Capital structure and firm performance –
A study of Swedish public companies
Bachelor’s thesis, Business Administration Accounting
Spring 2014
Supervisor: Johan Åkesson
Authors: Richard Dumont, Robert Svensson
Bachelor’s thesis Business Administration, Handelshögskolan Göteborgs Universitet,
Accounting, Spring 14
Authors: Richard Dumont and Robert Svensson Supervisor: Johan Åkesson
Title: Capital structure and firm performance – A study of Swedish public companies Background and problem: Developments in capital structure during the last 30 years have resulted in a number of capital structure theories. At the same time, and in spite of all research on the topic, capital structure policies were one of the reasons for many company problems when the financial crisis hit in 2008. It is therefore interesting to look at how capital structure has evolved in the last decade as well as to test the functional relationship between capital structure and firm performance on a large scale.
Purpose: To map and explain the development of capital structure and firm performance in Swedish companies during the last decade.
Limitations: The thesis will only focus on companies listed on the Swedish stock exchange and with yearly sales amounting to at least SEK 10m.
Method: A large-scale quantitative cross-sectional study including some 300 Swedish companies and 8 years of financial statements data. Relationships have been tested with a multiple regression model and developments of financial data have been tracked and compared over an 8-year period.
Results and conclusions: There is a negative relationship between debt-to-equity and return on equity for Swedish firms during 2005-2012. Companies can thereby increase their return on equity by decreasing their debt-to-equity levels.
Further research: A study of optimal capital structure for Swedish firms, using the latest developments in capital structure theory and using similar data as in this study.
Key words: Regression model, capital structure, return on equity
Table of Contents
1. Introduction ... 1
1.1 Purpose ... 2
2. Literature review ... 3
2.1 Trade‐off theory ... 3
2.2 Pecking order theory ... 5
2.3 Financial crisis ... 5
3. Methodology ... 7
3.1 Methodology for describing developments in firm performance and capital structure ... 7
3.1.1 Use of measures ... 7
3.1.2. Definition of measures ... 8
3.1.2.1 Return on assets ... 8
3.1.2.2 Return on equity ... 9
3.1.2.3 Debt‐to‐equity ratio ... 9
3.1.2.4 Average interest on debt ... 10
3.2 Methodology for linear regression analysis ... 11
3.2.1 Description of control variables ... 12
3.2.2 Hypothesis development ... 13
3.2.3 Multicollinearity ... 14
3.2.4 Goodness‐of‐fit ... 14
3.2.5 Causality versus correlation ... 15
3.3 Data collection ... 16
4. Results ... 17
4.1 Variable developments ... 17
4.2 Regression analysis ... 20
5. Analysis ... 22
5.1 Discussion of financial ratios ... 22
5.2 Analysis of regression results ... 23
6. Conclusions ... 26
7. References ... 27
Appendix ... 30
1
1. Introduction
Capital structure is an important issue from a financial standpoint because it is linked to the firm’s ability to meet the objectives of its stakeholders (Simerly & Li, 2000).
Modigliani and Miller (1958) argued that in a perfect market, the value of a firm is unaffected by how it is financed. When imperfections of real markets are taken into account, however, capital structure can have a substantial impact on firm performance (Abor, 2005; Denis 2012). Modigliani and Miller (1963) found that higher leverage leads to increased performance due to tax benefits.
Developments in capital structure research during the last 30 years have resulted in a number of capital structure theories that predicts somewhat contradictory results (Baker
& Martin, 2011). For instance, trade-off theory suggests a positive relationship between firm performance and leverage (Margaritis & Psillaki, 2010), whereas the pecking-order theory predicts a negative relationship (Baker & Martin, 2011). Additionally, research has shown that the effect of leverage on firm performance might depend on the specific environment of the firm (Simerly & Li, 2000).
Because of the inconsistent theories that exist within the field it is important to provide empirical results that can help validate or disprove these theories. Bertmar and Molin (1977) carried out a comprehensive study during the period 1966 to 1972 to map the developments and relationships between a set of financial measures and capital structure.
They found that the level of debt financing increased in Swedish firms during the studied period. Perhaps Bertmar and Molin’s (1977) analysis of financial performance and capital structure of Swedish companies is the most widely known field study in this area. Apart from this, not much research has been done on capital structure and firm performance among Swedish companies. It is therefore interesting to update some of the research and compare and analyze the result in light of the new important developments in capital structure research (Denis, 2012).
In addition, the financial crisis during 2008 and 2009 also makes it interesting to look at
the developments of capital structure. This is because many problems that companies
experienced were created specifically from capital structure policies (Baker & Martin,
2011). Financial economists have critically evaluated capital structure theory as a result
(Baker & Martin, 2011). The years before, during and after the financial crisis thereby
provides a good period for studying the relationship between firm performance and
capital structure in different economic environments, which has been found to have an
effect on the relationship (Margaritis & Psillaki, 2010).
2
The present thesis aims to study firm performance and capital structure in a similar manner as Bertmar and Molin (1977). This will be done by answering the following research questions:
How have ROE and ROA developed between 2005 and 2012?
How has average interest on debt developed between 2005 and 2012?
How has debt-to-equity ratio developed between 2005 and 2012?
How do capital structure choices influence firm performance?
1.1 Purpose
The objective of the present study is twofold. The first objective is to quantitatively
describe developments of firm performance and capital structure before, during, and after
the financial crisis 2005 to 2012. Secondly the objective is to find how capital structure
choices have influenced firm performance during the period and to understand if the
relationship is different during the boom, crisis years and the following recession.
3
2. Literature review
This section primarily reviews literature connected to capital structure and the relationship between capital structure and firm performance. Moreover, the financial crisis is described with regards to causes and possible effects, on the economy and businesses in general and on capital structure in particular.
The studies of capital structure are usually said to have started with Modigliani and Miller (1958). They described how capital structure affects a company’s value in a perfect economy. In that case, companies get an increase in profitability from the higher leverage that exactly corresponds to the rise in discount rate due to the higher risk, thus a company’s valuation would be unaffected by its capital structure. However, Modigliani and Miller (1963) later concluded that, in practice, capital structure does affect company valuation since higher leverage gives greater benefits from tax shielding. Kraus and Litzenberger (1973) expanded on the concept of tax shields when they introduced the trade-off theory and the notion of an optimal capital structure. This theory, along with another important capital structure theory; the pecking order theory (Myers and Majluf, 1984), will be further reviewed in sections 2.1 and 2.2 respectively.
In an attempt to explain the relation between firm profitability and capital structure, Johansson and Runsten (2005) presented the following formula:
Equation 1.
In the formula, IR represents the average interest of a company’s loans, TL/E is the total liabilities-to-equity ratio, ROA is return on assets and ROE is return on equity. Important to note is that a company can increase their ROE by changing their capital structure. As long as the return on assets is higher than the average interest rate, the ROE will increase with higher leverage. Several studies have been done on the relationship between firm performance and capital structure (see for example Simerly & Li, 2000; Margaritis &
Psillaki, 2010). Bertmar and Molin (1977) studied capital structure and profitability in Swedish companies from 1966 to 1972. They found that debt-to-equity ratio and return on equity have a negative correlation.
2.1 Trade-off theory
In contrast to dividend, interest payments on debt reduces a company’s taxable income.
At the same time, debt increases the likelihood of bankruptcy for a company. According
to the trade-off theory, capital structure reflects the trade-off between tax-benefits and
expected costs of bankruptcy. (Kraus & Litzenberger, 1973)
4
Equation 2. The firm’s value in trade-off theory
1
2
2 1
The firm’s choice of leverage is then determined by maximizing V in the equation above.
In the model, R is a random cash flow of a firm. T denotes the constant corporate tax rate and D is the required debt payments. The first-order condition with respect to D is:
Equation 3.
1
A number of relationships can be derived from the formulas and can hence be used to explain capital structure decisions by the companies studied (Baker & Martin, 2011).
When the tax increases in the equation, the debt should also increase since higher tax will give higher tax advantages. There should also be a positive relationship between debt and profitability because expected bankruptcy costs and tax shields are more valuable for profitable firms. There is, however, mixed empirical results for the relationships predicted by the formula (Baker & Martin, 2011).
One of the reasons that capital structure research has been generating a variety of different results ever since the work of Modigliani & Miller (1958; 1963) is the
complexity of measuring tax benefits (Graham, 2000). The reason for the complexity has been mainly data problems and the complex tax codes (Graham, 2000). In addition, quantifying the interest taxation effects and understanding the bankruptcy process and financial distress is also prominent issues (Graham, 2000). Graham (2000) developed a new measure of tax benefit that included the entire tax benefit function. As a result, it is concluded that the typical US firm can double tax benefits by issuing debt until the marginal tax benefit begins to decline (Graham, 2000).
It has been a central issue in financial research that firms are consistently having lower debt levels than what is predicted as optimal levels by the trade-off theory (Ju et al., 2005). The traditional literature on optimal capital structure using the trade-off theory usually only includes bankruptcy costs and tax shields (Ju et al., 2005). The problem with these studies is that they are static and hence do not incorporate the rights of bondholders to force firms in to bankruptcy (Ju et al., 2005). As a result, few traditional studies have not yet provided compelling response to what the optimal value-maximizing capital structure is, Ju et al. (2005) being among the exceptions.
Ju et al. (2005) suggests that the traditional research with a trade-off theory perspective
usually suggests that firms are overleveraged. That is, firms would benefit from reducing
their debt-to-equity ratio. When managers make capital structure decisions with the
5
objective to maximize firm performance, however, Ju et al. (2005) finds empirical evidence that the median firm in Standard & Poor’s Compustat database is
underleveraged and hence would benefit from increasing their debt-to-equity ratio.
One major problem with the trade-off theory is that it assumes market efficiency and symmetric information (Baker & Wurgler, 2002). The decision to issue new equity depends on the stock market performance. The market timing theory suggests that companies are issuing debt depending on the business cycle. When the general economy is in bad condition, firms will not issue equity. When the economy is booming, however, equity issuance is high. That is, during good times, the debt-to-equity ratio should decrease (Baker & Martin, 2011).
2.2 Pecking order theory
The pecking order theory was popularized by Myers (1984) and Myers and Majluf (1984) and is based on asymmetric information between firms and investors (Baker & Martin, 2011; Frank & Goyal, 2008). The principle of pecking order theory is that equity is a less preferred way to finance a firm because investors believe that managers will only issue new equity when the equity is overvalued. Specifically, managers of a firm are supposed to use a pecking list when they need to finance their operations (Graham & Harvey, 2001). As a result, investors will pay a lower value for the issued equity (Myers &
Majluf, 1984). Empirical evidence also suggests that issuance of new equity result in stock price reductions (Baker & Martin, 2011).
The pecking-order theory does not suggest a specific debt ratio as optimal, but instead seek external financing only when there are insufficient internal funds (Graham &
Harvey, 2001). And when they do seek external financing, they always prefer debt over issuing new equity (Myers, 1984). The pecking order theory suggests a negative correlation between debt and profitability (Baker & Martin, 2011). This is because high- quality firms tend to use internal funds for financing its operations, whereas low-quality firms have to seek external financing, usually debt, in absence of profits. There is also a correlation between the level of asymmetric information and the incentive to issue new equity (Baker & Martin, 2011).
Tong and Green (2005) set up a study to test whether the trade-off theory or the pecking order theory best predicted the performance of Chinese companies. They found
statistically significant support for the pecking order theory over the trade-off theory.
Specifically, they found that there was a negative correlation between debt and profitability.
2.3 Financial crisis
The financial crisis started in 2007 in the United States before spreading to Europe in the middle of 2008 (Crotty, 2009). It caused indexes on several global stock markets,
including the U.S and Sweden, to fall by over 50% and is considered the worst crisis
since the Great Depression in the 1930s (Crotty, 2009). One of the main reasons for the
6
crisis was the loose monetary policy employed by the Federal Reserve. For a couple of years before the crisis, the federal funds interest rate was set well below the level
historical experience would suggest given the situation (Taylor, 2009). This led to a large amount of mortgage sales and a housing boom, which subsequently resulted in a bust and thereafter a bank run that gave many financial institutions liquidity problems (Crotty, 2009).
Cornett et al. (2011) conclude that the liquidity crisis that many banks experienced led to a decrease of credit supply, thus making it more expensive for firm’s to borrow money.
New lending volumes in the US fell with 47% during the fourth quarter of 2008 to a level 79% lower than during the credit peak in 2007 (Ivashina & Scharfstein, 2010). The effect of this is highlighted in a study by Campello et al. (2010), where a majority of the
surveyed companies were said to be affected by the worsened access to credit markets.
Affected firms were restricted in terms of investments and a majority of them were forced to cancel or turn down interesting projects, resulting in hindered economic growth
(Campello et al., 2010).
Most studies have focused on the effects of the decreased credit supply (see for example Popov & Udell, 2012; Chor & Manova, 2012) but apart from this, the crisis also brought about lower demand for goods and increased risk. In an empirical study, Kahle and Shulz (2011) find that net equity issuance starts to decrease before net debt issuance and
remains on a low level during the crisis. This finding implies that credit supply may not be the dominating factor behind firms’ restrained financial and investment policies.
Reasons for this would be the fact that the increased risk results in a higher cost of equity
and that the expected cash flows as well as investment opportunities decrease (Kahle and
Shulz, 2011). Altogether, the drop in credit supply resulting in higher interest rates and
the regression of capital demand mean that lending volumes decreased even more than
during a normal recession (Ivashina & Scharfstein, 2010).
7
3. Methodology
The methodological approach for the present study is divided into two main parts. The source data is extracted from the same database and the same definitions apply to the whole study. The first part of the method presents the approach taken to answer the first part of the study; describing how the financial ratios have developed during the studied period. This section includes definitions of the financial ratios and motivations of the various choices taken in terms of P&L and balance sheet items. The second part of the section presents the selected research approach for the regression analysis used to study the relationship between debt-to-equity and return on equity. Lastly, this section also includes a description of how the data were collected.
The objective of this study is to map and describe the development of key financial ratios and the relationship between return on equity and debt-to-equity. Therefore, results and analysis are separated into different sections in order to be able to emphasize the empirical findings of the study, rather than the analysis.
3.1 Methodology for describing developments in firm performance and capital structure
This section will describe the methodology that was used in order to fulfill the first part of the research objective, describing the developments in company performance and capital structure for Swedish companies. Firstly, the measures that have been used are presented along with explanations as to why they are relevant. Secondly, the chosen measures are defined in more detail.
3.1.1 Use of measures
When evaluating a company’s performance, Johansson and Runsten (2005) argue that net income is useless as a measure. Standing alone, it does not give information about
whether a company generates any return to its shareholders, which is necessary in order to ensure the company’s survival. Net income is instead meaningful first when it is put in relation to the capital that was required in order to produce the profit. Based on this discussion, a company’s performance will in this study be measured by using the financial ratio return on equity after taxes (ROE). According to Johansson and Runsten (2005), this is the most important financial ratio from a shareholders’ perspective since it can easily be compared to the market cost of capital. The measure is calculated as the percentage of the net income compared to shareholder equity.
Return on equity is influenced by firms’ decisions regarding capital structure. A
performance measure free from the influence of financial decisions, return on assets
(ROA), has also been used in this study to assess companies’ performance. While return
on equity can be made to look good by utilizing high leverage, return on assets considers
both debt and equity and is thus not affected by the leverage. However, return on assets is
a more industry-dependent variable and does not offer the same comparability as return
on equity (Hawawini et. al, 2003). Altogether, by utilizing both the described measures
8
this report aims to present a clear picture of Swedish companies’ performance during the studied period.
For describing the development in capital structure for Swedish companies, the debt-to- equity (D/E) ratio will be calculated. Also included as a variable for this study is the companies’ average interest on debt. In the literature review 2.3, it was mentioned how low interest rates were one of the drivers for the financial crisis and therefore it is interesting to see how this number have changed over the studied period. Furthermore, there is a probable correlation between average interest on debt and the debt-to-equity ratio. For example, Bertmar and Molin (1977) found that average interest on debt and debt-to-equity ratio have a positive correlation. Therefore, the average interest measure may help to explain the development in capital structure.
The results are presented with the median and one aggregate number for each measure.
The aggregate number is the result of all companies’ numbers weighted together, e.g. the combined earnings before interest expenses divided by the combined average total assets give the aggregate number for ROA. Thus, larger companies will contribute to a greater extent to this number. The median does not take any company specific variables into account but is simply the middle number in a sample arranged from lowest to highest value. Because of the characteristics of the data, the median rather than the average was used (Bertmar & Molin, 1977). Several smaller companies have numbers that can be considered outliers and would skew the result if using the average. The aim of the median is to show the value for the “typical” company in the data set.
Data is gathered from companies’ yearly income statements and balance sheets. The income statement shows the sum of a company’s revenues and expenses over a period, usually a calendar year. The balance sheet summarizes a firm’s assets, equity and liabilities at a certain point in time, usually the end of a calendar year. Usually, return on assets and the average interest on debt is calculated based on an average of the opening and closing balances (Johansson & Runsten, 2005). This approach will also be used in this study and in order to be consistent, all numbers taken from balance sheets will represent averages.
3.1.2. Definition of measures
The four financial ratios are defined more closely in this subsection. Each financial ratio will be broken down in its component parts, in order to increase the transparency and motivate the choices of the study.
3.1.2.1 Return on assets
The first performance measure, return on assets is defined the following way:
9
Equation 4.
For the calculations, the definition from Bertmar and Molin (1977) has been applied.
Here, earnings before interest expenses (named EBI) consist of the following entries in the income statement: profit/loss after financial expenses (RR1), extraordinary income (E1), extraordinary expenses (EE), group contribution (E2), external interest costs (IC1) and interest expenses to group companies (IC2).
Equation 5.
1 1 2 1 2
Accordingly, the metric does not consider taxes, neither year-end accounting
adjustments. Group contribution is excluded since it cannot be derived directly from the day-to-day activities of the company while other extraordinary incomes and expenses can, at least from a long-term perspective (Bertmar & Molin, 1977). Interest costs are excluded in order to make the metric independent of the firm’s capital structure. Average total assets are simply generated from the Total Assets line in the balance sheet and represents an average, i.e. average total assets for 2012 corresponds to the average of the numbers from the year-end reports of 2011 and 2012.
3.1.2.2 Return on equity
Return on equity is calculated from the following numbers:
Equation 6.
1
The bottom line from the income statement, net income, is used since that represents the profit attributable to the shareholders. Again, the definition from Bertmar and Molin (1977) has been applied to calculate equity in the denominator. More specifically, the entries gathered from the balance sheet were Total Equity and Untaxed reserves. Untaxed reserves are postponements of taxable income and therefore, one part of it can be
considered as unpaid tax debt, while the other is considered as equity (Hoogendoorn, 1996). As tax rate, the Swedish corporate tax for each year has been used, meaning 28.0% for years 2005-2008 and 26.3% for years 2009-2012 (Ekonomifakta, 2014).
Again, an average is used for the numbers taken from the balance sheet.
3.1.2.3 Debt‐to‐equity ratio
The debt-to-equity measure is used to represent the capital structure of the firm’s over the
studied period. There are a number of ratios used for this, where total liabilities divided
10
by equity is the basic one most often used by Swedish companies (Johansson & Runsten, 2005). However, this measure includes also non-interest-bearing debt, which does not inflict any financial costs on the company and thus is not associated with the same financial risk. Therefore, the number may not always provide an accurate picture of a firm’s financial position (Johansson & Runsten, 2005). For example, a company can increase their total liabilities-to-equity ratio if they gain a better market position that allows them to increase the days payable outstanding. This would result in a higher amount of accounts payable and thus a higher total liabilities-to-equity ratio, which could lead to the belief that the firm’s financial position has weakened when the company in fact has strengthened their market position. Due to this, the globally more widely used definition of debt-to-equity ratio, the interest-bearing liabilities divided by equity, is better suited for this study.
The data used for this study does not include any information about which parts of the liabilities in the balance sheet that is interest bearing and which is not. Consequently, we needed to decide on another measure that as accurately as possible could represent a firm’s financial position while being feasible to use given the data available. In the balance sheets, a distinction is made between short- and long-term liabilities. Short-term liabilities are loans and obligations that last less than a year. It includes for example accounts payable, tax debt and advance payments, which are normally categorized as non-interest bearing (Johansson & Runsten, 2005). Therefore, these liabilities have been excluded from the debt-to-equity ratio used in this study, which now can be defined by the following formula:
Equation 7.
/
1
Long-term debt includes the entries bond loans, long-term liabilities to credit institutions, long-term liabilities to group/associated companies and other long-term liabilities in the balance sheet. Equity has been calculated the same way as in 3.1.2.
3.1.2.4 Average interest on debt
Like with the debt-to-equity ratio, calculating average interest on debt based only on non- interest-bearing liabilities can provide a better picture of a firm’s financial position (Johansson & Runsten, 2005). Again, it is not possible to distinguish between interest- bearing and non-interest-bearing debt in the data so long-term debt will serve as an estimate. Also, using the same type of debt as for the debt-to-equity ratio gives the opportunity to better study a possible correlation between the two in the analysis. The following formula will be used:
Equation 8.
11
Interest costs include the entries interest expense to group companies and external interest costs. The long-term debt is calculated the same way as in 3.1.3.
3.2 Methodology for linear regression analysis
In order to test the relationship between different variables, a linear regression analysis will be carried out. Hypotheses will be formed about the linear relationships in the following form:
Equation 9.
The basic question of this equation is how y varies with x. In any econometric study, there are basically three issues that have to be covered (Wooldridge, 2012). The first issue concerns how other factors that influence y is allowed in the model. The second issue is to describe or determine the functional relationship between y and x. The third issue is to determine how the ceteris paribus effect between the variables can be captured.
The simple regression model represents the ambiguity of these three issues (see equation above).
The dependent variable, y, is representing the return on equity and is the variable that is supposed to be explained in the analysis (Montgomery & Runger, 2007). Debt-to-equity ratio is the dependent, or explanatory, variable. That is, the objective is to find out if, and in that case how, the dependent variable is causing any variance in the independent variable. Specifically, the intention is to understand if changes in return on equity can be explained by changes in debt-to-equity (Wooldridge, 2012).
As mentioned initially in this chapter, however, one of the main issues with the simple regression model for empirical studies is that it is difficult to make ceteris paribus conclusions about how the studied variables actually affect each other (Wooldridge, 2012). In our case, a simple regression analysis would assume that all other factors that affect ROE are completely uncorrelated with D/E, which is unrealistic. Therefore, a multiple regression model is necessary because it allows for several factors, which all together affect the independent variable. The power of the multiple regression model is that it makes it possible to change one dependent variable at the same time as the other dependent variables are held fixed (Montgomery & Runger, 2007). The multiple regression model is presented in the following equation.
Equation 10.
⋯
The disturbance term in the regression model contains all other factors that are not included as control variables. These are the unobserved factors of the model
(Wooldridge, 2012). The β
1–term in the equation will explain the functional relationship
between leverage and return on equity.
12
Throughout the study, ordinary least squares (OLS) method is used to estimate the unknown parameters in the linear regression model. The reason for choosing the OLS estimator method is justified by the Gauss-Markov Theorem, because it creates the result with the smallest variance in the coefficients (Wooldridge, 2012). That is the OLS estimators are the best estimators which minimize the vertical square errors between the data points and the regression line (Wooldridge, 2012).
3.2.1 Description of control variables
The criteria for selecting the control variables are that they are factors, which affect the ROE and are correlated with D/E. If these types of factors were not explicitly included as control variables in the multiple regression model, it would generate a biased OLS estimator of β
1(Wooldridge, 2012). That is, the benefit of explicitly including the
control variables in the regression model, as opposed to keep them in the error term of the model, is that the variables will be held fixed as the model is executed. Several other studies of similar nature were used to derive proper control variables and avoid under- specification of the model and biased results (Simerly & Li, 2000; Margaritis & Psillaki, 2010; Abor, 2005).
The control variables used in this study is size, growth, asset turnover, and business risk.
The size of a firm has been found to be influencing several important aspects of a firm, such as structure, decision-making, and performance (Simerly & Li, 2000; Abor, 2005).
The sizes of the sample companies are measured by the natural logarithm of yearly turnover (Margaritis & Psillaki, 2009). Size is therefore included explicitly in the study as a control variable. The sales is used in logarithmic form to make the data more normally distributed (Wooldridge, 2012). Researchers of social sciences often use the log values for large integer values (Wooldridge, 2012; Freund et al. 2006).
The second control variable is business risk, which is defined as the fluctuations in return on assets (Johansson & Runsten, 2005). Business risk will influence both D/E and ROE for firms. Since the business risk is often industry specific, firms will look at their total risk preference and then adjust the gap between the industry-specific business risk and the total risk they are willing to take (Johansson & Runsten, 2005). For instance,
commercial banks tend to have a very low business risk and hence is compensating with taking a higher financial leverage in order to generate an attractive return on equity for shareholders (Johansson & Runsten, 2008). Some similar studies have used industry as a control variable (Bradley et al., 1984). The reason industry is not included as a control variable in this study is two-fold. First, the industry data that could be generated in the employed database was of poor quality. The second reason is that industry is often correlated with business risk (Johansson & Runsten, 2005), and including such a variable would increase the multicollinearity of the results.
The third control variable is growth, defined as yearly growth in total assets. Firm growth
has been found to have a positive correlation with firm performance (King & Santor,
13
2008; Claessens et al., 2002). Growth has also been used as control variable in other studies (Abor 2005).
The fourth control variable is total asset turnover. The rationale for including this variable was the absence of good quality data on industry and asset turnover is rather correlated with different industries (Johansson & Runsten, 2005). For instance, capital intensive energy firms tend to have asset turnover ratios in ranging from 0.4 to 0.7, whereas manufacturing firms might have between 0.8 and 1.6 (Johansson & Runsten, 2005).
The general form of the multiple regression model was then developed to the following model, containing the main variable under investigation and the control variables.
Equation 11.
∗ ∗ SIZE ∗ GROWTH ∗ ASS. TO ∗ BUS. RISK
If a coefficient of a factor in the model is zero even after controlling for the other variables in the model, that factor is probably an inclusion of an irrelevant variable (Wooldridge, 2012). All coefficients of factors included in the model will be analyzed after the computations of the model in order to identify potential over-specifications in the model.
3.2.2 Hypothesis development
The problem set up for investigation in this study is to test whether the D/E has an effect on ROE. The null hypothesis is therefore H
0: β
1= 0. The alternative hypothesis is H
1: β
1≠ 0, which means that there is a statistically significant relationship between D/E and
ROE. A t-test was conducted in order to ensure that the estimated coefficient is not due to
sampling error. Rather than only selecting a specific critical value for the t statistic, an
analysis of the p-value will be conducted. This approach is more informative than using a
predetermined critical value for rejecting the null hypothesis (Wooldridge, 2012). The p-
value expresses the smallest significant value at which the null hypothesis would have
been rejected (Montgomery & Runger, 2007). Practically, this means that a small p-value
is evidence against the null hypothesis and vice versa. The significance level of the
regression model is selected as 99%. That is, the required p-value for rejecting the null
hypothesis was selected to be 0.01. The reason for selecting a high significance level is
that the sample size is fairly large (Wooldridge, 2012). Even though there are no standard
rules for selecting significant levels, 99% is common when having a few thousand data
points (Wooldridge, 2012). For example, if the p-value is 0.01, it means that if the null
hypothesis were true, one could observe an equally large t statistic 1% of the time,
providing some evidence against the null hypothesis. Specifically, the p-value is the
probability of obtaining a test statistic at least as extreme as the one that was actually
observed (Wooldridge, 2012). The null hypothesis is summarized in the table below.
14
Table 1
The F test is conducted in order to evaluate how much of the variation that was captured by the regression. The F test statistic is calculated as the mean squared error of regression (MSR), which is the mean squared error of the regression, divided by the mean squared error (MSE), which is the average square of errors of the OLS estimator. MSE is the squared distance that the OLS estimator is from the population value (Wooldridge, 2012), and it is that same as the variance plus the square of any bias. The F-value is then
translated to a significance level automatically by the model. The F value measures the significance of the multiple regression overall. The significance of the F test is equal to the p-value of the test.
Equation 12.
3.2.3 Multicollinearity
A multicollinearity test was conducted in order to ensure that the regression model had no multicollinearity. The test was performed with the NumXL Pro add-in in Microsoft Excel with the Variance Inflation Factor (VIF) method (See Appendix A). High correlation between two or more of the independent variables is referred to as multicollinearity (Wooldridge, 2012). The problem of multicollinearity is commonly discussed in econometric studies (Bradley et al., 1984), and sometimes even overly empathized by econometricians (Wooldridge, 2012). The effect of multicollinearity is basically the same as having a small sample size because both drive a high variance of the coefficients (Wooldridge, 2012). The only way to reduce the multicollinearity problems in this study would be to exclude one of the control variables. By doing excluding one of the
explanatory variables, however, might cause the results of the model to be biased (Wooldridge, 2012). In spite of the attempt to minimize multicollinearity, it is always a fundamental problem in cross-sectional regression studies.
3.2.4 Goodness-of-fit
H0 β1
= 0
There is no a statistically significant relationship between D/E and ROE H1:
β1≠ 0
Null hypothesis rejectedSignificant at 0,01
Hypothesis test ‐ Testing correlation between D/E and ROE
15
The coefficient of determination, or R-squared, in the regression analysis measures how much of the variation in ROE that can be explained by the debt-to-equity ratio
(Wooldridge, 2012). It is important to note that R-squared is not a variable that will influence the generation of the model. Some studies tend to construct their studies in order to get a desirable R-squared (Wooldridge, 2012). As a result, they are likely to introduce multicolliniearity in the model (Wooldridge, 2012). R-squared is defined as follows:
Equation 13.
In addition, the adjusted R-squared will also be generated in the model. This is calculated in addition to R-squared because R-squared is always increasing as more regressors are added to the model. The adjusted R-squared penalizes the R-squared for having
numerous explanatory variables according to the equation below (Wooldridge, 2012), where k denotes the number of explanatory variables and n is the total number of observations.
Equation 14.
1 1
1
It is important not to overemphasize the goodness-of-fit by including too many control variables in the model (Wooldridge, 2012; Freund et al. 2006). As mentioned earlier, the selection of control variables is a trade-off between biased results and introducing
multicollinearity. That also means that there is always a good idea to include explanatory variables that affect y and are uncorrelated with all the other independent variables (Wooldridge, 2012).
3.2.5 Causality versus correlation
The present study has a similar challenge as many other statistical studies, which is to evaluate whether the leverage actually has got a causal effect on firm performance as opposed to having merely a correlating relationship, which can be achieved by keeping all factors but the test variables equal (Wooldridge, 2012). It is rarely possible, however, to keep all other factors equal when performing analysis on economic data, but it is rather a question of holding enough other factors fixed to make a case for causality
(Wooldridge, 2012). In order to make the case for causality in this study, a number of control factors derived from theory will be included in the regression analysis (see
“description of control variables”). This is an attempt to isolate the ceteris paribus effect
of the studied variables, which is carried out using the Data Analysis add-in in Microsoft
Excel. By explicitly including factors that has been found to affect the ROE by other
authors, these factors can be held fixed and a ceteris paribus effect can be analyzed
(Freund et al., 2006).
16
One part of this study was to find evidence that the D/E does have an effect on the ROE.
An equally important part, however, was to describe the functional relationship between the variables. The coefficient β
1will explain that if that relationship is positive or
negative as well as to give an indication of the magnitude of the causality. It is also important to emphasize the difference between statistical significance and economic significance. Statistical significance is only related to the size of the t statistic, whereas the economic significance is more related to the size and sign of the coefficients (Wooldridge, 2012).
3.3 Data collection
The sample used for this study consists of Swedish public companies. Numbers are obtained from the database Retriever Business that has annual account figures for all companies listed on the Swedish stock exchange. Since the information dates back to 2004 and averages are used for calculating some of the variables, the presented numbers in the analysis start at year 2005. The majority of the companies have not registered numbers for year 2013 in the database, thus making 2012 the last year analyzed. The original sample consisted of 492 companies. Some of the companies have incomplete numbers for the period, i.e. they became listed on the stock exchange later than 2004.
After excluding these companies the sample consisted of 369 companies. Additionally, the sample was adjusted to exclude very small companies with below SEK 10m in revenues. Specifically, all companies that did not have at least SEK 10m in revenues every year of the studied period were excluded. The reason was that some of these smaller companies were having highly questionable data logged in the “Retriever Database”. For instance, some of these companies were not having any sales at all some of these years. After excluding these companies the final sample consisted of 299 companies. For a more detailed description of the studied companies, see Appendix D &
E.
For the regression analysis, another type of adjustment to the data was made, where a number of outliers were excluded. A residual output analysis was used to identify the outliers, see Appendix C. These single observations had a major influence on the result of the model. An outlier is defined as an observation that is so extreme that when dropping it, the OLS estimators change by a practically large amount (Wooldridge, 2012). Because the original sample is a few thousand data points, the impact on the quality of the study should not be noticeable (Wooldridge, 2012). After adjusting the data set, 2337 data points remains.
The data is based on figures presented by the studied companies, and can therefore be subject to accounting choices made by the companies. That is, we can only make
conclusions about the volatility in certain financial measures presented by the companies.
However, we have not used any of the KPIs already calculated by the companies and thus
the study will not be affected by company-specific choices regarding KPI definitions.
17
4. Results
In this section, the results of the study are presented. First, the developments in the different studied variables are described. Secondly, the results from the regression analysis are presented.
4.1 Variable developments
The description of firm performance is made based on the measures return on equity and return on assets.
Figure 1. Return on equity for Swedish companies 2005-2012
Figure 1 shows the return on equity by two metrics, the aggregate of all studied
companies and the median of the sample, over the years 2005 to 2012. It shows a steady decline in the aggregate metric between 2005 and 2007 before the financial crisis in 2008, which results in a steeper downwards curve, hitting the bottom at around 3%. The aggregate curve rebounds the year after with an increase of approximately 9 percentage points and in 2010 the number is back to the level of 2007. In 2011, the aggregate
number takes a downturn again from 17% to around 11% before a slight increase in 2012 to end the period.
For the median number, the fluctuations are smaller than for the aggregate number. The largest difference between two years for the median is 8 percentage points and 13 for the aggregate, both between 2007 and 2008. There are differences in the direction of the curves for several of the analyzed years. In 2005 as well as in 2011, the median number increases while the aggregate decreases and the opposite occurs in both 2009 and 2012.
The values for the median are generally lower than for the aggregate number with the exception of 2008 where the median value is 2 percentage points higher. Over the whole
2005 2006 2007 2008 2009 2010 2011 2012
Aggregate 21,3% 19,2% 16,5% 3,3% 12,2% 17,0% 10,8% 12,8%
Median 13,0% 14,4% 13,7% 5,7% 4,9% 9,3% 9,7% 6,3%
0,0%
5,0%
10,0%
15,0%
20,0%
25,0%
ROE
18
period, the aggregate number is on average 4.5 percentage points higher. Both curves have a substantial decrease in 2008 corresponding to the financial crisis and they end the period with significantly lower values than they started with in 2005.
Figure 2. Return on assets for Swedish companies 2005-2012.
Figure 2 shows the median and aggregate number for return on assets in Swedish public
companies over the period from 2005 to 2012. Similar to return on equity, the aggregate
number shows a steady decline between the years 2005 and 2007. During 2008, there is a
significant drop from 12% to 5%, which is the lowest point of the period. Then in the two
following years the measure recovers by increasing 3 percentage points each year before
turning down again in 2011. Altogether, the aggregate number drops 6 percentage points
from 14% to 8% between 2005 and 2012. While the aggregate number decreases between
2005 and 2007, the median increases during these years, though not by much. As with
return on equity, the median number for return on assets varies less than the aggregate
number. It also takes a lower value for the whole period with the aggregate number being
on average 4 points higher.
19
Figure 3. Debt-to-equity ratio for Swedish companies 2005-2012.
The description of capital structure is made based on the measure debt-to-equity ratio.
Figure 3 shows the aggregate and median value of the debt-to-equity ratio for Swedish companies during the years between 2005 and 2012. The values show that there is a significant difference between the aggregate number and the median company with the aggregate number being on average 157% larger over the period. While the median has been fairly steady over the period, the changes in the aggregate number is considerable.
The ratio decreases slightly during 2006 before starting to trend upwards. In 2008, there is an increase of approximately 0.100 followed by a slight increase also during 2009. The peak is reached in the last year of the period, 2012, when the debt-to-equity ratio
increases to 0.473, a value 52% higher than in 2005.
20
Figure 4. Average interest on debt for Swedish companies 2005-2012.
Figure 4 shows the average interest on debt for Swedish public companies from 2005 to 2012, both by the aggregate and median number. In this case, the median is consistently larger than the aggregate number. The curves are very similar, both in terms of their variation and in the direction of the slopes. Every time the median increases, so does the aggregate number. There is an evident increase in average interest at the beginning of the financial crisis in 2008, most clearly shown in the median curve. Thereafter comes a sharp decline until the lowest level for both curves is hit in 2010. From the start of the period to the last year in 2012 the aggregate number of average interest decreases by 2 percentage points. However, the median company has approximately the same average interest on debt in 2012 as in 2005.
4.2 Regression analysis
The result of the full period in of the multiple regression model is presented in table 2.
The Analysis of the Variance (ANOVA) table includes valuable data for understanding the characteristics of the test. As expected, the model has five degrees of freedom which corresponds to the number of explanatory variables in the regression model.
Table 2
R‐squared 24.1% F 148
Adjusted R‐squared 23.9% Significance F 1,8E‐136
N 2337 df 5
Regression statistics
Regression output
ANOVA
21
The purpose of the regression model is, naturally, to explain the variation in ROE with the variation in D/E. The F statistic is 148, which gives a significance for the entire multiple regression model of well under the selected significance level of 0.01.
The R-squared of the regression is 24.1% for the full period, which means that 24.1% of the total variation in ROE can be explained by the explanatory variables. In addition, the adjusted R-squared of the model is just slightly lower at 23.9%
The coefficients of the multiple regression model is read from table 3 and represents the estimates of the coefficients in the regression equation. The most interesting coefficient in this study is the coefficient for debt-to-equity, which explains the functional
relationship between ROE and leverage. The null hypothesis cannot be rejected in the period between 2010 and 2012. For the other two sub-periods and the full period, the null hypothesis can indeed be rejected and hence debt-to-equity has a significant statistical impact on ROE. The negative sign for β indicates a negative relationship between the debt-to-equity and ROE for all periods, with the strongest functional relationship during the crisis years in 2008 and 2009. That is, companies with lower debt-to-equity any given year, presents a higher ROE in the end of that year.
Table 3
Furthermore, it is also interesting to discuss the impact of control variables on the independent variable, ROE. Size has a p-value under 0.01 in all periods and hence the null hypothesis can be rejected, which indicate that larger companies are having higher ROE during the studies period. In contrast, the null hypothesis cannot be rejected for growth during any period at the 0.01-level. There are however large differences between the periods. When looking at the full period, the null hypothesis would have been rejected at a 3% significance level. For the periods 2005 to 2007 and 2012 to 2012, however, the null hypothesis would not have been rejected even at the 10% significance level, which is commonly used as the highest acceptable value by many researchers (Wooldridge, 2012).
The asset turnover ratio is positive and significant for all periods as the p-value is less than 0.01.The positive sign of the OLS estimators indicate that companies with higher total asset turnover are generally having a higher ROE. Lastly, the null hypothesis can be rejected for business risk in all periods except for the period 2010 to 2012. Business risk has the steepest slope of the OLS estimators in all periods, and has a negative sign, which indicate that companies with lower business risk are having higher ROE.
R2 Year β p‐value* β p‐value* β p‐value* β p‐value* β p‐value*
2005‐2007 ‐0,080 0,001 0,045 0,000 0,000 0,697 0,060 0,000 ‐3,721 0,000 33%
2008‐2009 ‐0,136 0,000 0,056 0,000 0,000 0,040 0,070 0,001 ‐1,121 0,000 24%
2010‐2012 ‐0,018 0,478 0,064 0,000 0,000 0,304 0,093 0,000 ‐0,318 0,099 23%
2005‐2012 ‐0,070 0,000 0,054 0,000 0,000 0,031 0,081 0,000 ‐1,444 0,000 24%
*significant at p<0.01; marked with bold text
Summary statistics of dependent variables ‐ size and significance
Debt‐to‐Equity Size Growth ATO Business risk
22
5. Analysis
The analysis of the presented results is divided into two parts. First, the financial ratios are analyzed and possible explanations for the developments are discussed. Regression model output is discussed in the second part, where the results are analyzed from the perspective of capital structure theories and other empirical studies.
5.1 Discussion of financial ratios
The performance of the companies during the studied years has varied greatly, in part due to the financial crisis, which takes place in the middle of the period. The median metric has a smaller variance in both of the performance measures. Looking at the
characteristics of the data, this is not all that surprising. A small number of companies have a large impact on the aggregate number, whereas all participating companies affect the median number equally. For example, in 2012 ten companies represented 70% of the aggregate net income. Consequently, a significantly changed result for one of the largest companies has considerable impact on the aggregate performance measures. It is also clear from the results that the financial crisis really hit Swedish companies in the year of 2008. Performance had been trending downwards in the years before as well but not to the same extent. There is a larger drop for the aggregate number of both the ROA and ROE measure, indicating that larger companies were more affected in that year.
However, they also seem to recover quicker as evidenced by the fact that the aggregate number changes direction in 2009 while the median keeps sloping downwards. A trend during the period is that performance turns for smaller companies seem to lag behind larger companies. This can be seen in 2005-2006 and 2011-2012 apart from the already mentioned 2009. The observation holds for both measures, even though the pattern is clearer in return on equity. Another observation is that the aggregate number outperforms the median over the whole period with the only exception being ROE in 2008, meaning larger companies in general perform better than smaller companies.
The slopes for ROA and ROE look very similar, which is to be expected since they both measure the profitability of a firm in some way. Furthermore, the value of ROE is generally higher than for ROA, which can be explained by financial decisions that generates leverage effects on ROE, see Equation 1. Leverage also explains the fact that the ROE measure is fluctuating more than the ROA. Only during 2008 does ROE fall below the value for ROA, indicating that the financial part of the equation has negative impact and thus that the ROA is lower than the average interest on loans
1.
Fluctuations in the performance measures can to a large degree be attributed to changes in the result rather than in equity or assets. While the aggregate profit/loss is very much affected by the unstable business cycle, the growth in assets and also equity is fairly stable over the period. The steady equity growth means that the increase in long-term
1