An Alternative Test of the Trade-Off Theory of Capital Structure

The purpose of this paper is to investigate the stochastic behavior of corporate debt ratios utilizing a balanced panel of 2,556 publicly traded U.S. firms during the period 1997-2010. We partition the panel into ten economic sectors and perform panel unit root tests on each sector employing book value and market value measures of debt ratio. First-generation panel unit root tests provide consistent evidence that debt ratios are mean reverting, which supports the trade-off theory. However, these tests rely on the assumption that the debt ratios are cross-sectionally independent, but tests of cross-sectional independence fail to uphold this assumption. Thus, utilizing a second-generation panel unit root test that controls for cross-sectional dependence, we uncover evidence showing that debt ratios are not mean reverting, which contradicts the trade-off hypothesis. We find that the recent macroeconomic developments triggered by the financial crisis and the Great Recession have considerable explanatory power over the dynamics of the debt ratios. In fact, when we exclude the years of the recent global financial crisis, the unit root hypothesis is rejected in one half of the sectors. We interpret these results as indicative that the recent global events may have produced in these sectors a structural change in the underlying data generation process (DGP). Overall, then, we find mixed evidence on the stationarity of debt ratios.

A particular concern about these models that has emerged in recent years is that they fail to include an assessment of the stochastic properties of debt ratios and ignore the issue of cross-sectional dependence. The first problem has been discussed at length by Granger and Newbold (1974) and exposes the econometric results to the spurious regression problem when data are non-stationary, i.e., contain unit roots. The second problem is particularly important in dynamic panel regressions. As noted by Phillips and Sul (2003), this substantially complicates the estimation and inference in dynamic panel models. Phillips and Sul (2003) address this problem from a theoretical perspective and propose an approach that is based on a panel version of the median unbiased estimator (Andrews, 1993).
The motivation of this study is twofold. First, unlike the vast bulk of the extant literature that focuses on the determinants of corporate capital structure, we rely on recent developments in the econometrics of non-stationary dynamic panel data. Specifically, we approach the analysis of the trade-off theory by assessing the stochastic properties of corporate debt ratio from the perspective of the panel unit root methodology. If the debt ratio is represented by a stationary process, shocks affecting the series are transitory, and the debt ratio will eventually return to its target level. Thus, evidence of stationarity supports the trade-off theory, as it characterizes the dynamics of capital structure as mean reverting. This situation, in turn, could be interpreted as an indirect signal of industry stability. Conversely, if the debt ratio evolves as a unit root process, shocks affecting the series have permanent effects, shifting the corporate capital structure from one level to another, which contradicts the trade-off theory. Second, we directly address the question of cross-sectional dependence in panel unit root tests. The application of univariate unit root tests, such as the Augmented Dickey-Fuller (Said & Dickey, 1984) and the Phillips-Perron (Phillips & Perron, 1988) tests, is somewhat commonplace in studies employing time series data. In contrast, the use of unit root tests for panel data is more recent (Im, Pesaran, & Shin, 2003;Levin, Lin, & Chu, 2002;Maddala & Wu, 1999). It is by now a generally accepted argument that the commonly used univariate unit root tests lack power in distinguishing the null hypothesis of unit root from stationary alternatives, and utilizing panel data unit root tests is one way of increasing the power of unit root tests (Choi 2001;Im et al., An alternative test of the trade-off theory of capital structure in the alternative. The LLC test, proposed by Levin et al. (2002), tests for the null hypothesis of the unit root against a homogeneous stationary hypothesis, i.e., the autoregressive parameter constrained to be the same across cross-section units, while the IPS test, suggested by Im et al. (2003), and the Fisher type tests developed by Maddala and Wu (1999) and Choi (2001) test for the null hypothesis of unit root against the heterogeneous alternative, i.e., the autoregressive parameter is allowed to vary across cross-section units. Surveys of panel unit root tests include, among others, Banerjee (1999), Breitung and Pesaran (2008), Gutierrez (2006), and Jang and Shin (2005). Unfortunately, however, testing the unit root hypothesis by employing panel data instead of individual time series is not without complications. In particular, the panel unit root literature has noted that in many empirical applications it may be inappropriate to assume that the cross-section units are independent. Observations on firms, industries, regions and countries normally tend to be crosscorrelated and serially dependent (Breitung & Pesaran, 2008). Thus, an important problem in panel unit root tests is whether the cross-sections of the panel are independent. On this issue, the panel unit root literature distinguishes between the first-generation tests, which are developed on the assumption of the cross-sectional independence, and the second-generation tests, which account for the dependence that might prevail across the different units in the panel. If the data are crosssectionally dependent, the panel unit root literature has demonstrated that the first-generation tests can generally be misleading, in the sense that they expose the tests to significant size distortions. That is, the tests tend to reject the null hypothesis of non-stationarity too often (see, for instance, Choi, 2001;Im et al., 2003;Levin et al., 2002;Maddala & Wu, 1999). Moreover, Pesaran (2007) demonstrates that panel unit root tests that do not account for cross-sectional dependence when cross-sectional dependencies are indeed present are seriously biased if the degree of cross-sectional dependence is sufficiently large. To date, only a few studies examine the corporate capital structure employing panel unit root tests. Chang, Liang, Su and Zhu (2010) use quarterly data over the period 1996:Q4-2007:Q3 from a panel of Taiwanese electronic firms and fail to reject the null hypothesis of unit root, except for the subsample of firms with low profitability. Bontempi and Golinelli (2001) (Levin et al., 1992) test, the IPS (Im et al., 2003) test, and the Maddala and Wu (Maddala & Wu, 1999) Fisher type tests. Bontempi and Golinelli (2001) apply the IPS test (Im et al., 2003), while Tasseven and Teker (2009) employ the LLC (Levin et al., 1992) Breuer, McNown and Wallace (2002), and the Cross-Sectionally Augmented ADF test (CADF) proposed by Pesaran (2007) address explicitly the problem of cross-sectional dependence.
The SURADF test is based on a system of augmented Dickey-Fuller (ADF) equations and estimates the autoregressive process by the Seemingly Unrelated Regression Equations (SURE) procedure; i.e., it accounts for cross-sectional dependence by directly incorporating the variance-covariance matrix of the residuals of the equations system in the estimation process. The advantage of this approach is that it allows identification of the cross-sectional units of the panel that contain a unit root (Lau, Baharumshah, & Soon, 2013 There are a variety of reasons why cross-sectional dependence may exist in an industry. Commonly, cross-sectional dependence reflects the fact that firms in the same industry respond to unobserved common stochastic shocks and are linked by unobserved common stochastic trends. Common shocks and common trends spread across all firms in an industry, thus engendering the panel feature of cross-sectional dependence. Monetary and fiscal shocks frequently provide the channels that generate common stochastic shocks. For example, monetary shocks in the supply of money and fiscal shocks in the supply of government debt affect the rate of inflation and the structure of interest rates, which in turn influence the firm's cost of capital and the equilibrium of financial markets, leading to changes in the financial constraints in the corporate sector and alternative representations of the corporate capital structure (Bokpin, 2009;Frank & Goyal, 2009). Furthermore, in a globalized economy, shocks generated in one country are known to cross national borders (Lau, Baharumshah and Soon, 2013). This phenomenon is especially true for oil shocks. The global financial crisis is arguably one of the deepest exogenous shocks that recently affected the corporate sector. The credit supply shock (Dang et al., 2014) originated by the subprime crisis has affected the corporate demand for and supply of funds and, consequently, the capital structure. Common stochastic trends, however, are another source of cross-sectional dependence, as they reflect the presence of corporate variables that tend to move together, i.e., are cointegrated in a VAR system (Granger, 1981). Empirical evidence, for instance, has found that stable relationships exist at the industry level between measures of firm performance, such as sales or profitability, and research and development expenditures (Chan, Lakonishok, & Sougiannis, 2001) and between the market value added of the firm (MVA), an external measure of a firm's performance, and several internal measures, such as earnings per share (EPS), free cash flow per share (FCF), return on equity (ROE), return on assets (ROA), and economic value added per share (EVA) (Bernier & Mouelhi, 2012).
This study contributes to the empirical capital structure literature in several ways. First, as mentioned above, our methodological approach enables us to fill a gap in the existing literature by focusing on an alternative stochastic process that might be more consistent with the long-run behavior of debt ratios. Existing empirical work has focused almost exclusively on the relationships between corporate capital structure and its determinants. While these studies have produced a great deal of evidence on the association between capital structure and its determinants, they have not been able to provide much evidence on the dynamics of debt ratios. Our methodology is based on a panel unit root test that allows for alternative assumptions of cross-sectional dependency for capital structure adjustments. Surveys of panel unit root tests include, among others, Breitung and Pesaran (2008), Banerjee (1999), Gutierrez (2006, and Jang and Shin (2005). Panel unit root tests exploit both the time-series (t = 1, 2…T) and cross-section (i = 1, 2…N) dimensions of the underlying data, thereby having more power and greater efficiency than conventional time series unit root tests (Baltagi, 2005). This type of analysis is not new in corporate finance. Tippett (1990), for example, models financial ratios in terms of stochastic processes, and Tippett and Whittington (1995) and Whittington and Tippett (1999) report empirical evidence that the majority of financial ratios exhibit random-walk behavior. A unit root process imposes no bounds on how a series moves. If the debt ratio really conforms to a random-walk process, then it is unpredictable.
A presumption of the trade-off theory is that managers make capital structure decisions based on a target debt ratio and that shocks affecting the debt ratio will prove transitory. This implies that debt ratios are mean reverting towards a target level and follow a stationary dynamic. Conversely, if managers do not make decisions based on a target debt ratio, shocks re-Electronic copy available at: https://ssrn.com/abstract=2548424 An alternative test of the trade-off theory of capital structure sult in permanent shifts in the debt ratio. In this case, change in the debt ratio evolves as a unit root, nonstationary process, which is consistent with alternative capital structure theories, such as the pecking order or the market timing theories (Baker & Wurgler, 2002;Myers & Majluf, 1984). Non-stationarity of the debt ratio differs from persistence. Persistence involves a slow process of adjustment to an optimal level, while non-stationarity implies that debt ratios fluctuate randomly, driven only by stochastic shocks without a tendency to return to a mean. Therefore, non-stationarity implies that firm debt ratios exhibit a unit root, while persistence suggests that firm debt ratios exhibit a near unit root. It is important to note that the dynamic partial adjustment models currently utilized in the literature are based on assumptions that capital structure adjustments are mean reverting and these adjustments are cross-sectionally independent across firms (Fama & French, 2002;Flannery & Rangan, 2006;Frank & Goyal, 2003;Huang & Ritter, 2009;Leary & Roberts, 2005;Shyam-Sunder & Myers, 1999;Welch, 2004). Evidence on the stochastic properties of the debt ratios also possesses welldefined implications for econometric modeling and forecasting. Failure to reject the unit root hypothesis potentially implies that debt ratios exhibit a long-run cointegrating relationship with other firm-level data, while rejecting the unit root hypothesis implies that debt ratios exhibit only a short-term relationship with other corporate series. Rejecting or not rejecting the unit root hypothesis, in turn, profoundly affects the forecasting process because forecasting based on a mean-reverting process proves quite different from forecasting based on a random walk process.
Second, we control for effects related to the economic sector when analyzing the stochastic properties of debt ratios. We accomplish this by stratifying the data into ten sectors and examining the stochastic properties of debt ratios within each sector. Debt ratios have been found to exhibit significant differences across sectors (Bradley et al., 1984;Lemmon et al., 2008). Graham and Harvey (2001) found that one third of their sample had debt ratios lower than 0.20, and another third had debt ratios higher than 0.40. This stratification is done because of distinct differences in debt ratios across economic sectors, and the extent and speed of reversion of a firm's debt ratio to its target may vary by sector. Firm-level data for each sector are obtained by partitioning a large panel of 2,556 U.S. public companies during the period 1997-2010. Because we partition the sample into sectors, we employ the average debt ratio for the sector as a benchmark. We first examine the evolution of the debt ratios over the entire sample period 1997-2010.
The financial literature, however, has recognized that the turbulent and volatile macroeconomic environment created by the recent financial crisis and the resulting Great Recession had severe effects on corporate financial policies (Campello, Graham, & Harvey, 2010;Campello et al., 2011;Duchin, Ozbas, & Sensoy, 2010). Thus, it would seem prudent to evaluate the robustness of the panel unit root results with the events of the global financial crisis and the Great Recession. To account for this problem, we date the financial crisis with the year of the Lehman Brothers bankruptcy. We then construct a "pre-crisis" sub-sample, 1997-2007, and investigate whether this sample reduction has affected our findings. In this respect, our paper adds to the nascent literature that documents the negative impact of the recent financial crisis on corporate debt ratios (Dang et al., 2014).
Third, we measure debt ratios using both book values and market values. Book value and market value debt ratios are conceptually different. Book measures are by definition "backward looking" because of their reliance on accounting data, whereas market values are generally held to be "forward looking". Therefore, differences between the movement of book value and market value debt ratios may be sizeable (Barclay & Morellec, 2006). Rajan and Zingales (1995) and Welch (2004) provide indepth rationale for analyzing both.
The main findings of our paper can be summarized as follows. First, we find that cross-sectional dependence does matter and substantially affects the outcome of the tests. When we apply conventional, first-generation panel unit root tests that are based on the assumption of crosssectional independence, we find results that lead to the rejection of the unit root hypothesis. This evidence is consistent with mean reversion of debt ratios and, therefore, supports the trade-off hypothesis. However, to determine if these first-generation tests are appropriate, we utilize diagnostic tests developed by Pesaran (2004) andFrees (1995;. Second, we find strong evidence of sub-stantial cross-sectional dependence within our sample indicating that the assumption of cross-sectional independence is inappropriate. Third, the Pesaran (2007) panel unit root test that allows for cross-sectional dependence consistently yields results supporting the unit root hypothesis, which is inconsistent with debt ratios being mean reverting. This evidence is contradictory to the trade-off hypothesis. Of course, the failure to formally reject a null hypothesis of unit root does not, on its own, rule out the existence of some important struc-

Panel Unit Root Tests and the Corporate Debt Ratio
In this section, we outline a dynamic panel model of corporate debt ratios that provides a theoretical background for the application of panel unit root tests. Let ε is a zero-mean white noise process. Equations (1) and (2) jointly imply the following stationary autoregressive process, Equivalently, equation (3) can be given the augmented where ∆ is the difference operator, and Solving equation (4) for i ρ = 0 reduces to the unit root Equation (5) implies that when there is a shock i t ε at time t, the debt ratio changes in the long run by . In other words, this suggests the shock has a permanent effect, which is inconsistent with the trade-off hypothesis. Under the null hypothesis, We also stratify the sample into ten economic sectors, following the Compustat economic sector (ECNSEC) classification scheme, and perform panel unit root tests on each sector utilizing our two alternative debt ratio measures. The ten sectors are (the number of firms is reported in parenthesis): 1) Materials (187); 2) Consumer Discretionary (420) Titman and Wessels (1988) and Rajan and Zingales (1995). Myers (1977) and Fama and French (2002) favor the use of book values, while Welch (2004) advocates the use of market values. Drobetz, Pensa and Wanzenried (2007) discuss the advantages and disadvantages of each measure. We follow Rajan and Zingales (1995) and define leverage as the ratio of financial debt to debt plus equity. We include short-term debt in the definition of the debt ratio as its omission may lead to an understatement of financial distress risk. We consider both the book and market values of equity because it is highly possible that some firms operate within a book value framework rather than a market value framework, and vice versa. The book value of the debt ratio i t BDR of firm i at time t is defined as follows: where i t η is the number of shares outstanding (Compustat annual data item 54) and i t P denotes the stock price (Compustat annual data item 199).
In Table 1   We also find that the empirical distributions of each measure of debt ratio are generally non-symmetric. In each case, the mean is greater than the median, implying that the distribution is positively skewed (longer tails to the right). there was evidence of extreme values, the analysis also conducted "winsorizing" of the top and bottom 5% of the data. This approach was used to eliminate any unexpected effects of outliers. There was no meaningful effect on the results. We do not report the winsorized results, but these are available upon request. Table 1 reveals the potential problem of aggregating all obser- Electronic copy available at: https://ssrn.com/abstract=2548424 An alternative test of the trade-off theory of capital structure

Empirical Results
We present the empirical evidence in three stages.
First, we perform the panel data statistical analysis utilizing Fisher type tests (Choi, 2001;Maddala & Wu, 1999). These tests are nonparametric and have the advantage of allowing for as much heterogeneity across units as possible. They belong to the first generation of panel unit root tests, which include among others, Levin et al., (2002), Im et al., (2003), Harris and Tzavalis (1999). The findings of these tests uniformly favor the trade-off hypothesis. However, the concern is that these tests are not robust in the presence of cross-sectional dependence. In other words, these first-generation tests employ a methodology that incorporates the often implausible assumption of cross-sectional independence and fail to discriminate between stationarity with cross-sectional independence and non-stationarity with cross-sectional dependence. The power of the conventional panel unit root tests is weakened by the presence of cross-sectional dependence. Therefore, we next test this assumption of cross-sectional independence utilizing the approaches suggested by Pesaran (2004) and Frees (2004). We find exhaustive evidence indicating the presence of heterogeneous cross-sectional dependencies among the time series, which calls for an alternative test methodology. Consequently, we utilize a second-generation panel unit root test that accounts for cross-sectional dependence based on the methodology of Pesaran (2007). We conclude this section with an assessment of the impact of the recent financial crisis on the results obtained using the full sample.

Results of the panel unit root tests under the assumption of cross-sectional independence
We first implement the Fisher type (Fisher, 1932) unit root tests developed by Maddala and Wu (1999) and Choi (2001) The m P test is a modified version of the Maddala and Wu's (1999) P test applied to large panels because in the limit the P test statistic has a degenerate distribution.
, where LL is the log of the likelihood function with k parameters estimated using T observations. In the Fisher type PP unit root test, the spectral regressions employ the Bartlett kernel in conjunction with the Newey-West bandwidth selection. For economy of space, we do not report the results of the * L and m P tests because they are uniformly consistent with the results of the P and Z tests. We do not include a time trend be-cause a time trend is not consistent with a long-run positive, non-accelerating target debt ratio. However, we do include an intercept because the average debt ratio is nonzero. We perform all tests using the "demeaned" version (i.e., we subtract the cross-sectional means from observed data to reduce the degree of contemporaneous correlation) and, in the Fisher type ADF tests, we include one lag (to account for serial correlation) chosen by AIC. Subtracting the crosssectional means from the observed data is a strategy suggested by Levin et al. (2002) and Im et al. (2003)     Electronic copy available at: https://ssrn.com/abstract=2548424 pendence is driven by a common factor that has a homogeneous effect on all firms in the industry, regardless of their size. This assumption is highly unrealistic for most practical settings because it ignores the heterogeneous impact of short-run co-movements (common cycles) and long-run comovements (common trends) on the dynamics of firms within the same industry (O'Connell, 1998).
The presence of heterogeneous cross-sectional dependencies undermines the power of the Maddala and Wu (1999) and Choi (2001) tests, leads to false rejections of the null hypothesis of the unit root, and may produce evidence of stationarity when the data are non-stationary. In the next section, we address this issue by testing for cross-sectional dependence using the diagnostic tests proposed by Pesaran (2004) and Frees (2004). Pesaran (2004) proposes a general test for crosssectional dependence referred to as the CD test.

Results of the tests for cross-sectional dependence
As demonstrated by Pesaran (2004), the CD test applies to a large variety of panel data models.
This includes stationary and non-stationary dynamic heterogeneous panel models having a small T (years) and a large N (firms), which is the case for the sample panel data employed in this study.
The CD test applies to both balanced and unbalanced panels, is robust to parameter heterogeneity and structural breaks in the slope coefficients and error variance, and performs well in terms of size and power. Under the null hypothesis, the covariance matrix of the residuals is diagonal, i.e., Frees (1995,2004) proposes a statistic that is not subject to this shortcoming. The statistic is based on the sum of the squared correlation coefficients and is given by: where i εˆand j εˆ are the residuals obtained from the same models estimated for the CD test. Frees (1995Frees ( , 2004 demonstrates that a function of 2 AVE R follows a joint distribution of two independent chi-square variables, i.e.
Electronic copy available at: https://ssrn.com/abstract=2548424 An alternative test of the trade-off theory of capital structure where q Q is the appropriate quintile of the Q distribution.
We report the findings of the two diagnostic tests in Tables 4 and 5. The outcomes of these tests clearly indicate the presence of cross-sectional dependence in both the book value and market value debt ratios. The tests strongly reject the null hypothesis of cross-sec-tional independence at any conventional significance level. This situation casts doubt on the statistical evidence in favor of stationarity by the Fisher type tests.
In addition, the estimates of the residuals correlation coefficients present a wide range of variability, suggesting that residual correlation is heterogeneous rather than homogeneous. For economy of space, the matrices of the estimates of residual correlation coefficients are not reported, but are available on request.
To summarize, the rejection of the null hypothesis of cross-sectional independence implies that tests for the presence of a unit root in book value and market value debt ratios should take this dependence into account to produce unbiased and reliable test statistics. These findings call into question any conclusions drawn from the Fisher type tests. The next section ad-

Results from the panel unit root tests under the assumption of cross-sectional dependence
In this sub-section, we investigate the stationarity property of the two measures of debt ratio by applying the panel unit root test developed by Pesaran (2007).
The test assumes that cross-sectional dependence is present in the data in the form of a single unobservable common factor. The test expands on the Im et al. The test is a two-step procedure. First, Pesaran (2007) proposes a test on the t-ratio of the OLS estimate of i β in the following cross-sectionally augmented ADF   Electronic copy available at: https://ssrn.com/abstract=2548424 An alternative test of the trade-off theory of capital structure null of non-stationarity, the CIPS statistic has a nonstandard distribution even for large N, but has good size and power properties even when T and N are relatively small. The critical values for 1%, 5% and 10%, however, are tabulated in Pesaran (2007). In addition, Pesaran (2007) constructs a truncated version of the CIPS, denoted as * CIPS , to avoid the problem of an extreme statistic in cases when T is small, and to ensure the existence of the first and second moments of The truncated test is given by where (22) ) 2 2 ( ) , (  Tables 2-3. While Tables   2-3 provide evidence of stationarity of debt ratios, the evidence presented in Table 6 demonstrates the opposite. After controlling for heterogeneous cross-sectional dependence, the evidence reveals a non-stationary (unit root) debt ratio process in the vast majority of the sectors. Therefore, we cannot reject the null hypothesis of a unit root in all ten sectors for the market value debt ratios, and in nine out of ten sectors for book value debt ratios. This failure to reject the stationarity of debt ratios is consistent with hypotheses that do not envisage the existence of a target debt ratio and an adjustment process toward it. It indicates that borrowing is not driven by an attempt to move toward a target capital structure, but instead indicates that borrowing is driven by a need for external funds that is consistent with the pecking order and market-timing hypotheses.
Furthermore, the acceptance of the unit root indicates that random shocks have permanent effects on a firm's capital structure. Debt ratios behave as a stochastic process driven year-after-year by external shocks that affect firms. To demonstrate the robustness of our findings, we checked whether our results are sensitive to our measures of debt ratio. We repeated the panel unit root tests presented in Tables 2-6

The Impact of the financial crisis on the stochastic properties of debt ratios
Do our results really invalidate the trade-off model?
We argue that it is premature to make such a conclusion. Up to this point, we have assumed that throughout the entire sample period the stochastic process representation of debts ratios does not exhibit structural change. In the case, when this assumption fails, the tests can be misleading and biased toward the non-rejection of the unit root hypothesis. Thus, some caution should be exercised in interpreting our findings of non-stationarity under cross-sectional dependence because the global financial crisis and the resulting Great Recession are included in our sample. As shown by Dang et al. (2014), the speed of adjustment of corporate debt ratios has been significantly affected by the financial crisis. Using the dummy variable approach, Dang et al. (2014)   The findings for the Fisher type tests do not modify the conclusions drawn for each sector using the original sample. The results of these tests, in both the ADF and the PP specifications, reject the unit root hypothesis for both the book value and the market value of the debt ratio series. Similarly, the findings of the CD and C AVE remain robust to the sample reduction. That is, we find strong evidence of cross-section dependence in each sector for both measures of debt ratio. We do not report these findings, but they are available on request. However, some of the findings for the CIPS* are sensitive to the time period. We find more evidence of stationarity in the "pre-crisis" sample than in the Electronic copy available at: https://ssrn.com/abstract=2548424  -2.19 and -2.07, and -2.00; T = 15, N = 50: -2.26, -2.11, and -2.03, and denoted as. *, **, and ***, respectively.
original sample. Furthermore, we find more evidence of stationarity for the market value debt ratio than the book value debt ratio. This outcome was to be expected because book values are largely unaffected by changes in stock prices. In Table 7, the CIPS* test rejects the null hypothesis of the unit root for the market value debt ratio in Materials, Consumer Staples, and Utilities at the 5% significance level, and Industrials at the 1% significance level. Similarly, the CIPS* test rejects the null hypothesis for the book value debt ratio in Health Care at the 5% significance level. Thus, at least one measure of the debt ratios in these sectors appears to exhibit a reversal in dynamics, from unit root to mean reversion. Because the results for the full sample indicate non-stationarity and the results of the reduced sample suggest stationarity for these five sectors, we conclude that in these sectors the financial crisis has destabilized debt ratios, switching the dynamics of the debt ratios from a mean reversion behavior to a random-walk dynamics. We interpret these results as in-dicative that for these sectors the recent global events may have triggered a structural break in the underlying data generation process. In the face of increased risk aversion by credit suppliers and widespread informational asymmetries, this outcome is not shocking.
In such an environment, a pecking order may be generated, where retained earnings represent the least expensive source of financing. For the remaining sectors (Consumer Discretionary, Energy, Financials, Information Technology, and Telecommunication Services) instead, the financial crisis does not appear to have affected the unit root dynamics of debt ratios. The resilience of these sectors to the crisis may be indirect evidence that internal financing plays a non-trivial role in the determination of debt ratios. This lack of uniformity of our findings is not surprising and is consistent with the idea that the recent financial crisis has not had a homogeneous impact on the U.S. economy (Dang et al., 2014). Thus, overall, we find that the evidence on the debt ratios dynamics is mixed.
ics of corporate capital structure. Do firms have target debt ratios? The literature on corporate capital structure suggests at least three possible mechanisms for explaining the determinants of debt ratios: the trade-off theory, the pecking order theory, and the market-timing theory. Existing empirical work has focused almost exclusively on the determinants of capital structure, and while they have produced substantial evidence on the relation between capital structure and its determinants, they have not been able to provide much evidence on the dynamics of debt ratios. This study brings new evidence to bear on this important issue. We approach the question from the viewpoint of the methodology of panel unit root tests and investigate whether debt ratios are mean reverting or alternatively exhibit a random-walk process. If the empirical findings provide evidence of stationarity, this is an indication that the dynamics of the debt ratios are mean reverting, and, consequently, firm financial behavior follows the trade-off theory. Otherwise, if the empirical results provide evidence of unit root dynamics, this signals that firm financial behavior evolves according to other theories of capital structure, such as the pecking order theory or the market-timing theory.
Employing a panel of 2,556 US public firms over the period 1997-2010, we investigate the stationarity properties of the book value and market value measures of debt ratios for ten economic sectors of the U.S. economy. We first employ Fisher type panel unit root tests and find evidence that is overwhelmingly favorable to a mean reversion, i.e., stationarity hypothesis, and, consequently, the trade-off theory. This finding is consistent with much of the literature. However, these first-generation tests rely on the assumption of cross-sectional independence. Our analysis provides evidence that this assumption is not supported by the data. Cross section dependence does matter and substantially affects the outcome of the tests. Thus, when we apply the second-generation panel unit root test developed by Pesaran (2007) that accounts for this dependence, the results challenge the notion that debt ratios are mean reverting. We view these findings as evidence that contradicts the trade-off theory, but is consistent with the pecking order and the market-timing theories. We perform a robustness check on our findings and consider whether the results of the panel unit root tests are sensitive to the selection of the sample period.
We find that the recent macroeconomic events of the global financial crisis and the Great Recession play a crucial role in our understanding of the dynamics of debt ratios. We construct a pre-crisis sample that excludes the last three years at the end of our full sample.
The results of this sample reduction generate more evidence of stationarity. Utilizing the market value debt ratio, four sectors (Materials, Consumer Staples, Utilities, and Industrials) exhibit stationary dynamics, while employing the book value debt ratio, one sector (Health Care) exhibits stationarity. We interpret these results as indicative that the recent global events may have produced in these sectors a structural change in the underlying data generation process (DGP).