THE IMPACT OF EFFECTIVE TAX RATES
ON LEVERAGE: EVIDENCE FROM EUROPE
Frank Brinkman Student Number: S2374072
University of Groningen Faculty of Economics and Business
MSc. Finance
Supervisor B. van Oostveen June 2018
ABSTRACT
Using a dataset of the largest 1,157 listed firms in Western-Europe, this paper examines the impact of the effective tax rate on leverage taking endogeneity issues into consideration. In an attempt to circumvent these issues, the Arellano-Bond estimator is utilised. The findings of this paper show that over the total sample the effective tax rate shows a significant positive relation with leverage with consistent results between the ordinary least squares and Arellano-Bond estimator. When testing for robustness, the sample is split up based on geographic- and time-based characteristics. The results show inconsistencies in the significance of the effective tax rate and also between the OLS and AB estimator, possibly indicating that endogeneity is an issue on this topic. The positive relation between the effective tax rate and leverage is in line with famous capital structure theories as the capital structure irrelevance principle and the trade-off theory. Summarized, there is a significant relationship between the effective tax rate and leverage. However, the reason for the remarkable results in the sub-samples is definitely reason for further research on this topic.
I.
INTRODUCTION
The impact of the effective tax rate on leverage is a topic with mixed empirical results. Indicators of endogeneity issues are present which could lead to biased estimations. This paper covers the impact of the effective tax rate on the leverage of a sample of 1,157 listed Western-European firms. Many researchers have made attempts to establish theories of optimal capital structures and many performed empirical research on what variables determine the capital structure of a firm in practice. This paper will try to determine what impact the effective tax rate has on capital structure, in order to add a conclusive variable to the list of determinants.
As mentioned, extensive research on the determinants of leverage has been published. Also, the impact of the effective tax rate has been covered by previous research. However, this paper takes a look at a different geographical region and, most important, employs the Arellano-Bond (AB) estimator in an attempt to circumvent the effects of endogeneity. Endogeneity occurs when an explanatory variable is correlated with the error term. In more simple terms, it means that there is a problem with causation. Common causes of causation issues are reverse causality, an omitted variable, or a selection bias.
Previous research performed by Biger, Nguyen, and Hoang (2008) and Gill and Mathur (2011) found significant impact of the effective tax rate on leverage. However, no measures were used to control for endogeneity issues. Both papers used basic econometrical methods to estimate coefficients and check for a significant correlation between the effective tax rate and leverage. As these researchers lacked controlling for endogeneity, the results could be inconsistent due to the causes mentioned above.
In contrast, Stickney and McGee (1982) examined the correlation between the effective tax rate and leverage using the effective tax rate as the dependent variable. Their research claims that instead of taxes influencing leverage, that the amount of leverage influences the amount of taxes paid. However, this paper does not use methods controlling for possible endogeneity issues. Therefore, still being inconclusive about the actual causational effect.
result in a higher leverage, and in turn a higher leverage will again result in a higher effective tax rate. This suggests that a change in one of the variables may cause them to spiral up or down together. When such results are observed it is imperative to take endogeneity issues into consideration in the research design.
This paper includes two measures to control for endogeneity issues. The first measure is to include a lagged level of the dependent variable, the debt to equity ratio, in the list of independent variables. This variable essentially captures the same effect as using fixed effects in a regular OLS model. It also removes most of the reverse causality that exists between leverage and the effective tax rate. The second measure regards the method by which the independent variables are estimated. They are estimated using lagged levels of the variables itself. These two measures are both incorporated in the Arellano-Bond estimator, which will be discussed in more detail in the methodology section.
The research-gap this paper attempts to fill is one of a more econometrical nature. The topic has been researched before, however, not including the before mentioned measures against endogeneity. This paper focusses on endogeneity issues that may exist between leverage and the effective tax rate. As results of conventional methods will also be presented, it will show the differences between geographical areas used in previous research.
The results of this paper are rather interesting. When measures are taken to avoid effects of endogeneity, the impact of the effective tax rate on leverage remains significant. Even though most previous researchers have overlooked the possible issues of endogeneity, this paper finds rather similar results for both regular OLS and the Arellano-Bond estimator. Both in the regular sample as in most robustness tests a positive relation between the effective tax rate and leverage is found. Not all robustness samples show consistent results, this will be elaborated on further in the results section.
consecutive effects on inflation, system risk, and/or exchange rates. However, these exact effects are beyond the scope and expertise of this paper.
This paper continues as follows. Section 2 gives an overview of the relevant literature on the basis of which hypotheses are developed. Famous capital structure theories such as the capital irrelevance theorem, pecking order theory, and the trade-off theory, as well as more recent empirical results are discussed. Section 3 explains the methodology applied, which mainly focusses on the Arellano-Bond estimator. Section 4 explains how the sample of firms is selected, the variables are measured, and describes the descriptive statistics. Section 5 presents the results on the basis of which the hypotheses are accepted or rejected. Section 6 gives the discussion, conclusions and the several limitations of the outcomes. This section also provides opportunities for further research.
II.
LITERATURE REVIEW
As mentioned before, leverage is a well-researched topic in finance. That exogenous variables exist that impact the amount of leverage, can be concluded from many different empirical papers. However, there are also several capital structure theories that claim that capital structure should be determined by for example only tax benefits, or that capital structure solely depends on a specific pre-determined order of preference for new capital. This paper focuses on the effective tax rate as a determinant of how a firm is leveraged.
The literature review will be structured as follows. First of all, it will discuss famous capital structure theories such as Modigliani and Miller’s capital structure irrelevance principle, the by Myers and Majluf’s modified pecking order theory, the trade-off theory, and Titman and Wessel’s paper on the determinants of capital structure. Second, several determinants of leverage will be discussed. Third, this paper will elaborate on the determinant in question, namely the effective tax rate. Fourth, the topic endogeneity will be discussed, and after that the hypothesis will be developed.
Capital Structure
the absence of taxes, bankruptcy costs, agency costs, and asymmetrical information, the capital structure does not influence the value of a firm. Evidently, when removing these simplifications, this theory no longer perfectly holds. In their second collaborative paper (Modigliani and Miller, 1963) they post a correction of their initial paper, noticing that due to taxes the cost of debt is lower than the cost of equity. The conclusion is that a firm should attract increasing amounts of debt until this tax shield fully removed all taxable income.
Another famous theory on capital structure is the Pecking Order Theory, originally developed by Donaldson (1961) and modified by Myers and Majluf (1984). This theory implies that firms roughly have three options for sourcing funds. Namely internal funding, debt, and equity. This theory prescribes that firms choose their financing instruments by a pecking order of availability. First, they choose internal funding, then debt, and last equity. This theory does not show the definition on a perfect capital structure, it states a preference order of new capital that is employed in practise.
The trade-off theory is a theory that balances the costs and benefits of using debt or equity. The costs implied are those of financial distress and the benefits are those of tax advantages. When a high tax rate is present, the benefits of debt financing become relatively high. Therefore, a high tax rate should lead to higher leverage. The theory implies that there is an optimal point where those costs and benefits balance.
Many other theories and research on the topic of capital structure exist. For example a paper by Fama and French (2012) testing the different theories on capital structure, or a survey by Brounen, de Jong, and Koedijk (2006) showing practical interesting insights into how markets behave or what an optimal capital structure is. However, discussing all theories on capital structure is beyond the scope of this paper.
Determinants of capital structure
Intangibility of assets
Backed by Myers and Majluf (1984) and Scott (1977), firms tend to increase their debt levels as they have more tangible assets available to put up for collateral. As the value of tangible assets can more easily be observed by outside investors, the costs associated with issuing securities will be lower, causing firms to increase debt levels. Also, in case of bankruptcy, tangible assets are often valued higher than intangible assets. This results in bondholders to require an extra premium for highly intangible firms, leading to a higher cost of debt. Another argument is that for firms not in a position to offer collateral on their debt, the moral hazard and agency cost increase (Jensen and Meckling, 1976).
In conclusion, a firm with large levels of intangible assets is likely to be levered relatively low.
Non-debt tax shields
In the paper of DeAngelo and Masulis (1980) the theory that non-debt tax shields lower the value of debt tax shields is discussed. When firms carry non-debt tax shields such as depreciation and investment tax credits, the tax shield resulting from debt may become less effective as there may be no taxable income left. This could result in firms carrying less debt. However, the data availability of this variable is very poor and would drastically reduce the size of the sample. Also, as Titman and Wessels (1988) find no significant relation between non-debt tax shields and leverage, this variable is not used as a control variable in this paper.
Growth
That a relationship exists between growth and leverage, is not a topic of discussion and has been proven frequently. However, the magnitude and sign of this relationship is quite a difficult topic. Some theories argue a positive sign, others argue for the contrary.
growing firms as they are more flexible in their choice of future investments. Therefore, a negative correlation between growth and leverage is expected.
On the contrary, the pecking order theory by Myers (1984) could imply a positive correlation. Firms with large growth opportunities often require large sources of new funding. According to the theory, a firm would first exhaust all of its internal funding resources. However, if those are depleted, a firm would raise additional debt, and last equity. Leading to firms requiring large amounts of extra capital to be relatively largely levered. Thus, a positive correlation.
Empirical results of more recent literature on this variable show mixed results. Both positive relationships are found (Biger et al., 2008; Ozkan, 2001; Vo, 2017) as negative relationships (Gill and Mathur, 2011). A possible explanation for this generally uncommon negative relationship is the time-frame of the sample. The paper by Gill and Mathur uses a sample during the period 2008-2010, a period of global financial recession. High growth firms that have increased their leverage in the period before, may have been under pressure during the recession to lower their leverage.
Uniqueness
Firms that are selling a very unique product often face substantial selling expenses. Therefore, the selling expense to sales ratio is expected to be positive correlated to the uniqueness of a firm (Titman and Wessels, 1988). However, as uniqueness was not identified as having a significant influence on capital structure it will not be included as a control variable.
Industry classification
Size
Size is found to be an important determinant of leverage. Marsh (1982) finds that most firms make their financing decisions based on a set target debt to equity ratio. However, that target ratio is determined by size. Marsh (1982) finds evidence that large firms use more debt financing than small firms.
The smaller a firm, the larger the relative bankruptcy costs become. Also, large firms are often more diversified and less vulnerable to bankruptcy. These findings lead to the conclusion that the larger a firm is, the more leveraged it should be. (Warner, 1977) (Ang, Chua, and McConnell, 1982). Empirical results on the impact of size are consistently positive. (Biger et al., 2008; Gill and Mathur, 2011; Ozkan, 2001; Vo, 2017)
Volatility
According to Titman and Wessels (1988) is it suggested that capital structure is negatively affected by an increased volatility of earnings. However, as volatility was not identified as having a significant influence on capital structure it will not be included as a control variable.
Profitability
The influence of profitability on leverage is a topic with diverse results. Profitability has a close relationship with the pecking order theory. According to this theory, firms that produce considerable internal revenue streams are not as dependent on external funding, and therefore use more internal financing than debt financing. This suggests a negative relationship between profitability and leverage.
On the other hand, the trade-off theory suggests that profitable firms use more debt financing as they have a lower probability of bankruptcy (Fama and French, 2002). Also, firms with a large profits are able to benefit more from the tax shields accompanied by debt, leading to a higher leverage (Modigliani and Miller, 1963). These two theories suggest that actually a positive relationship relation should exist between profitability and leverage.
In table I an overview of the used variables including expectations based on both theory and empirics can be found. As this table does not provide any certainty about the probable signs of the variables, it provides an additional reason to perform further research on this subject. Also, the sign of the variable of interest, the effective tax rate, shows uncertainty by both theory and empirics. In the next section the theories about the impact of effective tax rate on leverage will be discussed.
Effective tax rate as a determinant.
The determinants of Titman and Wessels (1988) already show many of the major variables that impact the capital structure. Besides these major determinants, other factors are likely to exist that also affect leverage. The focus of this paper will be on the effective tax rate. The connection between capital structure and the effective tax rate has been researched before, however, with mixed results.
As discussed above, according to Modigliani and Miller (1963), firms should increase their debt levels up to the point that their tax shield is sufficiently high to erase all profitable income. This implies that a relatively low tax rate should lead to high debt levels. As more debt is required to erase all profitable income.
This line of reasoning is coherent with the trade-off theory. As discussed above, when the tax benefits of debt become relatively large due to high tax rates, firms are inclined to increase their debt level. Thus, implying a positive relationship between debt and leverage. Huang and Song (2006) present the determinants of capital structure of Chinese firms. They use a sample of 1200 listed firms over the period of 1994-2003 and find a significant correlation between leverage and commonly known determinants as tangibility, profitability,
Table I
Determinants of leverage: Overview expected signs
Variable Expected sign by
theory
Expected sign by empirics
Effective tax rate +/- +/-
Intangibility - +
Growth +/- +/-
Size (ln) + +
size, growth, and specific industries. Also, more interestingly, they find a mixed result for the effective tax rate. Dependent on the type of measurement of leverage, the effective tax rate is significant or not. When the total debt to equity ratio is used as a measure, they find no effect. However, when using the market total debt to equity ratio a positive coefficient of 0.060 is found. Unfortunately, as this paper was not focused on the effective tax rate specifically, no further explanation is provided for this disparity.
In the years after, more researchers study the determinants of capital structure including an effective tax rate factor. Research by Biger, Nguyen, and Hoang (2008) on the determinants of capital structure indicate a significant negative coefficient of -0.090 for the effective tax rate. This paper, based on a sample of Vietnamese firms during 2002-2003, concludes that there are two factors influencing the impact of the effective tax rate. A positive effect is expected as firms benefit from the tax shield obtained by debt. However, a negative sign could be expected as the firm’s profitability decreases by a higher tax rate and therefore find it more difficult to obtain debt. Consequently, they conclude that the second effect if more powerful. The drawback of this paper is that is employs the relatively elementary OLS method and that a rather short time frame is employed.
Gill and Mathur (2011) also estimate the determinants of capital structure. They analyse the determinants of capital structure on a sample of 166 Canadian firms during 2008-2010. Their research indicates that the effective tax rate is negatively correlated to leverage with a coefficient of -0.163, slightly larger that found by Biger et al. (2008). Unfortunately, this paper is rather brief in its explanatory section and no further interpretation of this effect is discussed.
Most recently, Faulkender and Smith (2016) published a paper on the impact of taxes on leverage taking certain endogeneity issues into consideration. The method they used is described in the next section. They found a significant positive impact of the effective tax rate on leverage of 0.240.
Endogeneity
What most above-mentioned researchers lacked is looking at possible endogeneity issues between the effective tax rate and leverage. From all research mentioned so far, it is fair to conclude that there is a connection between these variables. However, it is interesting to employ a new statistical approach on this subject to assess which effect holds at the end. The hypothesis that is tested in this paper is:
Hypothesis 1: The effective tax rate has a positive impact on leverage.
Endogeneity is becoming a more influential topic. Faulkender and Smith (2016) find evidence on the relationship of taxes and leverage taking endogeneity issues into consideration. Instead of utilizing a statistical model that circumvents endogeneity issues, they adapt the data in an attempt to achieve the same goal. They measure the impact of taxes on leverage by keeping the division of income from each subsidiary of a firm constant to the level of the first year of measurement. Therefore, a change in leverage is not influenced by the endogenous influence of a change of location, but by the exogenous shift in tax rates. They find evidence that higher taxes are a cause of increased leverage.
Another recent paper taking endogeneity into consideration on the topic of capital structure is research by Vo (2017). This paper estimates the determinants of capital structure by the use of the Arellano-Bond estimator. The determinants this paper tests are growth, tangibility, profitability, size, and liquidity. Unfortunately, the effective tax rate has been left out of consideration in this paper. By circumventing endogeneity, Vo (2017) finds significant estimators for all control variables used in this paper.
This paper combines the concepts of the two papers mentioned above by estimating the impact of the effective tax rate by utilizing the Arellano-Bond estimator.
III. METHODOLOGY
(OLS). In this model it is possible to account for the omitted variable bias using a fixed-effects model. However, it leaves other major endogeneity issues as reverse causality and selection bias unaddressed. This results in that research on this topic may only show correlation and not prove actual causation. This paper tries to address the endogeneity issues by utilizing of the Arellano-Bond estimator.
In order to get more accurate coefficients for the effective tax rate, control variables are added to the equation. It is important that the correct control variables are used as they may take away a significant amount of the effect for the tested variable. The control variables, included in the equation, are based on research of Titman and Wessels (1988) and are already described in detail in the literature review. This paper uses four control variables, namely intangibility, growth, size, and profitability. Intangibility, measures to what extent a firm has the opportunity to offer collateral. It is measured as a ratio of intangible assets to total assets. Growth, measured by using percentage growth in assets as a proxy. This may be slightly inaccurate for low asset, high revenue firms. However, the best proxy available for growth as it is more stable than revenue. Size, included as a natural logarithm of revenue and profitability as a ratio of operating income to total assets.
The Arellano-Bond estimator
After extensive research to more comprehensive models, the two-stage AB estimator has been identified as an econometrical model capable of dealing with an endogenous variable as the effective tax rate, without the necessity of identifying new instrumental variables (Vo, 2017).
This estimator uses generalized method of moments (GMM) for linear dynamic panel-data models. GMM is a method to estimate parameters using the moments of the random variables. A panel-model is a dynamic model if it includes the lagged dependent variable as an independent variable. Normally, a GMM model requires new instrumental variables, which were not possible to be identified. However, the AB estimator uses lags of its own levels as instruments circumventing the issue of no instruments.
Unfortunately, the amount of available post-estimation methods is very limited. It is not possible to test for heteroscedasticity after using the AB estimator. Therefore, a regular regression has been used in combination with a White’s test for heteroscedasticity. The null hypothesis of this test had to be rejected, concluding presence of heteroscedasticity. The AB estimator is capable of correcting for this problem by employing robust standard errors. Presence of heteroscedasticity in a regular OLS regression is no definite proof of its presence in the AB model. However, from a conservative perspective, it is safe to assume that this model might also suffer from heteroscedasticity and therefore should be corrected for.
The single post-estimation method available is a test for serial correlation. The AB estimator is only consistent if no second-order serial correlation exists between error terms of the first-differences equation. To test for this presence, a post-estimation method is available. As long as this test does not reject the null for serial correlation in the second-differences errors, the AB estimator is a consistent method.
Problems and solutions
When estimating the impact of the effective tax rate on leverage there are four econometrical problems that occur. They will be discussed one by one, including their solutions discussed by Mileva (2007).
First, in a regular OLS model it is assumed that all independent variables are exogenous. This means that the variable is not explained by other variables in the model. As it is not proven that the effective tax rate is exogenous, it should be treated as an endogenous variable. Frequently, researchers use an instrumental variable approach such as two-stage least squares to overcome this issue. However, as explained before, no instruments with significant explanatory power have been identified. The AB estimator deals with this issue by using a lagged dependent variable and differenced control variables as instruments.
Second, there may be time-invariant characteristics present such as country or firm specific effects. These should be taken into consideration as otherwise they would distort the results. The AB estimator takes first differences in the first stage, thereby removing the fixed effects present in each observation. As mentioned before, this causes adding industry or country dummies to be unnecessary.
Third, when including a lagged dependent variable as an independent variable, it gives rise to autocorrelation. However, this first-differenced lagged dependent variable is also estimated by its past level. Therefore, including the lagged dependent variable does not give rise to autocorrelation.
Fourth, our panel consists of many firms but limited time periods. For many estimation methods this is an issue, however, the AB estimator was designed for panels with this characteristic and should therefore not cause any estimation biases.
Research design
In order to test if the data shows similar results to previous research employing similar research methods, first a fixed-effects OLS model is tested using formula 1.
𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖𝑡 = 𝑎 + 𝛽1𝐸𝐹𝑇𝐴𝑖𝑡 + 𝛽2𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖𝑡+ 𝜇𝑖 + 𝜈𝑖𝑡 (1)
In which 𝑖 is a firm index, 𝑡 is a time index, 𝑎 a constant, 𝛽1 and 𝛽2 are coefficients for the effective tax rate and the control variables, 𝜇 the fixed-effects term, and 𝜈 the random error term.
As this method does not fully account for endogeneity, the more sophisticated AB estimator will be used to test if the relationship also holds if measures are taken to prevent endogeneity.
After estimations on the entire sample, robustness tests will be performed. The sample will be split up in different time-frames and regions. Performing these robustness tests increases the reliability of the research. If inconsistent results come out of the test, there could be extra underlying reasons for these differences or the effect might not be as harmonious in general.
Dependent variable
The dependent variable in this paper, a measure of leverage, is a variable that requires some discussion. Leverage is measured as a ratio of debt to equity. Research on the topic of capital structure has utilized several methods to calculate this ratio. The measurement of equity is rather straightforward, total assets minus total liabilities. However, for the measurement of the debt level, several different methods are found. Two major differences are part of this discussion. First, a distinction is often made between measuring debt with its market- or book value. The second distinction is made on which components of debt are included. Long term, short term, or the total amount of liabilities.
𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖𝑡 =
𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡𝑖𝑡
𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡 (2)
The method is equation 1 is commonly used. Another calculation, measuring leverage as total debt to total value, should yield similar results.
Independent variables
In order to estimate the impact of the effective tax rate as accurate as possible, it is essential to include control variables in the equation. The control variables selected for this paper are intangibility, growth, size, and profitability. In this section the calculation of the independent variables will be discussed in more detail and finally their expected signs will be discussed.
Effective tax rate
For calculations in this paper the effective tax rate is measured as in equation 3. 𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒 𝑡𝑎𝑥 𝑟𝑎𝑡𝑒𝑖𝑡 = 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑 𝑡𝑎𝑥𝑒𝑠𝑖𝑡
𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑏𝑒𝑓𝑜𝑟𝑒 𝑡𝑎𝑥𝑒𝑠𝑖𝑡 (3)
In which 𝑖 is an index for firms and 𝑡 an index for time. The calculation of the effective tax rate is rather consistent in academic literature, using a ratio of reported taxes to earnings before taxes. As this is common practice, and also used in comparable papers by for example Huang and Song (2006), Biger, Nguyen, and Hoang (2008), and Gill and Mathur (2011), it is also used in this paper to be able to compare results properly.
Control variables
variable this paper attempts to test. The control variables this paper uses are results of the well-known paper by Titman and Wessels (1988). This, almost 7,000 times cited paper, is considered a well renowned source for determining our control variables. Recently, Faulkender and Smith (2016) again confirm the relevance and accuracy of these variables. Titman and Wessels find the four variables mentioned above to have a significant effect on leverage, leading to the use of those variables as control variables in this paper.
The formula’s this paper employs for the control variables are displayed in equation 4-7. 𝐼𝑛𝑡𝑎𝑛𝑔𝑖𝑏𝑖𝑙𝑖𝑡𝑦𝑖𝑡 = 𝐼𝑛𝑡𝑎𝑛𝑔𝑖𝑏𝑙𝑒 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡 (4) 𝐺𝑟𝑜𝑤𝑡ℎ𝑖𝑡 =𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡− 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑖𝑡−1 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡−1 (5) 𝑆𝑖𝑧𝑒𝑖𝑡 = 𝐿𝑛(𝑅𝑒𝑣𝑒𝑛𝑢𝑒)𝑖𝑡 (6) 𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖𝑡= 𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑛𝑔 𝑖𝑛𝑐𝑜𝑚𝑒𝑖𝑡 𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡 (7)
The control variables and the methods of calculation in this paper are based on the paper by Titman and Wessels (1988). However, these methods are similar as those used in other papers on determinants of capital structures by Gill and Mathur (2011) and Mukherjee and Wang (2013).
IV.
DATA
have become increasingly efficient and liquid. Therefore, including data from before 2000 might not represent the current market situation optimally.
This paper differentiates itself on two major aspects. First, most important, it deals with endogeneity. Second, previous research on this topic has mostly been conducted outside of Europe. Researching this demographic could lead to new interesting insights. In table II a detailed overview of the variables used, and their definitions can be found.
Descriptive statistics
In table III descriptive statistics can be found. In the second column the total amount of observations before trimming are displayed. After removing all the observations with incomplete and irrational data, 14,206 observations remained. For the variables leverage and intangibility, all values outside the 0-1 range have been removed as these are not reasonable. For the effective tax rate, a range of 0-0.7 has been applied. Effective tax rate below 0 and above 0.7 are highly exceptional cases as this is far below or above regular statutory tax rates of any country. Growth has been cut off around a range of -0.9 and 3. Firm’s decreasing by more than 90% or increasing by more than 300% may be considered highly exceptional situations. For size no trimming has taken place. Profitability has been cut off around a range of -0.5 and 1.
Considering the means of the sample, all variables but growth can be considered rather
Table II
Overview and definitions of variables used in estimation
Variable Definition Source
Dependent variable
Leverage Ratio of total debt to total equity Datastream
Independent variable
Effective tax rate Ratio of income taxes to earnings before taxes. Datastream
Control variables
Intangibility Ratio of intangible assets to total assets Datastream Growth Percentage growth between the current year and the
previous year.
Datastream
Size (ln) Natural logarithm of total revenue Datastream
average growth rate of 11.4% if well above the average growth of the economy, showing that firms within this sample are able to outperform the average growth of the economy on the long term.
Besides basic descriptive statistics, also skewness and kurtosis are presented in table III. Skewness and kurtosis are statistics that inform about the normality of the data. Skewness informs about the shape of the distribution and its symmetry. Kurtosis measures the fatness of the tails. A skewness of zero and a kurtosis of three would show a perfect normal distribution. If skewness and kurtosis are off these target numbers, the data may not be normally distributed. The skewness and kurtosis for size are exceptionally high. Therefore, this variable has been transformed by taking a natural logarithm. If data deviates significant from a normal distribution, it could be problematic for small samples. For this sample, non-normally distributed data is not an issue as it is considerably large.
Table III Descriptive statistics
In this table the descriptive statistics for the 1,157 largest listed Western-European firms active during (part of) 2000-2016, for which sufficient trustworthy data could be gathered, are presented. The observations show the total amount of observations available before trimming. The descriptive statistics for the other variables are based on the observations used in the actual calculations. That sample has 14,206 observations. All data is gathered from the Datastream database. Leverage is calculated as debt/equity. Effective tax rate is calculated as income taxes/earning before tax. Intangibility is calculated as intangible assets/total assets. Growth is calculated as percentage change in total assets in years. Size is a measure of total revenue. Profitability is calculated as operating income/total assets.
Variable Observations Mean St. dev. Min Max Skew. Kurtos.
Leverage 19,533 0.389 0.233 0.000 0.999 0.266 2.508
Effective tax rate 19,517 0.275 0.119 0.000 0.700 0.377 4.003
Tangibility 19,436 0.182 0.185 0.000 0.923 1.110 3.525
Growth 19,133 0.114 0.256 -0.888 2.978 4.046 30.311
Size 19,550 4.486e10 2.120e11 3.168e3 3.54e12 8.286 147.383 Profitability 19,548 0.078 0.076 -0.732 2.117 3.661 71.666
as the Netherlands and the UK are creditor-oriented. The main differences between these systems are present in the way that is being dealt with bankruptcy. In a debtor-oriented system the debtor is favoured in an impending bankruptcy, and likewise similarly for a creditor-oriented system. These country specific effects could be a determinant of leverage.
In table IV the amount of observations and the weighted mean value per variable per country can be observed. For none of the countries highly significant outliers are present. This can be explained by the fact that all these firms are operating in a highly competitive and efficient international market. The only slight outlier that is present is in Norway in total asset size. Compared to Finland, Norwegian firms seem to be 10 times smaller. This is not considered to be an issue for this paper as these firms are still reasonably large.
Table IV
Mean values by country for sampled listed Western-European firms in the period 2000-2016
In this table the mean values of all the variables used for all separate countries can be found for the 1,157 largest listed Western-European firms active during (part of) 2000-2016, for which sufficient trustworthy data could be gathered. All data is gathered from the Datastream database. The variables in the headers of each of the columns are abbreviations for Observations, Leverage, Effective tax rate, Intangibility, Growth, Size, and profitability. Leverage is calculated as debt/equity. Effective tax rate is calculated as income taxes/earning before tax. Intangibility is calculated as intangible assets/total assets. Growth is calculated as percentage change in total assets in years. Size is a measure of revenue and reported in billions of Euros. Profitability is calculated as operating income/total assets. The bottom row shows the total amount of observations and weighted averages for the variables.
Country Obs. LEVE EFTR INTA GROW SIZE PROF
Table V shows the correlation matrix. The largest correlation of -0.319 is between profitability and size. This does not cause for any suspicion for collinearity in the sample as it is sufficiently low. In the appendix details regarding the variance inflation factor can be found.
Another measure to test our dataset is the variance inflation factor (VIF). The VIF is a post-estimation test for a regular OLS regression. It measures whether there are signs of multicollinearity and to what extent the variables are sensitive for adding or removing variables from the model. Chatterjee (2012) defines that there is an issue with multicollinearity if one of the variables indicates a VIF larger than 10 or if the average value of the VIFs is substantially larger than 1. We can be seen in table A.I in the appendix, the values of the VIF test do not demonstrate any issues of multicollinearity.
V.
RESULTS
The results found in this paper confirm the positive relationship between the effective tax rate and leverage found by Huang and Song (2006) and Faulkender and Smith (2016). Generally, the OLS and AB estimator find a similar positive relation, therefore we do not reject our hypothesis.
Table V
Correlation matrix for sampled listed Western-European firms in the period 2000-2016
In this table the correlation matrix for the 1,157 largest listed Western-European firms active during (part of) 2000-2016, for which sufficient trustworthy data could be gathered, are presented. Leverage is calculated as debt/equity. Effective tax rate is calculated as income taxes/earning before tax. Intangibility is calculated as intangible assets/total assets. Growth is calculated as percentage change in total assets in years. Size is a measure of revenue. Profitability is calculated as operating income/total assets.
[1] [2] [3] [4] [5] [6]
[1] Leverage 1.000
[2] Effective tax rate 0.040 1.000
[3] Intangibility 0.011 0.010 1.000
[4] Growth 0.003 0.003 0.044 1.000
[5] Size (ln) 0.209 0.019 0.049 -0.102 1.000
Another interesting result to be found in this dataset is that results differ substantially between regions in Western-Europe and that there are also differences between time periods. In this section, first the results from the OLS and AB estimator on the entire sample will be discussed. Secondly, the robustness test results are presented.
Regression results
As discussed earlier, the first step taken in this paper is to compare the results of the dataset with the results of previous research utilising similar research methods. The papers basic results are compared to, are the papers by Huang and Song (2006) and Gill and Mathur (2011). They employ an ordinary least squares method to measure the relation between the effective tax rate and leverage. As mentioned before, these researchers found mixed results. Huang and Song (2006) found no effect when using the book debt to equity ratio and a small positive coefficient of 0.060 when using the market debt to equity ratio. Gill and Mathur (2011) found a negative coefficient of -0.163 for the effective tax rate.
On the left-hand side of table VI, the OLS results for this paper can be found. As can be seen, both the regression with and without control variables show similar positive highly significant coefficients. This is in line with the research of Huang and Song (2006) but contradicts with the research of Gill and Mathur (2011).
As has previously been discussed, the OLS model may potentially be considered a biased estimator due to endogeneity issues. In order to circumvent these issues, the AB estimator is used to provide unbiased estimators for the effective tax rate and the control variables. On the right-hand side of table VI these results can be found. The coefficients suggested by the AB estimator seem to be rather identical to coefficients from the OLS estimation. The signs are all identical and also the magnitude of the coefficients seem to be closely related. From this sample we might infer that endogeneity issues do not carry substantial weight to change the outcome of the estimations. In case severe endogeneity issues would be present, these coefficients are not likely to be similar.
described in section 2. Profitability shows a negative sign, also in line with the expectations. What is interesting to notice in table VI, is that intangibility is no longer significant under the AB estimator. In the next section this will be elaborated on more thoroughly by also taking the results from the robustness tests into consideration.
Table VI
OLS and Arellano-Bond estimation: Determinants of capital structure on the full sample
This table shows the results of the OLS and Arellano-Bond two-step estimations. The results are based on the full sample of the 1,157 largest listed Western-European firms. Columns 1 and 3 show results without the control variables. Columns 2 and 4 show results including control variables. All data is gathered from the Datastream database. The dependent variable is leverage, a ratio of debt to equity. The independent variables are: leveraget−1, a lagged level of the dependent variable;
effective tax rate, a ratio of income taxes to earnings before taxes; intangibility, a ratio of tangible assets to total assets; growth, a percentage change between years of total assets; size, a natural logarithm of revenue; profitability, a ratio of operating income to total assets. The numbers between brackets show the for heteroscedasticity corrected robust standard errors. *, **, And *** represent the significance level for the coefficients at respectively the 10%, 5%, and 1% level.
OLS Arellano-Bond [1] [2] [3] [4] Leveraget−1 0.517 (0.045) *** 0.640 (0.044) *** Effective tax rate 0.059
Robustness
In order to conclude that the results presented above do not only hold in our specific sample group, robustness checks are carried out. In table VIII and IX an overview of 16 different robustness tests can be found. The sample is differed by both geography- and time-based characteristics. The specifications of each sub-sample can be found in table VII.
In table VIII robustness tests can be found for the sample separated by geographical characteristics. First, especially looking at the results for the effective tax rate, it can be seen that the results between the OLS and AB estimator are again quite consistent. Showing no indication of endogeneity issues regarding this variable.
However, the results within the geographical samples are not consistent. The sub-samples for Eastern Europe and Scandinavia show insignificant results for the effective tax rate. However, the sub-samples for Western- and Southern Europe show significant results. This leads us to conclude that there are substantial differences between regions. Unfortunately, within the scope of this paper no relationship has been identified to explain the differences between these sub-samples.
Table VII
Overview of sub-sample descriptions for robustness tests
Column Sub-sample Specification
Table VIII Geography-based
[5] & [9] Eastern-Europe Greece and Czech Republic
[6] & [10] Scandinavia Sweden, Norway, Finland, and Denmark
[7] & [11] Western-Europe Ireland, United Kingdom, Netherlands, Germany, and Belgium
[8] & [12] Southern-Europe France, Portugal, Spain, and Italy
Table IX Time-based
Table VIII
OLS and Arellano-Bond estimation: Determinants of capital structure on geographical sub-samples
This table shows the results of the OLS and Arellano-Bond two-step estimations. The results are based on sub-samples of the total sample, split up by geographical qualifications. Columns 5 and 9 are based on a sub-sample of Greece and Czech-Republic; columns 6 and 10 are based on a sub-sample of Sweden, Norway, Finland, and Denmark; columns 7 and 11 are based on a sub-sample of Ireland, United Kingdom, Netherlands, Germany, and Belgium; columns 8 and 12 are based on a sub-sample of France, Portugal, Spain and Italy. All data is gathered from the Datastream database. The dependent variable is leverage, a ratio of debt to equity. The independent variables are: leveraget−1, a lagged level of the dependent variable; effective tax rate, a ratio of income taxes to earnings before taxes; intangibility, a ratio of tangible assets to total assets; growth, a percentage change between years of total assets; size, a natural logarithm of revenue; profitability, a ratio of operating income to total assets. The numbers between brackets show the for heteroscedasticity corrected robust standard errors. *, **, And *** represent the significance level for the coefficients at respectively the 10%, 5%, and 1% level.
OLS Arellano-Bond [5]*** [6]*** [7]*** [8]*** [9]*** [10]*** [11]*** [12]*** Leveraget−1 0.305*** (0.211)*** 0.577*** (0.078)*** 0.512*** (0.082)*** 0.623*** (0.051)*** Effective tax rate 0.032***
Table IX
OLS and Arellano-Bond estimation: Determinants of capital structure on time-based sub-samples
This table shows the results of the OLS and Arellano-Bond two-step estimations. The results are based on subsamples of the total sample, split up by time-based qualifications. Columns 13 and 17 are for 2000-2008, 14 and 18 for 2009-2016, 15 and 19 for 2004-2012, and 16 and 20 for 2007-2010. All data is gathered from the Datastream database. The dependent variable is leverage, a ratio of debt to equity. The independent variables are: leveraget−1, a lagged level of
the dependent variable; effective tax rate, a ratio of income taxes to earnings before taxes; intangibility, a ratio of tangible assets to total assets; growth, a percentage change between years of total assets; size, a natural logarithm of revenue; profitability, a ratio of operating income to total assets. The numbers between brackets show the for heteroscedasticity corrected robust standard errors. *, **, And *** represent the significance level for the coefficients at respectively the 10%, 5%, and 1% level.
OLS Arellano-Bond
[13]*** [14]*** [15]*** [16]*** [17]*** [18]*** [19]*** [20]***
Leveraget−1 0.541*** 0.694*** 0.622*** 0.534***
The results of the control variables also indicate inconsistencies between these sub-samples and the full sample. For the sub-sub-samples based on Eastern-Europe and Scandinavia, in columns 5, 6, 9, and 10 respectively, almost no significant determinants were identified. In table IX the results can be found for the time-sensitive robustness tests. In this table there are several distinctive results to be noticed. For the effective tax rate, our variable of interest, the magnitude and the signs of the coefficients are still rather consistent. However, the results from the AB estimator are not consistent with the results from the OLS. This could indicate presence of endogeneity issues. For the control variables no major issues can be observed, except for intangibility. This variable loses significance with the AB estimator. This could imply that there are also endogeneity issues present for this variable.
Summarized, measured over our full sample, the effective tax rate appears to be a significant determinant of capital structure. However, when the sample is decreased in size and split up by several types of characteristics, it loses its significance in 6 out of 16 samples. This leads us to conclude that there is definitely a relationship between the effective tax rate and leverage. However, the possibility that there is another characteristic influencing this effect is plausible. Research into this characteristic or variable could be reason for additional research on this topic.
VI.
DISCUSSION & CONCLUSIONS
This paper measures the impact of the effective tax rate on leverage over a sample of the largest 1,157 listed Western-European firms. Mixed results and uncertainty on the direction of causation of the effective tax rate on capital structure are important indicators for endogeneity issues. This paper uses an uncommon method to measure the impact of the effective tax rate on a firm’s capital structure. By employing the Arellano-Bond estimator, issues regarding endogeneity are aimed to be circumvented.
However, it remains interesting to possibly find other methods to account for endogeneity and measure if results remain consistent.
The results found in this paper are in line with previous research by Huang and Song (2006) and Faulkender and Smith (2016). In their papers they found a positive relation between the effective tax rate and leverage. On the other hand, it directly contradicts research by Biger et al., (2008) and Gill and Mathur, (2011) which show a negative relationship.
When taking the robustness tests into consideration, the impact of the effective tax rate become slightly uncertain. All the coefficients for the effective tax rate show a consistent positive sign. However, the significance of this coefficient become too small in several cases.
The results for the geography-based sub-samples are consistent between the OLS and the AB estimator. Therefore, we conclude that there is no relation between the effective tax rate in Western-Europe and Scandinavia.
However, for the time based sub-samples, the results are not consistent between the OLS and the AB estimator. This indicates that signs of endogeneity issues are present. For the OLS it concludes significant outcomes over all time periods with rather similar magnitudes in the coefficients. The AB estimator only shows slightly significant results for two out of the four sub-samples.
There are several limitations to this paper in the context of data and research methods. These limitations could also be seen as opportunities for further research on this topic, attempting to overcome them. Firstly, the data collection process. This paper collected data on initially 1,500 firms in Western-Europe. After removing firms that lacked important data points, 1,157 firms were left. However, not all data potentially required was viable to retrieve. Information on non-debt tax shields was very limited and not readily accessible. Even though it was not recognized as a significant determinants in the paper by Titman and Wessels (1988), it could be interesting to measure its effect.
The second limitation is one in the design of the AB estimator. This estimator assumes linearity in its model. However, it is clear that the effective tax rate does not increase linear constantly and a curved model would be a better fit. However, as outliers have been removed from the sample, no significant estimation errors are expected. Nevertheless, it remains a topic of improvement.
A final limitation and reason for further research is the inconsistency in the significance of the coefficients for the effective tax rate. No explanation has been found for this occurrence. Possibly an extra characteristic is present that influences this outcome.
VII. REFERENCES
Ang, J. S., Chua, J. H., McConnell, J. J. 1982. The administrative costs of corporate bankruptcy: A note. The Journal of Finance 37, 219–226.
Arellano, M., Bond, S. 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies 58, 277.
Biger, N., Nguyen, N. V, Hoang, Q. X. 2008. The determinants of capital structure: evidence from Vietnam. International Finance Review 8, 307–326.
Brounen, D., de Jong, A., Koedijk, K. 2006. Capital structure policies in Europe: Survey evidence. Journal of Banking and Finance 30, 1409–1442.
Chatterjee, S., and Hadi, A. S. 2012. Regression analysis by example. 5th ed. New York: Hoboken, NJ.
DeAngelo, H., Masulis, R. W. 1980. Optimal capital structure under corporate and personal taxation. Journal of Financial Economics 8, 3–29.
Donaldson, G. 1961. Corporate debt capacity: A study of corporate debt policy and the determination of corporate debt capacity. Boston: Division of Research, Harvard School of Business Administration.
Fama, E. F., French, K. R. 2002. Testing trade-off and pecking order predictions about dividends and debt. Review of Financial Studies 15, 1–33.
Faulkender, M., Smith, J. M. 2016. Taxes and leverage at multinational corporations. Journal of Financial Economics 122, 1–20.
Gill, A., Mathur, N. 2011. Factors that Influence financial leverage of Canadian firms. Journal of Applied Finance & Banking 1, 19–37.
Huang, G., Song, F. M. 2006. The determinants of capital structure: Evidence from China. China Economic Review 17, 14–36.
Jensen, M. C., Meckling, W. H. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3, 305–360.
Marsh, P. 1982. The choice between equity and debt: an empirical study. The Journal of Finance 37, 121–144.
Mileva, E. 2007. Using Arellano – Bond Dynamic Panel GMM Estimators in Stata. Fordham University.
Modigliani, F., Miller, M. H. 1958. The cost of capital, corporation finance and the theory of investment. The American Economic Review 48, 261–297.
Modigliani, F., Miller, M. H. 1963. Corporate income taxes and the cost of capital: a correction. The American Economic Review 53, 433–443.
Mukherjee, T., Wang, W. 2013. Capital structure deviation and speed of adjustment. Financial Review 48, 597–615.
Myers, S. C. 1984. The capital structure puzzle. The Journal of Finance 39, 575–592 . Myers, S. C., Majluf, N. S. 1984. Corporate financing and investment decisions when firms
have information that investors do not have. Journal of Financial Economics 13, 187– 221.
Ozkan, A. 2001. Determinants of capital structure and adjustment to long run target: evidence from UK conpany panel data. Journal of Business Finance & Accounting 28, 175–198.
Scott, J. H. 1977. Bankruptcy, secured debt, and optimal capital structure. The Journal of Finance 32, 1–19.
Finance 43, 1–19.
Vo, X. V. 2017. Determinants of capital structure in emerging markets: Evidence from Vietnam. Research in International Business and Finance 40, 105–113.
Warner, J. B. 1977. Bankruptcy costs: Some evidence. The Journal of Finance 32, 337–347.
VIII. Appendix A
Table A.I VIF estimates
This table shows the Variance Inflation Factors (VIF) for the independent variables of this sample. Leverage is calculated as debt/equity. Effective tax rate is calculated as income taxes/earning before tax. Intangibility is calculated as intangible assets/total assets. Growth is calculated as percentage change in total assets in years. Size is a measure of total revenue. Profitability is calculated as operating income/total assets.
Effective tax rate 1.00
Intangibility 1.02
Growth 1.01
Size 1.02
