Evolution of Leverage and its Determinants in Times of Crisis

Transcription

1 Evolution of Leverage and its Determinants in Times of Crisis Master Thesis Tilburg University Department of Finance Name: Tom Soentjens ANR: Date: 27 June 2013 Supervisor: Prof. M. Da Rin

2 ABSTRACT This thesis investigates the evolution of leverage and its determinants in times of crisis. I focus on a sample consisting out of 2,205 US-listed firms that present 7 years of non-missing and consecutive data. The empirical analyses focus on the time period 1999Q3 2012Q4 in which a benchmark and crisis period respectively cover the quarters 1999Q3 2008Q2 and 2008Q3 2012Q4. Consistent with the findings of Lemmon, Roberts & Zender (2008) I show that leverage ratios are persistent and converging over a time horizon of 11 years. Furthermore I present evidence that leverage ratios do seem to increase in times of crisis and that this is especially the case for the High and Very High levered firms. With respect to the traditional determinants of leverage I present results in line with as well the Trade-Off Theory of Debt as Pecking Order Theory. The impact of the determinants Growth Opportunities, Tangibility and Industry Median Leverage increases in times of crisis. However, for the variables Growth Opportunities and Tangibility these results are only significant with respect to market leverage ratios. The impact of the variable Initial Leverage decreases during the crisis period in as well the book as market leverage model. Furthermore indicate the results of a Variance Decomposition Analysis that capital structure decisions are driven by unobserved and firm-specific characteristics that are not completely captured by the traditional determinants of leverage. Moreover is presented that the vast majority of variability in leverage ratios can be explained by cross-sectional differences in firm characteristics, rather then by time-dependent factors. Key Words: Capital Structure; Leverage; Determinants of Leverage; Trade-Off Theory; Pecking Order Theory; Evolution of Leverage; Crisis 2

3 INTRODUCTION Since Modigliani & Miller s Irrelevance Proposition, capital structure design became an important part of the financial academic literature. Several studies tried to explain cross-sectional and time-varying differences in leverage based on theoretical and empirical research. The most predominating theories in this field are currently the Trade-Off Theory of Debt, the Pecking Order Theory and Market Timing Theory. In addition to these three general theories, a great amount of publications tried to link capital structure decisions to firm-specific characteristics. These characteristics are better known as the Determinants of Leverage. Despite the fact that the Holy Grail with respect to explaining capital structure decisions is not found yet, the relevance of several determinants of leverage is commonly accepted. This paper is an addition to the available literature in a sense that it incorporates the effect of a macro-economic crisis in capital structure decisions. Initially will be investigated how leverage ratios evolve in a crisis period and if this is significantly different compared to a normal market situation. Secondly, I try to explain crosssectional differences in capital structure design by the use of the most important and commonly accepted traditional determinants of leverage. Besides that I will verify whether of the impact of these determinants is in line with theoretical predictions and previous research, I will also analyze whether of the impact of these determinants changes in times of crisis. In order to realize these goals I conduct several empirical analyses that are, for the vast majority, in line with Lemmon, Roberts & Zender (2008). I use a sample that consists out of 85,089 firm-specific observations relating 2,205 US-listed companies, covering the time window 1999Q3 2012Q4. The remaining part of this paper is structured as follows. In section 1 I review the most important academic research relating capital structure design and determinants of leverage. Furthermore I present my hypotheses in this section. Section 2 will describe the dataset and sample selection. In section 3 I conduct a sorting exercise to reveal any remarkable patterns in leverage and its determinants during the crisis period. In section 4, 5 and 6 I present respectively the results of an Event-Time Study, Multivariate Regression Analysis and Variance Decomposition Analysis. Section 7 will present the most important findings and conclusions of this master thesis. 3

4 Ι. LITERATURE REVIEW & HYPOTHESES FORMULATION For decades, capital structure design and its determinants form an important part of the academic research in the field of corporate finance. The origin of this research can be found in 1958, when Modigliani and Miller presented their Capital Structure Irrelevance Theorem. Ever since, capital structure design became a cornerstone in the academic literature. A new flow of publications showed up which presented completely new theories about capital structure decisions or built further on Modigliani and Miller s Irrelevance Theorem. This section is divided in three parts. The first part focuses on the main capital structure theories available in current literature. The second part introduces the most predominating theories concerning the traditional determinants of leverage, together with the hypotheses underlying this research. In the last part of this section an overview of the formulated hypotheses is presented. Ι.Ι Capital Structure Theories As mentioned before, the capital structure debate finds its origin in the presentation of Modigliani & Miller s Capital Structure Irrelevance Theorem. This theory argued that managers are indifferent about the capital structure composition as long as the firm operates in a perfect market without taxes, bankruptcy costs, information asymmetry and agency costs. Modigliani and Miller proved this theory by showing that arbitrage opportunities exist if two identical firms with different leverage ratios would have deviating market values and/or returns to investors. However, are the assumptions of the MM Proposition realistic in practice? This question formed the start of a new stream of academic research in the field of capital structure design. As an example, Bolton & Scharfstein (1990) published a paper in which they proved the influence of debt financing in the trucking industry of the United States. They found evidence that, when relaxing Modigliani & Miller s assumptions, not the most efficient firms but firms with deepest pockets have a greater chance of survival. Or as Bolton & Scharfstein describe it: the process of natural selection of the fittest, and not the fattest, does not work in practice. Besides Bolton & Scharfstein several other publications openly questioned the 4

5 assumptions underlying the MM Proposition or presented completely new capital structure theories. The three most important will be described in the remaining part of this section. Trade-Off Theory of Debt The Trade-Off Theory of Debt is a direct result of Modigliani & Miller s Irrelevance Theorem and was initially developed by Kraus and Litzenberger in The annexation of corporate income tax in the Irrelevance Theorem (Modigliani & Miller, 1963) led to the relative extreme prediction that firms should completely rely on debt financing. This due to the tax-deductibility of the costs of debt and therefore the financial benefits accompanied with a corporate tax shield. However, Graham (2000) showed that a lot of firms finance their operations more conservative. The reason for this is that, besides tax advantages, debt financing also entails specific costs. These costs of debt include for example (indirect) bankruptcy costs, costs of financial distress and agency costs (Jensen & Meckling, 1976). The Trade-Off Theory of Debt describes that firms should choose a capital structure that optimizes the balance between the advantages of a corporate tax shield and the costs of debt. Or in other words, the marginal tax shield benefits should equal the marginal costs of debt. The Trade-Off Theory of Debt therefore believes that companies pursue an Optimal Leverage Ratio of which, according to Myers (1984), deviations are smoothed over time. Myers therefore makes the distinction between the Static Trade-off Theory (single period trade-off benefits and costs of debt) and the Dynamic Trade-off Theory ( Target Adjustment Behavior ). Pecking Order Theory The Pecking Order Theory was originally initiated by Donaldson (1961) and further developed by Myers (1984) and Myers and Majluf (1984). The Pecking Order Theory prescribes that managers apply a hierarchical preference with respect to capital structure decisions. This pecking order finds mainly its basis in adverse selection problems (Myers & Majluf, 1984), agency problems (Myers, 2001), tax consideration (Stiglitz, 1973) and information asymmetry. 5

6 In general firms can raise money by either retaining profits, issuing debt or by emitting new shares. Assuming that the management of a company serves the interest of the current shareholders, new shares will never be issued below their real fundamental value. This results in the consequence that potential investors will perceive the new shares as overvalued, especially when other financing alternatives (retained earnings or debt financing) are available. Ultimately this will decrease the price of these shares and impede the equity offering. This is also empirically substantiated by several studies like Asquith & Mullins (1986), Masulis & Korwar (1986) and Barclay & Litzenberger (1988). Debt financing eliminates a large part of this information asymmetry problem. Debt holders have a priority claim on the firm s assets and will therefore be less harmed by uncertainties about the firm s valuation (Myers, 2001). Unless the costs of debt financing are too high (i.e. due to financial distress) firms will therefore always prefer debt financing to equity financing. Based on above-mentioned intuition, the Pecking Order Theory emerged. Concretely the Pecking Order Theory describes that managers prefer the use of internal funds before the external capital market will be approached (Kayhan & Titman, 2007). However, if internal funds are insufficient, firms will first focus on debt financing before emitting new equity. Furthermore, in the case of debt financing, riskless debt will be preferred over risky debt. So based on the capital needs and financing capabilities, the firm will gradually switch from the safest financing options to the more risky ones. The leverage ratio is therefore a reflection of the firm s capital needs. Consequently, in contrast to the Trade-Off Theory, the Pecking Order Theory does not speak of an optimal leverage ratio. Market Timing Theory Finally, the Market Timing Theory of Debt can be seen as an important explanation of capital structure decisions. Baker and Wurgler (2002) developed the Market Timing Theory, inspired by different surveys and publications such as Graham and Harvey (2001) and Hovakimian, Opler and Titman (2001). In general, the Market Timing Theory states that managers are indifferent between debt and equity financing. The choice for issuing debt or equity is therefore determined by which market offers the most favorable conditions at the moment the firm needs to raise money. In a period of 6

7 high market-to-book ratios, managers will therefore choose to issue equity in order to benefit from the overvalued shares. The opposite is the case for periods with low market-to-book ratios and/or relative low interest rates. According to Baker and Wurgler the capital structure of a firm can therefore be seen as an accumulation of historical market timing decisions. Despite the fact that several papers (Baker & Wurgler, 2002; Welch, 2004 and Kayhan & Titman, 2007) find a persistent markettiming trend, others argue that this effect is short-term and fades away over longer time periods (Leary & Roberts, 2005; Alti, 2006; and Hovakimian, 2006). But who is / are right? As presented in this section, a lot of different theories try to explain capital structure decisions. Theories became more sophisticated over time and tried to tackle the problem out of different perspectives. Unfortunately the theories often point into conflicting directions and until now an ambiguous answer to the question who is or who are actually right cannot be provided. The Pecking Order Theory seems to be the best alternative for large, mature firms that have good access to the external debt market. The more stable character of these firms enables them to finance themselves for the largest part with retained earnings and potentially raise debt when necessary. In contrast, the smaller, younger and fast growing firms with less financial stability initially focus more on issuing equity. In this case, the Trade-Off Theory seems to be the more appropriate alternative. These factors are empirically substantiated by several studies Determinants of Leverage Due to the conflicting characters of the capital structure theories, a new stream of publications showed up that introduced a completely different approach in explaining capital structure decisions. Instead of looking for a general and all-explaining theory they tried to link leverage ratios to several firm-specific characteristics, the so-called Determinants of Leverage. This approach was followed by several studies like Titman & Wessels (1988), Rajan & Zingales (1995), Frank & Goyal (2003) and 1 For an overview of these studies and their results see: Brealey, Myers and Allen, Corporate Finance, McGraw-Hill international edition, 8e editie, 2006, p

8 Lemmon, Roberts & Zender (2008). The most important and commonly accepted determinants of leverage are firm size, growth opportunities, tangibility, profitability and initial leverage. The remaining part of this section will explain each of the abovementioned determinants of leverage and link these to the Pecking Order Theory and Trade-Off Theory of Debt. With respect to the main investigation objective, also the hypotheses underlying this research will be presented. Firm Size Several studies have analyzed the relationship between firm size and leverage. This size effect can intuitively and empirically be explained by as well the Trade-Off Theory as Pecking Order Theory. With respect to the Trade-Off Theory can be stated that the size of a firm is negatively related to the risk of bankruptcy. This due to the more diversified character and more stable earnings of larger firms (Fama & French, 2002). Larger firms are therefore less likely to be exposed to the costs of debt and will therefore relatively take on more leverage (Rajan & Zingales, 1995). In contrast the Pecking Order Theory states that information asymmetry problems, and thus costs of debt, are reduced as firm size increases. This makes debts more attractive towards other sources of funding and will therefore more likely be used in larger firms compared to smaller firms (Degryse, de Goey & Kappert, 2009). Therefore, both the Trade-Off Theory and Pecking Order Theory predict a positive relationship between size and leverage. H1 Leverage and size are positively related Translating the interpretation of the Trade-Off Theory and Pecking Order Theory to a crisis period, an increase in magnitude of the size effect can be expected. Since firm size proxies for the likelihood of bankruptcy, degree of diversification and stability of cash flows it can actually be interpreted as a risk factor. In times of crisis, characterized by decreasing trust and increasing uncertainty, the above-mentioned factors become even more valuable. For that reason I expect the magnitude of the firm size effect to increase in times of crisis. H2 The magnitude of the size effect increases in times of crisis 8

9 Profitability Whereas the current literature provides an unambiguous view about the relationship between size and leverage, this is not the case for profitability. On the one hand, the Cash Flow Theory (Jensen, 1986) states that leverage should be positively related to a firm s profitability. This since debt limits a firm s free cash flows and therefore withholds managers to invest in negative NPV projects. However, the Pecking Order Theory prescribes the opposite scenario. Highly profitable firms have the ability to finance their activities for a larger part with retained earnings. Since the Pecking Order Theory states that firms prefer internal financing to external financing, highly profitable firms will therefore rely less on debt financing. The negative relationship between leverage and profitability is empirically substantiated by several studies such as Toy et al (1974), Kester (1986), Titman & Wessels (1988), Rajan & Zingales (1995) and Fama & French (2002). H3 Leverage and profitability are negatively related From the perspective of external investors profitability is an important measure to evaluate a firm s performance. Indirectly it can therefore also be seen as a proxy for the likelihood that a firm meets its financial obligations. In a certain extent higher profitability gives creditors more certainty. Especially in times of economic turndowns this certainty is highly desired and does it facilitate debt financing. For that reason I expect that the magnitude of profitability will increase in times of crisis. H4 The magnitude of profitability increases in times of crisis Tangibility In as well the Pecking Order Theory as the Trade-Off Theory, a positive relationship between asset tangibility and leverage is prescribed. Asset tangibility is a direct measure for how much collateral is available within a firm. Since collateral reduces the information asymmetry problem (Pecking Order Theory) and agency problems concerning the debt holders (Trade-Off Theory) a positive relationship between tangibility and leverage is in line with both theories. This relationship is also empirically substantiated by several studies such as Bradley et al (1984), Titman & Wessels (1988) and Rajan & Zingales (1995). 9

10 H5 Leverage and tangibility are positively related In times of crisis it can be expected that external lenders demand more securities due to the decreasing trust, uncertainty about fundamental values and therefore increasing agency problems. For that reason investors will put more emphasis on relative safe securities, such as collateral, in their financing decisions. Therefore I expect the magnitude of this determinant to increase in times of crisis. H6 The magnitude of tangibility increases in times of crisis Growth Opportunities Comparable to the relationship between profitability and leverage, the current literature does not provide a concise statement towards the link between growth opportunities and leverage. In the perspective of the Pecking Order Theory, a positive relationship can be expected. This because firms with enormous growth opportunities have more potential to take on additional leverage and, as the theory describes, prefer debt to equity financing. Future investments will therefore rather be financed with debt. However, empirical evidence points out a contrary relation, which origin can be found in Myers (1977) Agency Theory. Intuitively this negative relationship is caused by the fact that highly levered firms might have to refuse profitable and positive NPV projects to prevent themselves from indirect bankruptcy costs. Companies that experience large growth opportunities will therefore focus relatively less on debt financing to sustain financial flexibility. This negative relationship is empirically substantiated and fine-tuned by several studies such as Titman & Wessels (1988), Chung (1993), Barclay et al (1995) and Rajan & Zingales (1995). H7 Leverage and growth opportunities are negatively related With respect to the impact of growth opportunities during crisis periods I refer to Titman & Wessels (1988). Titman & Wessels found a negative relationship between R&D expenses, as a proxy for growth opportunities, and leverage. They state that growth opportunities are actually capital assets that increase firm value but can poorly 10

11 be collateralized. In accordance with hypothesis 6 I therefore expect that the magnitude of this relationship will increase during crisis periods. H8 The magnitude of growth opportunities increases in times of crisis Initial leverage The last classical determinant of leverage is a firm s initial leverage ratio. In their paper Back to the Beginning: Persistence and the Cross-Section of Corporate Capital Structure, Lemmon, Roberts & Zender (2008) present the relative importance of a firm s initial leverage ratio on future capital structure decisions. LRZ show that leverage ratios are persistent over long time horizons. Concretely this implies that highly levered firms stay highly levered for more then 20 years. The same principle counts for firms that have low leverage ratios. This indicates that a firm s long-term target leverage ratio is closely related to their original capital structure. LRZ find a strong and economical significant positive relationship between initial and future leverage. A change of one standard deviation in initial leverage increases book (market) leverage on average with 7% (10%). Furthermore they conclude that the impact of initial leverage is much bigger compared to the other traditional determinants of leverage as previously presented in this section. H9 Leverage and initial leverage are positively related The role of initial leverage during crisis periods is not investigated and identified yet. Besides the fact that leverage ratios show a stable development, LRZ concluded that capital structure decisions mainly depend on time-invariant factors. However, intuitively the opposite scenario might sound reasonable. The exogenous shock of a crisis might force firms to step away from their original financing policy and adapt their capital structure decisions to the requirements of the changing external environment. In this scenario I would expect that the magnitude of initial leverage decreases in times of crisis. H10 The magnitude of initial leverage decreases in times of crisis 11

12 1.3 Hypotheses Overview Hypothesis Variable Expectation H1 Size Positive H2 Size (Crisis) Increases H3 Profitability Negative H4 Profitability (Crisis) Increases H5 Tangibility Positive H6 Tangibility (Crisis) Increases H7 Growth Opportunities Negative H8 Growth Opportunities (Crisis) Increases H9 Initial Leverage Positive H10 Initial Leverage (Crisis) Decreases Table 1: Overview of hypotheses 12

13 ΙΙ. DATA AND SAMPLE SELECTION The initial dataset is downloaded from Standard & Poor s Quarterly Fundamentals Compustat Database. The primary sample includes all firm-quarter observations between 1999Q3 and 2012Q4 of all listed enterprises in the United States. At this primary stage of the sample selection the raw dataset consisted out of 600,084 firmspecific observations relating 21,361 US-listed enterprises. Out of statistical considerations, the initial sample is cleaned according to various prerequisites. At first are companies with total book assets less then $10 million eliminated from the sample to stay in line with previous research (e.g. Lemmon, Roberts & Zender, 2008). Furthermore should all firm-quarters have non-missing data for book leverage, market leverage and all the other variables that will be used in the regression analyses. Firm-quarter observations that seemed illogical and/or unrealistic (i.e. negative sales and negative leverage) are cleared from the dataset. Moreover are all the variables winsorized at the upper and lower 0,5-percentiles to avoid the negative effect of outliers and should the dependent variables book and market leverage fall within the Closed Unit Interval. Last but not least will the dataset only include firms that present 7 year (28 quarters) of consecutive and non-missing data, the so-called survivors. This in order to make the dataset better controllable and to exclude the effect of firms entering and exiting the sample over the timespan of the investigation. At first is the threshold of 28 quarters in line with previous studies (e.g. Lemmon, Roberts & Zender, 2008), which also apply a benchmark of approximately 50% of the total timeframe to determine the survivor sample. Furthermore prevents the threshold that an individual firm represents two disconnected individual series of data in the survivor dataset. However, a disadvantage of this approach is that it makes the dataset vulnerable to a survivorship bias. Since several studies (e.g. Lemmon, Roberts & Zender, 2008) have pointed out that the influence of this potential bias is minimal and that results for the complete and survivorship sample are nearly similar, this research will solemnly focus on the survivor sample. 13

14 Table 1 presents the summary statistics for as well the complete as the survivor sample. All the variables are created in accordance with the research of Lemmon, Roberts and Zender (2008). The actual definitions and methods of calculation for the variables are presented in appendix 1. After cleaning the dataset, the survivor sample consists out of 85,089 firm-specific observations relating 2,205 individual companies. In line with initial expectations the survivors are on average larger, slightly less levered, more profitable and tangible firms that experience larger turnovers, more stable cash flows and have less growth opportunities compared to the complete sample. Furthermore are the survivors on average more likely dividend-paying firms and are the initial leverage ratios of the survivor sample, on average, quite surprisingly a bit higher then these of the complete sample. 14

15 Table 1 Summary Statistics Table 1 presents the summary statistic for both the complete as survivor sample. The dataset consists out off all firm-quarter observations between 1999Q3 and 2012Q4 and is downloaded from Standard & Poor s Quarterly Fundamentals Compustat Database. For all the variables the mean, median and standard deviation with respect to both samples are presented. The definitions and methods of calculation for the variables are presented in appendix 1. Mean Variable [Median] Book Leverage 0.22 [0.18] Market Leverage 0.23 [0.15] Log(Sales) 4.27 [4.13] Market-to-Book 1.49 [1.08] Profitability 0.02 [0.03] Tangibility 0.29 [0.19] Cash flow vol [0.52] Median Industry Book Leverage 0.17 [0.17] Median Industry Market Leverage 0.16 [0.13] Dividend Payer 0.45 [0.0] Intangible Assets 0.15 [0.06] Book Assets 3,868 [313] Initial Book Leverage 0.22 [0.19] Initial Market Leverage 0.23 [0.15] All Firms (SD) Mean [Median] (0.21) 0.21 [0.18] (0.25) 0.22 [0.16] (2.09) 4.67 [4.58] (1.31) 1.47 [1.09] (0.05) 0.03 [0.03] (0.26) 0.31 [0.23] (4.58) 0.50 [0.53] (0.11) 0.17 [0.18] (0.12) 0.16 [0.14] (0.50) 0.56 [1.0] (0.19) 0.15 [0.07] (14,892) 4,084 [441] (0.21) 0.23 [0.21] (0.25) 0.24 [0.17] Obs. 194,236 85,089 Survivors (SD) (0.19) (0.22) (1.97) (1.21) (0.04) (0.26) (3.89) (0.11) (0.13) (0.50) (0.19) (13,603) (0.20) (0.24) 15

16 ΙΙΙ. SORTING In order to get a first impression about the evolution of leverage and its determinants during crisis periods I conduct a simple sorting exercise. Table 2 presents the mean values of book and market leverage with respect to the crisis period and the normal market situation. The crisis period is initiated by the bankruptcy of Lehman in September 2008 and therefore covers the time window 2008Q3 2012Q4. The benchmark period, or the normal market situation, respectively relates the timeframe 1999Q3 2008Q2. Table 2 Evolution of Leverage Table 2 presents the average values for book and market leverage during the normal market situation and the crisis period. The benchmark and crisis period respectively cover the time windows 1999Q3 2008Q2 and 2008Q3 2012Q4. Besides the mean values the table present the results of a Two-Group Mean-Comparison T-Test. Columns 4 and 5 present the 95% Confidence Interval of the difference in mean values between two periods. Column 6 contains the T-Statistic associated with the test that the mean value during the crisis period statistically differs from the mean value during the normal market situation (H0: x!"#$%& x!"#$#$ = 0). The definitions and methods of calculation for the variables are presented in appendix 1. Survivor Sample Variable Normal Crisis 95% CI of Difference T-Statistic Book Leverage Market Leverage The results out of table 2 show an interesting fact. At first glance, the mean values of book and market leverage both seem to increase during the crisis period. However, for book leverage the difference between the benchmark period and the crisis period cannot statistically be distinguished from zero. With respect to market leverage this is actually the case. The difference between the two periods is statistically significant and economically large. Compared to the normal market situation, the average market leverage ratio increases with approximately 17% in times of crisis. Still an additional concern is that the increase in market leverage is not caused by an increase in leverage but by a decrease of market equity values. However, the result of a Two-Group Mean- Comparison T-Test relating the variable Market Equity points out that this is not the case. 16

17 The same methodology is repeated for the determinants of leverage as discussed in section 1. I divided the determinants of leverage into quartiles and sorted them from the lowest to highest. Table 3 presents the mean values for book and market leverage for each individual quartile per determinant of leverage. Table 3 Sorting Exercise This table presents the results of a sorting exercise based on the survivor sample. I sorted each potential determinant of leverage in quartiles and created a dummy variable that equals the value zero (one) if the firm-specific observation falls in the normal market situation (crisis period). The table reports the mean value for as well book as market leverage over the different quartiles during the normal market situation and crisis period. Furthermore are the differences between the mean values reported and tested on statistical significance by a Two-Group Mean-Comparison T-Test (H0: x!"#$%& x!"#$#$ = 0). *, **, *** indicates statistical significance at respectively the 10%, 5% and 1% level. Book Leverage Market Leverage Quartile Normal Crisis Diff. Quartile Normal Crisis Diff. Initial Leverage Initial Leverage *** *** *** *** *** *** *** Log(Sales) Log(Sales) *** *** *** *** ** *** *** Profitability Profitability *** *** *** *** *** *** Market-to-Book Market-to-Book *** *** *** *** * ** ** Tangibility Tangibility *** *** *** *** *** 17

18 The results out of table 3 indicate a positive relationship between as well leverage and size as leverage and tangibility. This is in line with the predictions of the Trade-Off Theory of Debt and Pecking Order Theory. Concerning the growth opportunities (market-to-book), only a clear pattern can be distinguished with respect to market leverage. In accordance with Myers (1977) Agency Theory, a clear negative relationship becomes visible. Furthermore show the quartiles of initial leverage an increasing development in mean values for as well book leverage as market leverage. This is consistent with the findings of Lemmon, Roberts & Zender (2008). With respect to the question whether of the impact of the determinants of leverage changes during the crisis period, it is difficult to draw up an initial conclusion based on the results out of table 3. Analyzing the differences in mean leverage ratios over the crisis and normal market situation does not reveal any clear and consistent patterns. ΙV. EVENT-TIME STUDY In order to further analyze the evolution of leverage during the crisis period, I conduct an Event-Time Study in accordance with the methodology applied by Lemmon, Roberts & Zender (2008). The Event-Time Study analyzes the development of crosssectional leverage ratios over the period 1999Q3 2012Q4. I refer to this period as the Event Time. For each individual quarter in the Event Time the firms are divided into four different quartiles based on their leverage ratio. These quartiles represent respectively the Low, Medium, High and Very High levered firms. In the subsequent quarters these portfolios are kept constant and are all quarter-averages for the remaining periods in the Event Time stored. This process is repeated for all individual quarters in the Event Time, where the quarter in which the portfolio is created is denoted as event-quarter zero. Ultimately this results in 54 strings of quarter averages, one for each quarter in the Event Time. At last, the average of these quarter averages is calculated for each portfolio. This leads to a set of portfolio quarter averages that represent the portfolios evolution of leverage over the Event Time. The analysis is executed for as well book as market leverage. Important is to stress some specific remarks to the above-described approach. At first has to be mentioned that, mainly with respect to the first 9 quarters, the dataset consists out of a limited number of individual firms. This leads to very small and 18

19 potentially biased portfolios. Since this could have a substantial impact on the findings in the latter part of the Event Time, I decided to exclude these quarters from the analysis. Secondly is the effect of firms exiting the sample limited due to the fact that the dataset only includes surviving firms. Last but not least cannot all portfolios be followed for an equal amount of quarters because of the restricted time horizon of this research. Since previous work (Lemmon, Roberts & Zender, 2008) has tested and refuted the latter two remarks, it can be assumed that this does not influence the results in a significant way. Figure 1 presents the results of the Event-Time Study. The findings are generally consistent with the initial findings of Lemmon, Roberts & Zender (2008). At first shows figure 1 that there is a large dispersion between the leverage ratios at the initial portfolio creation date. The difference between the highest and lowest portfolio is approximately 47% (53%) for book (market) leverage at time zero. Secondly, the evolution of leverage clearly shows a converging pattern. With respect to book leverage the most levered portfolio decreases from 47% to 35% and the least levered portfolio increases from 0.26% to 11%. For market leverage, the magnitude of this effect is even bigger. Here the Very High portfolio decreases from 53% to 38% and the Low portfolio increases from 0.18% to 10%. Furthermore it is interesting to see that most of the converging effect takes place in the first quarters after the portfolio creation date. The differential between the portfolios for book (market) leverage are respectively 8.7% (9.7%), 10.5% (11.8%) and 13.4% (16.6%) for the Low, Medium, High and Very High portfolios. Compared to the average within-firm standard deviation of book (market) leverage of 7.45% (9.6%) these differentials are economically large. Furthermore, over the entire Event Time, none of the portfolios intersect. This implies that highly levered firms remain highly levered in the subsequent quarters. The same principle counts for the Low, Medium and High levered companies. This substantiates the persistence of leverage ratios as indicated by Lemmon, Roberts & Zender (2008). 19

20 Figure 1 Event-Time Study This figure presents the results of the Event-Time Study, which follows the same methodology as applied by Lemmon, Roberts & Zender (2008). For each individual quarter, the firms are allocated to four different quartiles based on their leverage ratio. These quartiles respectively represent firms that have Low, Medium, High and Very High leverage ratios. Subsequently these portfolios are kept constant and followed for the remaining quarters of the Event Time. Each quarter I stored the average leverage ratio per portfolio. This results in 54 strings of quarter averages, where the quarter in which the portfolio is created is denoted as event-quarter zero. After that I calculated the average of these portfolio quarter averages for each individual event-quarter. This leads to a set of portfolio quarteraverages that represent the portfolios evolution of leverage over the Event Time. Panel A and B respectively present the results for book and market leverage. Panel A: Book Leverage Panel B: Market Leverage 20

21 Despite the fact that a clear converging trend becomes visible in figure 1, some remarkable fluctuations in the quarters prior to and during the crisis period show up. To analyze these fluctuations in more detail, I compare the results of the Event-Time Study with the theoretical predictions of Lemmon, Roberts & Zender (2008). Suppose we would live in a theoretical world where LRZ findings concerning persistent and converging leverage ratios would perfectly hold. We then would expect to find the following results: 1. The High and Very High levered portfolios have a negative, but gradually increasing growth rate over the entire Event Time window. Furthermore should the average growth rate of the High portfolio fall below the average growth rate of the Very High portfolio. This would lead to a convex evolution of leverage. 2. The Low and Medium levered portfolios would have a positive, but gradually decreasing growth rate over the entire Event Time window. Furthermore should the average growth rate of the Low portfolio fall below the average growth rate of the Medium portfolio. Contrary to the highly levered portfolios, this would lead to a concave evolution of leverage. In table 3 I present the average quarterly growth rates of the different portfolios during the normal market situation and crisis period, together with the results of some statistical tests. Out of these results can be concluded that the theory of LRZ only partly holds. Looking at the Low and Medium portfolios, the growth rates during as well the normal market situation as the crisis period are statistically larger then zero (second and fourth column of panel C and D). This is in line with initial expectations. However, the result of the Two-Group Mean-Comparison T-Test shows that the difference between the two average growth rates is not statistically significant distinguishable from zero. For that reason can neither be concluded that the growth rate of the Low and Medium portfolios gradually decreases over the Event-Time, nor that the evolution of leverage with respect to these portfolios changes during the crisis period. With respect to the High portfolio the results are more striking. Initially we would expect a negative growth rate during both periods. However, the results out of panel C 21

22 and D point out a contrary situation. Only for book leverage the growth rate during the normal market situation is negative. For market leverage and the crisis period the growth rates are statistically significant larger then zero. Furthermore is the average growth rate during the crisis period statistically bigger then the average growth rate during the normal market situation for as well book leverage as market leverage (Panel C and D column 5). These findings suggest that the High portfolios increase leverage in times of crisis. The same principle counts for the Very High portfolios. At first fall the average growth rates during the normal market situation statistically significant below zero. This is also in line with the initial expectations. Furthermore are, comparable to the High portfolio, the average growth rates during the crisis period statistically larger then the average growth rates during the normal market situation. This could substantiate the expected convex evolution of the most highly levered portfolios. However, since the average growth rate (book leverage) becomes positive during the crisis period there can be concluded that also the Very High portfolio increases leverage during the crisis period. This conclusion cannot be drawn up with respect to market leverage due to insignificant results in column 3 and 4 of panel D. 22

23 Table 3 Growth Rates Different Market Conditions This table presents the average quarterly growth rates for the different portfolios during the two market conditions, combined with the outcomes of several statistical tests. Respectively panel A and B present the growth rates for book and market leverage together with the 95%-Confidence Interval and T- Statistic of the Two-Group Mean-Comparison Test. Panel C and D respectively present the p-values for the statistical tests as described in the header of the column. Hereby x n represents the average growth rate during the normal market situation and x c the average growth rate during the crisis period. Prior to the statistical testing and calculation of the mean values the growth rates are winsorized at the upper and lower 2.5-percentiles to mitigate the effect of unwanted outliers. *, **, *** indicates statistical significance at respectively the 10%, 5% and 1% level. Panel A: Book Leverage Portfolios Portfolio Normal Crisis Diff. 95% CI of Difference T-Statistic Low 8.63% 5.45% 3.19% Medium 1.06% 0.88% 0.18% High -0.14% 0.84% -0.98%*** Very High -0.92% 0.24% -1.16%*** Panel B: Market Leverage Portfolios Portfolio Normal Crisis Diff. 95% CI of Difference T-Statistic Low 9.11% 6.29% 2.82% Medium 1.78% 1.32% 0.46% High 0.21% 1.24% -1.03%*** Very High -0.91% -0.18% -0.73%*** Panel C: Book Leverage Statistical Tests Portfolio x n < 0 x n > 0 x c < 0 x c > 0 x n < x c x n > x c Low *** ** Medium *** ** High ** *** *** Very High *** * *** Panel D: Market Leverage Statistical Tests Portfolio x n < 0 x n > 0 x c < 0 x c > 0 x n < x c x n > x c Low *** ** Medium *** ** High * *** *** Very High *** ***

24 V. MULTIVARIATE REGRESSION ANALYSIS Whereas the previous section focused on the evolution of leverage, this section analyzes the determinants of capital structure decisions during crisis periods by the use of a Multivariate Regression Analysis. Within the regression analysis the crosssectional and time-varying book and market leverage ratios will be explained by a subset of potential determinants of leverage. Despite the fact that this report mainly focuses on the traditional determinants of leverage as presented in section 1, I also incorporate some other determinants of leverage based on Lemmon, Roberts & Zender (2008). The regression analysis includes two interaction terms for each determinant of leverage to reveal a potential deviating impact of these factors during the crisis period and normal market situation. This methodology leads to the regression equation as presented below. Leverage!" = α + β! Crisis + β! Normal X!"!! + β! Crisis X!"!! + β! Normal Leverage!" + β! Crisis Leverage!" + υ! + γ + ε!" i t Crisis Normal(X!"!! ) Crisis(X!"!! ) Normal(Leverage!" ) Crisis Leverage!" υ! γ ε!" Indexes the individual firms. Indexes the time (quarters). Dummy variable that equals 1 if the observations falls within the crisis period (2008Q3 2012Q4) and zero otherwise. Potential determinants of leverage lagged for one quarter and interacted with a dummy variable that equals 1 if the observation falls within the normal market situation (1999Q3 2008Q2) and zero otherwise. Potential determinants of leverage lagged for one quarter and interacted with a dummy variable that equals 1 if the observation falls within the crisis period (2008Q3 2012Q4) and zero otherwise. Firm s initial leverage ratio interacted with a dummy variable that equals 1 if the observation falls within the normal period and zero otherwise. Firm s initial leverage ratio interacted with a dummy variable that equals 1 if the observation falls within the crisis period and zero otherwise. Quarter Fixed Effects. Industry Fixed Effects, based on the two digits SIC-code. Random error term. At first are the independent variables out of the regression analysis tested for multicollinearity. The correlation matrixes relating as well the book as market leverage model are presented in appendix 2. As shown in figure 3 and 4 the correlations between the variables are not disturbingly positive or negative. For that reason, no variables will be excluded from the analysis. 24

25 In table 4 I present the results of the Multivariate Regression Analysis together with the outcomes of some statistical tests. For a detailed explanation of the contents of the table I refer to the caption and appendix 1. Prior to the actual interpretation of the results has to be mentioned that, except for the variable Crisis, the signs and statistical significance of the independent variables in the standard OLS regression are identical to these of the Tobit-regression. This implies that the capped character of the dependent variables does not seem to influence the results out of the standard OLS regression into large extents. In the remaining part of this section I will present the most important findings of the Multivariate Regression Analysis per individual variable. Crisis The results of the standard OLS regressions of table 4 present a negative relationship between the dummy variable Crisis and leverage. According to the results in columns 1 and 2 (4 and 5), book (market) leverage decreases with approximately 0.05 (0.04) during the crisis period. Compared to the mean values of book and market leverage this implies change of respectively 24% and 26%. These results are therefore also economically significant. An intuitive interpretation of this finding would be that the crisis on itself has a decreasing influence on leverage ratios. This does not necessarily imply that leverage ratios decline in times of crisis. I will further elaborate on this topic in the next part of this section. At last should be stated that the abovementioned intuition cannot be adopted without concerns. In the Tobit-regressions the results for book leverage become insignificant and does the sign of the betacoefficient change in the market leverage model. For this reason no unambiguous conclusion can be drawn up with respect to the stand-alone effect of the crisis on leverage ratios. Evolution of Leverage In the previous part of this section I stated that a negative coefficient on the dummy variable Crisis does not necessarily imply that leverage ratios decrease in times of crisis. This because the variable Crisis only captures the stand-alone effect of the crisis period on leverage ratios. The overall effect of the crisis might be different because of the aggregate impact of all the other independent variables out of the model. In panel C I present the average predicted leverage ratio during the normal 25

26 market situation and crisis period based on the regression results. Furthermore I calculated the percentage of firms that have a higher predicted average leverage ratio during the crisis period compared to the normal market situation. Last but not least I present the p-value of a Two-Group Mean-Comparison T-Test that analyses whether of the average predicted leverage ratio during the crisis period is statistically significant larger compared to the normal market situation. All three of the tests show an increase in predicted leverage during the crisis period. This implies that, despite the stand-alone effect of the crisis is negative, leverage increases in times of crisis. This result is in line with the findings for the High and Very High levered portfolios out of the Event-Time Study and the initial sorting exercise out of section 3. Initial Leverage The results out of table 4 present a positive relationship between initial leverage and book and market leverage. These results are also statistically significant and economically large. A one-standard deviation change in initial leverage increases book (market) leverage with approximately 38% (41%) and 62% (55%) in respectively the crisis and normal market situation. In line with the findings of Lemmon, Roberts & Zender (2008), the magnitude of this effect is much bigger compared to the other independent variables. Furthermore presents panel B that the interaction terms are statistically different from each other in all the regression analyses. As shown in panel A, the magnitude of the effect clearly decreases in times of crisis. Based on the above-mentioned results, hypotheses 9 and 10 can objectively be accepted. Size Comparable to the findings of initial leverage, the results of the Multivariate Regression Analysis indicate a positive relationship between size and leverage. This is in line with the predictions of the Trade-Off Theory of Debt. Obviously the parameter estimates of the variable Size are much smaller compared to these of initial leverage. However, the results are still economically significant. For as well the normal market situation as crisis period, a one-standard deviation change in size leads to an increase of approximately 10% (9%) in book (market) leverage. In the Tobitregressions this impact becomes even larger. Despite the fact that both interaction terms do not seem to differ from each other in the regression analyses, the results of 26

27 panel B show the opposite scenario. With respect to the book leverage model the difference between the two interaction terms can statistically be distinguished from zero on the 1% level. For market leverage this is not the case. However, since this difference has to be found three places to the right of the decimal point, the impact is economically small and negligible. Due to the fluctuating results for book and market leverage and the relative small economic difference between the normal market situation and crisis period, I do not accept hypothesis 2. Since all regression analyses indicate a positive relationship between size and leverage, I accept hypothesis 1. Growth Opportunities In line with Myers (1977) Agency Theory, the results of table 4 indicate a statistical significant and negative relationship between growth opportunities and leverage. Especially with respect to the market leverage model, the results are also economically large. Hypothesis 7 can therefore objectively be accepted. Concerning the difference between both interaction terms, again a distinction has to be made with respect to the book and market leverage model. The results of panel B indicate a statistical significant difference in the market leverage model, but not in the book leverage model. Focusing at the results for market leverage, this differential is also economically large. Clearly the impact of a one-standard deviation change during the crisis period is 0.03 bigger compared to the normal market situation. When looking at the mean value of market leverage, this differential represents a 14% further decrease in leverage during the crisis period. Hypothesis 8 can therefore objectively be accepted for the market leverage model. This is not the case for book leverage because of the statistical insignificant difference between both interaction terms. Profitability According to the Pecking Order Theory profitability should be negatively related with leverage. This is also motivated by the results out of table 4. The beta-estimates fall in all the regression analyses statistically significant below zero. Furthermore are the findings economically large, what implies that hypothesis 3 can be accepted. With respect to the question whether of the impact of this variable changes in times of crisis, no statistically substantiated conclusion can be drawn up. As shown in panel B no significant difference exists between both interaction terms within as well the book as market leverage model. For this reason should hypothesis 4 be rejected. 27

28 Tangibility Comparable to the findings of the growth opportunities and leverage, the variable Tangibility shows different results in the book and market leverage model. At first has to be mentioned that a positive relationship between tangibility and leverage is found in all regression analyses. These results are statistically significant, economically large and in line with as well the Pecking Order Theory and Trade-Off Theory of Debt. Hypothesis 5 can thus be accepted. Furthermore it is interesting to see is that only for market leverage a statistical significant difference between both interaction terms can be distinguished (panel B, columns 4-6). Combined with the findings out of panel A can therefore be concluded that the impact of this variable, with respect to market leverage, increases in times of crisis and hypothesis 6 should be accepted. Due to the insignificant results in the book leverage model, the same conclusion cannot be drawn up for the relationship between tangibility and book leverage. Other variables Despite the fact that this report mainly focuses on the traditional determinants of leverage, it is also worthwhile to briefly discuss the most important findings concerning the other variables out of the regression analyses. The results out of table 4 indicate a positive relationship with respect to the variable Industry Median Leverage and a negative relationship between the variable Dividend Payer and leverage. These findings are consistent with the predictions of the Trade-Off Theory of Debt and conclusions of previous work (e.g. Frank & Goyal; 2004 and Lemmon, Roberts & Zender; 2008). Concerning the variable Industry Median Leverage the difference between the coefficients on both interaction terms can statistically be distinguished from zero in as well the book as market leverage model. With respect to book (market) leverage the differential between both interaction terms is approximately 0.02 (0.01). Compared to the average value of book (market) leverage this differential represents a percentage of 10% (5%), what is economically large. Clearly this would imply that the impact of this variable increases in times of crisis. Due to the insignificant results out of panel B this cannot be concluded for the variable Dividend Payer. 28

29 Table 4 Results Multivariate Regression Analysis This table presents the results of the Multivariate Regression analysis according to formula 1. Panel A reports the parameter estimates multiplied by the standard deviation of the underlying variable, together with the t-statistic (in brackets). The variable Crisis is a dummy variable that equals the value one if the firm-specific observation falls within the crisis period (2008Q3 2012Q4) and zero otherwise. X*Normal (X*Crisis) are the potential determinants of leverage interacted with a dummy variable that equals one (zero) if the observation falls within the normal market situation and zero (one) otherwise. Columns 1, 2, 4 and 5 contain the results of a standard OLS-regression, columns 3 and 6 present the marginal effects calculated at the variables mean of a Tobit-regression. The Tobit-regression is conducted due to the fact that the dependent variables are capped from above and below. All the variables are winsorized at the upper and lower 0.5-percentiles and the regression analyses are clustered at firm level. Panel B reports for each independent variable the p-value of the test X*Normal = X*Crisis. Panel C presents for each regression analysis the average predicted value of book and market leverage and the percentage of firms for which the average predicted leverage ratio during the crisis period was bigger compared to the normal market situation. Furthermore presents Panel C the p-value of the Two-Group Mean- Comparison Test that investigates whether of the average predicted leverage ratio during the crisis period is larger then the average predicted leverage ratio during the normal market situation. *, **, *** indicates statistical significance at respectively the 10%, 5% and 1% level. Panel A Book Leverage Market Leverage Variable (1) (2) (3) (4) (5) (6) Crisis -0.05*** (-4.63) -0.05*** (-4.34) (-0.75) -0.04** (-2.43) -0.04*** (-2.60) 0.05*** (3.87) Initial Leverage*Crisis 0.08*** (18.10) 0.08*** (18.64) 0.09*** (19.14) 0.09*** (16.31) 0.09*** (17.28) 0.09*** (16.70) Initial Leverage*Normal 0.13*** (35.00) 0.12*** (34.96) 0.14*** (37.14) 0.12*** (30.26) 0.12*** (31.37) 0.13*** (30.98) Log(Sales)*Crisis 0.02*** (6.73) 0.02*** (8.36) 0.04*** (8.31) 0.02*** (5.48) 0.02*** (7.29) 0.04*** (8.14) Log(Sales)*Normal 0.02*** (7.04) 0.02*** (8.41) 0.02*** (9.17) 0.02*** (7.87) 0.02*** (9.72) 0.04*** (10.79) Market-to-Book*Crisis -0.01** (-2.03) -0.01*** (-2.64) -0.02*** (-3.94) -0.05*** (-11.93) -0.05*** (-13.20) -0.08*** (-14.13) Market-to-Book*Normal -0.01*** (-5.85) -0.01*** (-7.18) -0.02*** (-8.66) -0.02*** (-12.44) -0.02*** (-14.26) -0.05*** (-14.86) Profitability*Crisis -0.01*** (-4.04) -0.02*** (-4.86) -0.02*** (-4.23) -0.02*** (-8.39) -0.03*** (-10.00) -0.03*** (-8.74) Profitability*Normal -0.01*** (-6.73) -0.01*** (-7.46) -0.02*** (-6.63) -0.02*** (-12.53) -0.02*** (-13.70) -0.03*** (-12.50) Tangibility*Crisis 0.02*** 0.03*** 0.02*** 0.02*** 0.04*** 0.03*** 29

30 Tangibility*Normal 0.02*** (6.07) Industry Med. Lev.*Crisis 0.04*** (7.50) Industry Med. Lev.*Normal 0.02*** (6.10) Cash Flow Vol.*Crisis 0.00 (1.19) Cash Flow Vol.*Normal 0.00 (1.25) Dividend Payer*Crisis -0.01** (-2.12) Dividend Payer*Normal -0.02*** (-3.36) (4.55) (6.02) (5.18) (5.20) (6.71) (6.32) 0.03*** (7.09) ### (###) -0.02*** (-4.78) 0.00 (1.16) 0.00 (1.22) -0.01** (-2.13) -0.02*** (-3.41) 0.02*** (7.14) 0.04*** (7.55) 0.02*** (5.98) 0.00 (1.50) 0.00 (1.61) -0.02** (-2.49) -0.02*** (-3.88) 0.01*** (3.90) 0.04*** (7.93) 0.03*** (9.32) 0.00* (1.92) 0.00** (2.04) -0.03*** (-3.74) -0.02*** (-3.61) 0.02*** (5.91) ### (###) -0.01** (-2.49) 0.00* (1.86) 0.00** (2.03) -0.03*** (-3.88) -0.02*** (-3.83) 0.02*** (6.11) 0.04*** (7.91) 0.03*** (9.05) 0.00** (1.98) 0.00** (2.12) -0.04*** (-4.46) -0.03*** (-4.44) Quarter Fixed Effects Yes Yes Yes Yes Yes Yes Industry Fixed Effects No Yes No No Yes No Adjusted R Obs. 82,884 82,884 82,884 82,884 82,884 82,884 Panel B Book Leverage Market Leverage (1) (2) (3) (4) (5) (6) Initial Leverage *** *** *** *** *** *** Log(Sales) *** *** *** Market-to-Book *** *** *** Profitability Tangibility *** *** *** Industry Med. Leverage *** *** *** *** *** *** Cash Flow Volatility Dividend Payer * Panel C Book Leverage Market Leverage (1) (2) (3) (4) (5) (6) Predicted Leverage Crisis Predicted Leverage Normal Lev Crisis > Lev Normal 61.6% 60.9% 61.7% 86% 85.5% 82.7% Pred. Lev. Crisis > ** ** *** *** *** *** Pred. Lev. Normal 30

31 VΙ. VARIANCE DECOMPOSITION ANALYSIS In this section will be analyzed how the variability of leverage ratios is composed and can be allocated to the individual determinants of leverage. Hereby is made use of a Variance Decomposition Analysis. The general regression model underlying this analysis is presented below by formula 2. The estimated model is consistent with Lemmon, Roberts and Zender (2008) and similar to formula 1 out of section 5. However, in contrast to formula 1, the Variance Decomposition Analysis neglects initial leverage ratios but incorporates Firm Fixed Effects (η! ). Leverage!" = α + β! Crisis + β! Normal X!!! + β! Crisis X!!! + η! + υ! + γ + ε!" (2) The results of the Variance Decomposition Analysis are presented in table 5. Panel A shows the results for book leverage and Panel B for market leverage. For as well book as market leverage seven different models are estimated, each comprising different determinants of leverage and/or fixed effects. Table 5 presents for each determinant of leverage the fraction of the total Type 3 Partial Sum of Squares within a specific model. Furthermore are the adjusted R-squares of the estimated models presented in the last row of the panels. For a detailed description of the applied methodology I refer to the caption of table 5. The results of the Variance Decomposition Analysis are generally consisted with the findings of LRZ (2008). When looking to columns A and B, immediately a striking fact becomes visible. In these columns are the leverage ratios solemnly explained by Firm Fixed Effects and Quarter Fixed Effects. As shown in column A, the Firm Fixed Effect Model realizes an adjusted R-square of 0.76 (0.72) and thus captures 76% (72%) of the total variance of leverage. In contrast, the Quarter Fixed Effect model realizes an adjusted R-square of 0.01 (0.03). This enormous deterioration indicates that leverage ratios depend for the vast majority on time-invariant factors and that heterogeneous capital structure decisions arise due to cross-sectional firm differences. In column D I incorporate the determinants of leverage as proposed by Rajan & Zingales (1995). Clearly the Industry Fixed Effects and variables Tangibility and Market-to-Book capture the most variation in leverage ratios. However, the model only realizes an adjusted R-square of 0.28 (0.37). This is significantly lower 31

32 compared to the Firm Fixed Effect model out of column A. When I include Firm Fixed Effects (column E) the adjusted R-square almost triples to 0.78 (0.77) and captures by far the largest part of the total variance in leverage. This again points out the relative importance of cross-sectional firm differences in capital structure decisions. The same principle counts for columns F and G, in which I also include other potential determinants of leverage (Frank & Goyal, 2004) in the model. When looking at the models that incorporate Firm Fixed Effects (columns A, C, E and G) it becomes obvious that most of the variation in leverage can be attributed to this firm-effect. The adjusted R-squares are by far the highest and remain relative stable over the different models. This leads to some interesting conclusions. At first indicates the firm-effect that capital structure decisions are driven by unobserved and firm-specific characteristics that are not completely captured by the determinants of leverage that are currently included in the models. Secondly this points out that the vast majority of variability in leverage ratios can be explained by cross-sectional differences in firm characteristics, rather then by time-dependent factors. The results of the Variance Decomposition Analysis would therefore suggest that crisis periods in itself are not a determinant of leverage. But does this mean that the determinants of leverage are actually useless in explaining capital structure design? No, this does not have to be the case. If the determinants of leverage mainly capture cross-sectional variation, instead of time-dependent variation, their explanatory power will naturally drop when Firm Fixed Effects are added to the model. The results of table 5 point out that this is actually the case. Comparing the adjusted R-squares out of columns A and E show that the explanatory value of the model increases with only 2% (5%) when including the determinants of leverage. However, the determinants on itself capture 28% (37%) of the total variation in leverage (column D). Despite the fact that the determinants of leverage do seem to capture a significant part of the cross-sectional variance in leverage ratios, this is much lower compared to the Firm Fixed Effects (book leverage: 28% vs. 76%; market leverage (37% vs. 72%). 32

33 Table 5 Variance Decomposition Analysis This table presents the results of the Variance Decomposition Analysis. Panel A shows the results for book leverage, Panel B for market leverage. For as well book as market leverage seven different models are estimated, each with different determinants of leverage and/or fixed effects. For each different model I estimated a regression analysis incorporating all the mentioned independent variables and fixed effects. I refer to this model as the complete model. For the complete model I stored the adjusted R-squared (presented in the last row of the table) and the Sum of Squares due to Regression (SSR). Subsequently, for each independent variable or fixed effect, I estimated the same regression model but then without the variable of interest. I refer to this model as the incomplete model. After that, the difference between the SSR of the complete model and the incomplete model was calculated, also referred to as the Type 3 Partial Sum of Squares. The Type 3 Partial Sum of Squares therefore shows, leaving all the other variables constant, how much additional variance of the dependent variable is captured by the regression analysis when adding the variable of interest to the model. To normalize the results, I summed the Type 3 Partial Sum of Squares across the independent variables and divided the Type 3 Partial Sum of Squares of each individual effect by this aggregate. This forces each column to count up to one and is in accordance with the methodology applied by Lemmon, Roberts and Zender (2008). Variable Panel A Book Leverage (a) (b) (c) (d) (e) (f) (g) Firm Fixed Effects Quarter Fixed Effects Dummy Crisis Log(Sales)*Crisis Log(Sales)*Normal Market-to-Book*Crisis Market-to-Book*Normal Profitability*Crisis Profitability*Normal Tangibility*Crisis Tangibility*Normal Industry Med. Lev.*Crisis Industry Med. Lev.*Normal Cash Flow Vol.*Crisis Cash Flow Vol.*Normal Dividend Payer*Crisis Dividend Payer*Normal Industry Fixed Effects Adjusted R

34 Variable Panel B Market Leverage (a) (b) (c) (d) (e) (f) (g) Firm Fixed Effects Quarter Fixed Effects Dummy Crisis Log(Sales)*Crisis Log(Sales)*Normal Market-to-Book*Crisis Market-to-Book*Normal Profitability*Crisis Profitability*Normal Tangibility*Crisis Tangibility*Normal Industry Med. Lev.*Crisis Industry Med. Lev.*Normal Cash Flow Vol.*Crisis Cash Flow Vol.*Normal Dividend Payer*Crisis Dividend Payer*Normal Industry Fixed Effects Adjusted R

35 VΙΙ. CONCLUSIONS This master thesis investigates the evolution of leverage and its determinants in times of crisis. The study focuses on a sample that consists out of 85,089 observations relating 2,205 US-listed firms that present 7 years on non-missing and consecutive data. The dataset is downloaded from Standard & Poor s Quarterly Fundamentals Compustat Database and covers the time period 1999Q3 2012Q4. In order to analyze potential differences in capital structure design, the time horizon of the research is divided in a crisis period (2008Q3 2012Q4) and benchmark period (1999Q3 2008Q2). Evolution of Leverage Consistent with the findings of previous research, the evolution of leverage shows a persistent and converging trend. Intuitively this implies that, compared to peer groups, highly levered firms remain highly levered for more then 11 years. Furthermore decreases the differential between the respective leverage groups over time. The magnitude of this converging effect is especially large for the lowest and highest levered firms and occurs mainly in the first few quarters of the Event Time. With respect to the evolution of leverage in times of crisis, the results of a Multivariate Regression Analysis indicate two important findings. At first does a crisis period on itself seem to decrease leverage. This is motivated by the negative beta-estimates of a dummy variable that equals the value zero (one) if the firmspecific observation falls within the benchmark period (crisis period). Yet, this conclusion should be adopted with care due to deviating findings in a Tobitregression. Taking into consideration the aggregate effect of all determinants of leverage incorporated in the regression models, leverage seems to increase in times of crisis. This is empirically substantiated by the finding that for all regression analyses the average predicted leverage ratio in times of crisis is statistically significant larger compared to the average predicted leverage ratio during the benchmark period. Based on these results can be concluded that, despite the decreasing stand-alone effect of a crisis period, leverage ratios seem to increase in times of crisis. A more detailed analysis of individual leverage groups points out that this is especially the case for the High and Very High levered firms 35

36 Determinants of Leverage Also with respect to the determinants of leverage can be stated that most of the findings fall in line with previous research. At first I find evidence that bigger and more tangible firms have on average higher leverage ratios. This is consistent with the predictions of the Trade-Off Theory of Debt. In contrast, highly profitable firms tend to have on average lower leverage ratios. Intuitively this effect can be explained by the Pecking Order Theory. The fourth determinant of leverage that is analyzed are a firm s growth opportunities. Consistent with Myers (1977) Agency Theory a negative relationship between leverage and market-to-book ratios is found. Furthermore I find a positive relationship between as well initial leverage and leverage as the variable Industry Median Leverage and leverage. Last but not least indicate the results of the Multivariate Regression Analysis that dividend paying firms experience, on average, lower leverage ratios. With respect to the question whether of the impact of the determinants of leverage changes in times of crisis some interesting evidence is found. At first does the impact of the variables Tangibility and Growth Opportunities increase during the crisis period. However, these findings are only significant regarding the market leverage model. The results for the variables Initial Leverage and Industry Median Leverage are less ambiguous. The impact of initial leverage decreases economically and statistically significant in as well the book as market leverage model. The same counts for Industry Median Leverage, yet here an increasing impact is found. Despite the fact that the existing determinants of leverage seem to explain capital structure decisions in a fairly good manner, the results of a Variance Decomposition Analysis point out the contrary situation. The relative importance of Firm Fixed Effects indicates that capital structure decisions are driven by unobserved and firmspecific characteristics that are not completely captured by the traditional determinants of leverage. Secondly this points out that the vast majority of variability in leverage ratios can be explained by cross-sectional differences in firm characteristics, rather then by time-dependent factors. With respect to future research it would therefore be interesting to find new determinants of leverage that focus on the unexplained variation in leverage ratios and simultaneously alleviate the problems of heterogeneous intercepts in the regression analyses. 36

37 VΙΙΙ. REFERENCES Alti, A. (2006), How Persistent is the Impact of Market Timing on Capital Structure, Journal of Finance, Vol. 61, No. 4 Asquith, P. and D.W. Mullins (1986), Equity issues and offering dilution, Journal of Financial Economics 15, p Baker, M. and J. Wurgler (2002), Market Timing and Capital Structure, Journal of Finance, Vol. 57, No. 1, pp Barclay, M.J. and R.H. Litzenberger (1988), Announcement effects of new equity issues and the use of intraday price data, Journal of Financial Economics 21, pp Barclay, M.J. and C.W. Smith (1995), The maturity structure of corporate debt, Journal of Finance, Vol. 50, Bolton, P. and D.S. Scharfstein (1990), A theory of predation based on agency problems in financial contracting, American Economic Review, Vol. 80, No. 1, pp Bradley, M., G.A. Jarrell and E.H. Kim (1984), On the Existence of an Optimal Capital Structure: Theory and Evidence, The Journal of Finance, Vol. 39, No. 3, pp Brealey, Myers and Allen (2006), Corporate Finance, McGraw-Hill International Edition, 8th edition, pp. 494 Chung K.H. (1993), Asset Characteristics and Corporate Debt Policy: An Empirical Investigation, Journal of Business Finance and Accounting, Vol. 20, No. 1, pp Degryse, H., P. de Goeij and P. Kappert (2009), The Impact of Firm and Industry Characteristics on Small Firm s Capital Structure: Evidence from Dutch Panel Data, CentER Discussion Paper, No Donaldson, C. (1961), Corporate debt capacity. Harvard University. 37

38 Fama, E. and K. French (2002), Testing trade-off and pecking order predictions about dividends and debt, Review of Financial Studies, Vol. 15, No. 1, pp Frank, M.Z. and V.K. Goyal (2003), Testing the pecking order theory of capital structure, Journal of Financial Economics, Vol. 67, pp Frank, M.Z. and V.K. Goyal (2004), Capital structure decisions: Which factors are reliably important? Financial Management, Vol. 38, pp Graham, J.R. (2000), How big are the tax benefits of debt? Journal of Finance, Vol. 55, pp Graham, J.R. and C.R. Harvey (2001). The Theory and Practice of Corporate Finance: Evidence from the Field, Journal of Financial Economics, Vol. 60, pp Hovakimian, A. (2006), Are Observed Capital Structure Determined by Equity Market Timing? Journal of Financial and Qualitative Analysis, Vol. 41, No. 1 Hovakimian, A., T. Opler and S. Titman (2001), Debt-equity choice, Journal of Financial and Quantitative Analysis, Vol. 36, No. 1, pp Jensen, M.C. (1986), Agency Costs of Free Cash Flow: Corporate Finance and Takeovers, American Economic Review, Vol. 76, No. 2, pp Jensen, M.C. and W.H. Meckling (1976), Theory of the Firm: Managerial Behaviour, Agency Costs and Capital Structure, Journal of Financial Economics, Vol. 3, No. 4 pp Kayhan, A. and S. Titman (2007), Firm s Histories and Their Capital Structure, Journal of Financial Economics, Elsevier, vol. 83(1), pp Kester, C.W. (1986), Capital and Ownership Structure: A Comparison of United States and Japanese manufacturing Corporations, Financial Management 15, 5-16 Kraus, A. and R.H. Litzenberger (1973), A State-Preference Model of Optimal Financial Leverage, Journal of Finance, Vol. 33, pp

39 Leary, M.T. and M.R. Roberts (2005), Do firms rebalance their capital structures? Journal of Finance, Vol. 60, pp Lemmon, M.L., M.R. Roberts and J.F. Zender (2008), Back to the Beginning: Persistence and the Cross-Section of Corporate Capital Structure, The Journal of Finance, Vol. 63, No. 4, pp Masulis, R.W. and A.N. Korwar (1986), Season equity offerings: An empirical investigation, Journal of Financial Economics, Vol. 15, p Modigliani, F. and M. H. Miller (1958), The Cost of Capital, Corporation Finance and the Theory of Investment. American Economic Review. Vol. 48(3), pp Modigliani, F. and M. H. Miller (1963), Corporate Income Taxes and the Cost of Capital: A Correction. American Economic Review, Vol. 53(3), pp Myers, S.C. (1977), Determinants of corporate borrowing, Journal of Financial Economics, Vol. 5, pp Myers, S.C. (1984), The Capital Structure Puzzle, Journal of Finance, Vol. 34(3), pp Myers, S.C. (2001), Capital Structure, Journal of Economic Perspectives, Vol. 15, No. 2, pp Myers, S.C. and N.S. Majluf (1984), Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have, Journal of Financial Economics, Vol. 13, pp Rajan, R.G. and L. Zingales (1995), What Do We Know about Capital Structure? Some Evidence from International Data, The Journal of Finance, Vol. 50, No. 5, pp Stiglitz, J.E. (1973). Taxation, Corporate Financial Policy and the Cost of Capital, Journal of Public Economics, Elsevier, Vol. 2(1), pp Titman, S. and R. Wessels (1988), The Determinants of Capital Structure Choices, The Journal of Finance, Vol. 43, No. 1, pp

40 Toy, N., Stonehill, A., Remmers, L., Wright, R. and Beekhuizen, T. (1974), A Comparative International Study of Growth, Profitability and Risk as Determinants of Corporate Debt Ratios in the Manufacturing Sector, Journal of Financial and Quantitative Analysis, Vol. 9, No. 5, pp Welch, I. (2004), Capital structure and stock returns, Journal of Political Economy, Vol. 112, No. 1, pp

41 - Appendix 1 - The overviews presented below show the definitions and methods of calculation for the key variables that are used in the research. Furthermore are the Compustat Quarterly Fundamentals Descriptions and Item Numbers for the variables and its components elucidated. The variable construction is based on, and for the majority consistent with, Lemmon, Roberts and Zender (2008). Since Lemmon, Roberts and Zender used Standard & Poor s Annual Compustat Database, the variable definitions and item numbers are adjusted to Compustat s Quarterly Fundamentals Database. Variable Total Debt Market Equity Book Leverage Market Leverage Log(Sales) Market-to-Book Profitability Tangibility Cash flow volatility Median Industry Book Leverage Median Industry Market Leverage Dividend Payer Intangible Assets Book Assets Initial Book Leverage Initial Market Leverage Definition Short-Term Debt + Long-Term Debt Stock Price * Shares Outstanding Total Debt / Book Assets Total Debt / (Total Debt + Market Equity) Natural logarithm of Total Sales (Market Equity + Total Debt + Preferred Stock Deferred Taxes and Investment Tax Credit) / Book Assets Operating Income Before Depreciation / Book Assets Net Property, Plant & Equipment / Book Assets Standard Deviation of Historical Operating Income Before Depreciation scaled by the Historical Mean of Operating Income Before Depreciation Median Level of Book Leverage within the firm-specific industry, based on two-digit Standard Industrial Classification Code (SIC). Median Level of Market Leverage within the firm-specific industry, based on two-digit Standard Industrial Classification Code (SIC). Dummy variable that is equal to 1 if the firm has paid any dividends in the timeframe 1999Q3 2012Q4. Intangible Assets / Book Assets Total Book Assets (in millions) First non-missing observation of Book Leverage for that specific firm in the dataset First non-missing observation of Market Leverage for that specific firm in the dataset 41

42 Variable Definition Compustat Quarterly Fundamentals Database Code (Item Number) Short-Term Debt Debt in Current Liabilities DLCQ (45) Long-Term Debt Long-Term Debt Total DLTTQ (51) Stock Price Price Close Quarter PRCCQ Shares Outstanding Common Shares used to Calculate CSHPRQ (15) Earnings Per Share Basic Book Assets Assets Total ATQ (44) Total Sales Sales/Turnover (Net) SALEQ (2) Preferred Stock Preferred / Preference Stock (Capital) PSTKQ (55) Total Deferred Taxes and Investment Deferred Taxes and Investment Tax TXDITCQ (52) Tax Credit Credit Operating Income Before Operating Income Before Depreciation OIBDPQ (21) Depreciation Quarterly Net Property, Plant & Equipment Property, Plant and Equipment Total PPENTQ (42) (Net) Dividend Payer Dividend Per Share Exdate Quarter DVPSXQ (16) Intangible Assets Intangible Assets Total INTANQ (52) 42

43 - Appendix 2 - This appendix presents the correlation matrixes for the independent variables out of the Multivariate Regression Analyses. Figure 3 relates the independent variables out of the Book Leverage Regression Analysis and figure 4 the independent variables out of the Market Leverage Regression Analysis. * represents statistical significance at the 1% level. Figure 3 Correlation Matrix Book Leverage Figure 4 Correlation Matrix Market Leverage 43