (1996b). Therefore, this variable should also be included in the model. MacKie-Mason has defined the bankruptcy probability as:

(2)

This equation is a tailored version of the original Altman (1968) formula. An apt remark, by MacKie-Mason (1990), is that equation (2) is not exactly calculating the probability of bankruptcy, as it is not restricted to lie between the 0 to 1 range. However, it does convey information regarding the bankruptcy possibility of a firm by the level of .

(30)

Page | 30 in the sample are divided into six General Industry Classification Code subgroups (see

Appendix A for the definition). For each subgroup, a dummy variable is created representing the specific industry that can take on a value of either 1 or zero. Also a dummy variable for time is introduced. The dummy will be constructed in a similar matter as the industry dummy, thus for each year a dummy is set up which can take on a value of 1 or zero. An often reoccurring problem with the usage of dummy variables is the so called dummy variable trap (Brooks, 2008). When a constant term together with dummy variables are entered into the equation, multicollinearity exists. Therefore a constant term is not included in equation (1).

3.2. Regression model

The data used in this research can be classified as panel data since it consist of both time series and cross-sectional elements. For this research we use the Ordinary Least Squares method (OLS) with the inclusion of both time and industry fixed effects to verify if the stated effects of the determinants on the capital structure of a firm are indeed present. However, this method implies that the components of the regression equation are stationary. To check whether the dependent variable, the debt ratio, is stationary, we will perform a unit root test in section 5.1. A non-stationary variable implies that the underlying probability distribution changes over time and the process does not have a finite mean and variance. This could lead to wrong inferences such as spurious regressions but it can also influence the properties and behavior of the variable itself. The unit root test performed is the Levin, Lin and Chu test (2002), which is solely used in panel settings.

(31)

Page | 31 heteroscedasticity. The White’s period standard errors are used. This type of standard error is justified by the fact that the number of cross-sections is larger than the number of periods.

Autocorrelation is present when the covariance between the error terms over time or cross-sectionally are different from zero. Ignoring autocorrelation will result in unbiased estimates, however they will be inefficient (Brooks, 2008). A lag of the debt ratio will be included in the regression equations to reduce the potential autocorrelation. An additional advantage of the lag is the ability to distinguish between short and long term impacts of the explanatory variables.

Endogeneity within a variable exist when the variable is correlated with the error term of the regression equation. Endogeneity can have several causes, such as simultaneity, measurement error, omitted variables and autoregression with autocorrelated errors. By using several control variables in the regression equation, the possibility of omitted variable bias in the model will be reduced. Also, endogeneity will be tried to overcome by using other proxies for some of the control variables. This will reduce the chance of measurement error.

The explanatory power of the estimated model will be evaluated via the adjusted R2_{measure. This measure, unlike R}2_{, takes into account that adding extra} regressors to the model will reduce the degrees of freedom. The adjusted R2_measure will therefore be a helpful tool in determining whether a variable should be included in the model by looking if the overall explanatory power of the model increases or decreases when the variable is added or omitted from the analysis.

(32)

Page | 32

4. Data

In this section, the data underlying this study will be introduced and the sample selection procedure will be conveyed. Lastly, the descriptive statistics and the corresponding correlation matrix of the data used throughout the remainder of this paper will be presented.

4.1. Sample selection

The data comprising this study consists of 2,497 companies in 73 countries for a period of 15 years (2000 to 2014). The companies are selected by the availability of loss carry back and carry forward data. This information was gathered via the PKF worldwide tax guides and the Ernst & Young worldwide corporate tax guides for the years of 2003 to 2013. Many countries impose a carry forward period that lasts for infinity. As this number mathematically cannot be handled, the countries that impose an infinite period of tax loss deduction will be set equal to the maximum period that is defined in the data sample (for carry back this period is 3 years, for carry forward this period is 20 years).

If data on the tax loss compensation periods was available for a certain country, then a corresponding list of publicly listed companies within this country was gathered from the Datastream database. We choose to use only one database for the accounting observations to reduce accounting measurement error, and thus endogeneity. However, every country often has different accounting standards and could therefore potentially lead to measurement error. This should be kept in mind when interpreting the results. From these indices the data of every company comprising those lists were gathered for the years of 2000 to 2014.

(33)

Page | 33 Also, a measure for the fiscal freedom within a country is used in the regressions. This measure is composed yearly by the Heritage Foundation3_.

Some adjustments to the data sample have been made. First, 72 firms with no corresponding name, country or industry distributed by the database were deleted. Next, after checking the sample for companies appearing twice or more in the dataset, 53 companies were deleted. Lastly, 5 companies were eliminated because they were incorporated in countries from which no tax indicators were available (Bermuda and the Faroe Islands). Also, not all tax and tax related variables were available for the complete data range 2000 to 2014. This results in a data sample consisting of 2,367 companies situated in 71 countries for the period 2003 to 2013. In table 4.1 the sample composition is displayed in more detail, however for convenience the table is subdivided into continents. Also, the companies are ordered by the industry they are in. The General Industry Classification code (GIC) per company was for this purpose also extracted from Datastream. In Appendix B an extensive version of the table is available containing the number of companies per country and industry.

3

(34)

Page | 34 Table 4.1 sample composition per industry and continent

This table contains the composition of the data gathered ordered by industry and continent. The table conveys information regarding the number of companies within a continent in a specific industry. The industry groups are classified by the General Industry Classification Code, where 1 is Industrial; 2 is Utility; 3 is Transportation; 4 is Bank/Savings and Loan; 5 is Insurance and 6 is Other Financial. In Appendix B, a more elaborate table is presented with the number of companies per industry and country.

Continents Industries 1 2 3 4 5 6 Total % of total Africa 162 10 3 43 14 11 243 10.3% Asia 503 77 25 128 29 41 803 33.9% Europe 605 68 19 82 33 28 835 35.3% North America 173 7 7 10 3 2 202 8.5% Oceania 95 12 8 7 5 21 148 6.3% South America 87 21 2 20 0 6 136 5.7% Total 1,625 195 64 290 84 109 _2,367 % of total 68.7% 8.2% 2.7% 12.3% 3.5% 4.6%

This sample can be classified as an unbalanced panel, due to the missing observations (both cross-sectional as well as over time). These missing observations are recognized by the statistical program conducting the regression (Eviews, 7th_{edition). When a} missing observation is identified, all the data of that specific firm in that specific year are not taken into account. This will therefore lead to different number of observations in different regressions. The number of observations will always be mentioned in the regression result tables in section 5.

4.2. Descriptive statistics

(35)

Page | 35 Table 4.2a Descriptive statistics of the dependent and independent variable This table contains information regarding the descriptive statistics of the dependent and independent variables of the sample. The dependent variable DEBT represents the total debt divided by total assets. LAG DEBT is the one period lag of the dependent variable DEBT. The independent variables are defined as follows: CIT, statutory corporate income tax rate per country; CBP, carry back period per country; CFP, carry forward period per country; FF, represents the fiscal freedom of a country; ETR, the effective tax rate, is represented by income taxes paid divided by EBIT; CIT·CBP and CIT·CFP are interaction terms of CIT with CBP and CFP.

DEBT LAG DEBT CIT CBP CFP FF ETR CIT·CBP CIT·CFP

(36)

Page | 36 Table 4.2b Descriptive statistics of the control variables

In this table the descriptive statistics of the control variables in the sample are depicted. PROFIT represents the profitability of an enterprise. It is calculated as the operating income over total assets. SIZE represents the size of the company and is calculated as the natural logarithm of total assets. TANG conveys information regarding the tangibility of assets of the firm. This is calculated as the net property, plant and equipment over total assets. GROWTH represents the growth opportunities of a company and is calculated as the market value of the company divided by the share holders’ equity. AGE represents the firm age and is calculated as the present year minus the founding or incorporation date. Lastly, ZPROB measures the probability of bankruptcy of a firm (see equation 2).

PROFIT SIZE TANG GROWTH AGE ZPROB

Mean _0.069 _14.480 _0.323 _0.002 _48.544 _1.113 Median _0.061 _14.610 _0.284 _0.002 ₃₄ _0.746 Standard Deviation _0.477 _2.480 _0.269 _0.025 _46.506 _58.025 Kurtosis _14,757.29 _3.137 _2.168 _13,767.94 _14.448 _9,460.109 Skewness _-110.743 _-0.199 _0.510 _-100.464 _2.336 _-52.388 Minimum _-66.438 _1.792 _0.000 _-3.335 ₀ _-6,671.447 Maximum _1.849 _21.905 _1.612 _1.044 ₅₄₂ _4,396.479 No of observations _26,144 _26,230 _26,130 _23,112 _20,942 _22,267

The dependent variable, total debt ratio, takes on a very wide range of values as can be seen in table 4.2a. The ratio can be as low as -0.001 and as high as 21, which indicates that there is 21 times more leverage than assets in a firm. A value above one can occur if a firm has more debt than equity. Also the statutory tax rate is varying measure as the lowest rate is 9%. However, the highest rate is 52%, which is almost 6 times as high. Also the effective tax rate draws our attention as companies can have an effective tax rate as low as -3,900% or as high as 12,500%. This is often a combination of low Earnings Before Interest and Taxes (EBIT) and high income taxes (attributable to items subject to special taxation rules, for example R&D) or vice versa.

(37)

(38)

Page | 38 Table 4.3 Correlation matrix of the regression variables

This table displays the correlation between the dependent, independent and control variables in the sample. The dependent variable DEBT represents the total debt divided by total assets. LAG DEBT is the one period lag of the dependent variable DEBT. The independent variables are defined as follows: CIT, statutory corporate income tax rate per country; CBP, carry back period per country; CFP, carry forward period per country; FF, represents the fiscal freedom of a country; ETR, the effective tax rate, is represented by income taxes paid divided by EBIT; CIT·CBP and CIT·CFP are interaction terms of CIT with CBP and CFP. The control variables are: PROFIT represents the profitability of an enterprise. It is calculated as the operating income over total assets. SIZE represents the size of the company and is calculated as the natural logarithm of total assets. TANG conveys information regarding the tangibility of assets of the firm. This is calculated as the net property, plant and equipment over total assets. GROWTH represents the growth opportunities of a company and is calculated as the market value of the company divided by the share holders’ equity. AGE represents the firm age and is calculated as the present year minus the founding or incorporation date. Lastly, ZPROB measures the probability of bankruptcy of a firm (see equation 2). The order of statistical significance is conveyed via asterisks, where the level of significance of 0.1, 0.05 and 0.01 will be indicated with respectively *, ** and ***.

DEBT LAG DEBT CIT CBP CFP FF ETR CIT·CBP CIT·CFP PROFIT SIZE TANG GROWTH AGE ZPROB

(39)

Page | 39

5. Results

This section contains an overview of the empirical results. First, we will investigate whether or not the dependent variable is stationary. Thereafter the main results are presented. Lastly, the robustness of the findings is investigated by performing multiple robustness checks.

5.1. Stationary features of the dependent variable

As was mentioned in section 3.2, the OLS method used for our regressions assumes that the dependent variable is stationary. However, this is not necessarily true. We therefore test whether the debt ratio is indeed stationary by employing a unit root test. The unit root test performed is the Levin, Lin and Chu test (2002), which is solely used in panel settings. The output is given in table 5.1.

Table 5.1 Unit root test dependent variable DEBT

This table conveys the output of theLevin, Lin and Chu (2002) unit root test performed on the dependent variable DEBT. The first column represents the unit root method employed. The fourth and fifth column represent the number of firms that report a debt ratio and the total number of observations of the debt ratio in the data set. The third column indicates the probability of the null hypotheses being correct. The order of statistical significance is conveyed via asterisks, where the level of significance of 0.1, 0.05 and 0.01 will be indicated with respectively *, ** and ***.

Under the null, the LLC test assumes a common unit root process. This hypothesis is firmly rejected, indicating that no (common) unit root is present. Therefore we do not have to fear any spurious results originating from the fact that the dependent variable is non-stationary.

Method Statistic Prob. Cross-sections Obs

(40)

Page | 40 5.2. Estimation results

In this section we present our findings of the estimation of equation (1). We start by including all explanatory (independent and control) variables and test if the inclusion of time and industry fixed effects is appropriate. Thereafter, we will try to get a more parsimonious model with fewer explanatory variables as this allows us to estimate the parameters more precisely on the basis of a larger number of observations.

We will first examine whether the inclusion of fixed effects, which are used to control for unobserved heterogeneity between industries and over time, is appropriate in our setting. In table 5.2, regression equation (1) is estimated several times: without fixed effects, with time fixed effects (TFE), with industry fixed effects (IFE) and with both TFE and IFE. The output is given in table 5.2 on the next page.

Table 5.2 Regression results with and without fixed effects

In this table, the results are presented of various regressions with the in or exclusion of time fixed effects (TFE) and industry fixed effects (IFE). They are conveyed in the third and fourth last row. The dependent variable DEBT represents the total debt divided by total assets. LAG DEBT is the one period lag of the dependent variable DEBT. The independent variables are defined as follows: CIT, statutory corporate income tax rate per country; CBP, carry back period per country; CFP, carry forward period per country; FF, represents the fiscal freedom of a country; ETR, the effective tax rate, is represented by income taxes paid divided by EBIT; CIT·CBP and CIT·CFP are interaction terms of CIT with CBP and CFP. The control variables are: PROFIT represents the profitability of an enterprise. It is calculated as the operating income over total assets. SIZE represents the size of the company and is calculated as the natural logarithm of total assets. TANG conveys information regarding the tangibility of assets of the firm. This is calculated as the net property, plant and equipment over total assets. GROWTH represents the growth opportunities of a company and is calculated as the market value of the company divided by the share holders’ equity. AGE represents the firm age and is calculated as the present year minus the founding or incorporation date. Lastly, ZPROB measures the probability of bankruptcy of a firm (see equation 2). The coefficient of the determinants is given and, beneath the coefficient, the standard error is given in parentheses. The order of statistical significance is conveyed via asterisks, where the level of significance of 0.1, 0.05 and 0.01 will be indicated with respectively *, ** and ***. The last two rows indicate the explanatory power of the model

via the adjusted R2_{and the number of observations used in the regression. To cope with heteroscedasticity}

issues, the regressions are run with White’s period standard errors.

(41)

Page | 41 Tax CIT -0.0278 -0.0044 -0.0274 -0.0097 (0.0347) (0.0361) (0.0354) (0.0364) CBP 0.0120 ** 0.0120 ** 0.0123 ** 0.0123 ** (0.0054) (0.0053) (0.0054) (0.0053) CFP -0.0016 ** -0.0013 * -0.0016 ** -0.0014 * (0.0008) (0.0008) (0.0008) (0.0008) FF -0.0004 ** -0.0004 *** -0.0003 ** -0.0004 ** (0.0002) (0.0002) (0.0002) (0.0002) ETR -0.0004 -0.0002 -0.0003 -0.0002 (0.0008) (0.0008) (0.0008) (0.0008) CIT·CBP -0.0428 ** -0.0415 ** -0.0462 ** -0.0447 ** (0.0186) (0.0184) (0.0185) (0.0184) CIT·CFP 0.0050 ** 0.0038 0.0050 ** 0.0041 (0.0025) (0.0025) (0.0025) (0.0025) Control PROFIT -0.1889 ** -0.1899 ** -0.2045 ** -0.2048 ** (0.0850) (0.0846) (0.0869) (0.0866) SIZE 0.0014 0.0011 0.0028 * 0.0024 (0.0017) (0.0017) (0.0017) (0.0017) TANG 0.0457 *** 0.0459 *** 0.0286 *** 0.0294 *** (0.0139) (0.0139) (0.0109) (0.0110) GROWTH -0.0573 *** -0.0542 ** -0.0604 *** -0.0572 *** (0.0217) (0.0210) (0.0214) (0.0209) AGE -0.0000 -0.0000 0.0000 0.0000 (0.0000) (0.0000) (0.0000) (0.0000) ZPROB -0.0000 -0.0000 0.0000 0.0000 (0.0000) (0.0000) (0.0000) (0.0000)

TFE No Yes No Yes

IFE No No Yes Yes

Adjusted R² 0.6921 0.6936 0.6953 0.6966 Observations 10,951 10,951 10,951 10,951

(42)

Page | 42 To check whether the addition of fixed effects to the equation is appropriate, a Redundant Fixed Effects test can be implemented to check if the inclusion of period fixed effects adds to the significance of the model. Under the null hypothesis, all time fixed effects are equal to zero (that is, in equation (1) for all time periods), whereas under the alternative at least one is different from zero. Similarly, we can do the same test for industry fixed effects, via this so-called Wald test. Both test results are given in table 5.3.

Table 5.3 Verification correct usage of fixed effects

This table conveys information regarding the correct usage of fixed effects in the regression equation. The first column represents F-statistic concerning the period or industry fixed effects. In the second column, the used test is depicted. The last column displays the probability that the fixed effects have no effect in the regression equation. The order of statistical significance is conveyed via asterisks, where the level of significance of 0.1, 0.05 and 0.01 will be indicated with respectively *, ** and ***.

Test Statistic Method Statistic d.f. Probability

Period F-statistic Wald Test 5.759130 (10, 10,921) 0.0000 ***

Industry F-statistic Wald Test 3.321097 (5, 10,921) 0.0053 ***

From table 5.3 one can see that the inclusion of both time and industry fixed effects is appropriate, because the null, stating that including fixed effects is superfluous, is strongly rejected. We therefore from now on will include time fixed effects as well as industry fixed effects in our regressions.

5.2.1. Checking control variables

(43)

Page | 43 variables add significant information to the model by eliminating the insignificant control determinants one by one (starting with the least significant). The results can be seen in table 5.4.

Table 5.4 Regression results checking control variables

The table reports the results of several regressions to identify the correct control variables adding explanatory power to the final model. The dependent variable DEBT represents the total debt divided by total assets. LAG DEBT is the one period lag of the dependent variable DEBT. The independent variables are defined as follows: CIT, statutory corporate income tax rate per country; CBP, carry back period per country; CFP, carry forward period per country; FF, represents the fiscal freedom of a country; ETR, the effective tax rate, is represented by income taxes paid divided by EBIT; CIT·CBP and CIT·CFP are interaction terms of CIT with CBP and CFP. The control variables are: PROFIT represents the profitability of an enterprise. It is calculated as the operating income over total assets. SIZE represents the size of the company and is calculated as the natural logarithm of total assets. TANG conveys information regarding the tangibility of assets of the firm. This is calculated as the net property, plant and equipment over total assets. GROWTH represents the growth opportunities of a company and is calculated as the market value of the company divided by the share holders’ equity. AGE represents the firm age and is calculated as the present year minus the founding or incorporation date. Lastly, ZPROB measures the probability of bankruptcy of a firm (see equation 2). The coefficient of the determinants is given and, beneath the coefficient, the standard error is given in parentheses. The order of statistical significance is conveyed via

asterisks, where the level of significance of 0.1, 0.05 and 0.01 will be indicated with respectively *, ** and

***. The last two rows indicate the explanatory power of the model via the adjusted R2_{and the number of}

(44)

Page | 44 (0.0025) (0.0024) (0.0023) (0.0035) Control PROFIT -0.2048 ** -0.1958 ** -0.2503 *** -0.3413 * (0.0866) (0.0805) (0.0639) (0.1931) SIZE 0.0024 0.0024 0.0046 *** 0.0079 *** (0.0017) (0.0016) (0.0013) (0.0022) TANG 0.0294 *** 0.0292 *** 0.0419 *** 0.0476 *** (0.0110) (0.0107) (0.0148) (0.0095) GROWTH -0.0572 *** -0.0396 -0.0178 (0.0209) (0.0317) (0.0520) AGE 0.0000 0.0000 (0.0000) (0.0000) ZPROB 0.0000 (0.0000) Adjusted R² 0.6966 0.7082 0.5835 0.4803 Observations 10,951 12,204 16,947 18,369

In all the foregoing regression results, the bankruptcy probability (ZPROB) was highly insignificant. We therefore left it out of the regression equation in column 4, because it will both increase the number of observations and it will result in more accurate coefficients estimates of the other parameters without losing anything, due to the fact that ZPROB was never significant. As the last two rows of column 4 indicate, the adjusted R2_{rises whilst also a rise of observations occur by approximately 20%. The} same is done for firm age, because also this determinant has never been significant in any of the forgoing specifications. However, in this specification the adjusted R2_drops by almost 20%. When we look at the data sample that is added to the total sample due to the elimination of firm age (4,743 observations), we see that a few of the tax variables that were significant in the last specification are not significant in this subsample4_{, such} as the carry forward period, the fiscal freedom and the interaction term of the statutory tax rate and the carry forward period. This results in a weak adjusted R2_{of the} subsample. Therefore, the overall adjusted R2_{significantly drops in the specification of} the 4th_{column. In the last column, a model is specified without growth opportunities} (GROWTH). This result should however be interpreted with caution. In the foregoing

4

Tax system differences and their impact on capital structure