What determines the speed of adjustment to the target capital structure?

A dynamic adjustment model and panel methodology are used to investigate the determinants of a time varying target capital structure. Because firms may temporarily deviate from their target capital structure in the presence of adjustment costs, the adjustment process is also endogenized. Specifically, we analyse the impact of firm-specific characteristics as well as macroeconomic factors on the speed of adjustment to the target debt ratio. The sample comprises a panel of 90 Swiss firms over the years from 1991 to 2001. We document that faster growing firms and those that are further away from their optimal capital structure adjust more readily. The results also reveal interesting interrelations between the adjustment speed and well-known business cycle variables. Most important, the speed of adjustment is higher when the term spread is higher and when economic prospects are good.


I. Introduction
Capital structure is arguably at the core of modern corporate finance. While Modigliani and Miller (1958) derived conditions under which capital structure is irrelevant for firm valuation, the subsequent theoretical literature has shown that a firm can influence its value and improve its future prospects by varying its optimal ratio between debt and equity. 1 The empirical literature, especially for European countries, could not keep up with the pace of theoretical developments for two reasons. First, reliable data has been available only recently for European firms. This paper employs a panel of 90 Swiss firms over the period from 1991 to 2001.
Second, and perhaps more relevant, early tests of capital structure theories suffered from several shortcomings that are partly explainable by a lack of appropriate econometric methods. While the well-established theories explain differences in the optimal debt-equity ratio across firms, most of the empirical literature applied a static framework, using the observed debt ratio as a proxy for a firm's optimal leverage ratio. For example, Titman and Wessels (1988) for US data and Rajan and Zingales (1995) for international data document that leverage is related to firm-specific characteristics, such as profitability, investment opportunities, tangibility of assets, and earnings volatility. 2 However, as forcefully argued by Heshmati (2001), theories of the capital structure do not explain observed differences in debt ratios but rather the differences in optimal leverage ratios across firms. Using observed debt ratios is particularly problematic if adjustment to the optimal capital structure is costly. In the presence of adjustment costs, it might be cheaper for firms not to fully adjust to their targets even if they recognize that their existing leverage ratios are not optimal.
Standard capital structure models cannot capture dynamic adjustments in leverage ratios. Recent survey evidence by Graham and Harvey (2001), Brounen et al. (2004) and Drobetz et al. (2006) strongly suggests that firm decision-makers seek a target debt-equity ratio. Their main objective in setting debt policy is not to minimize a firm's weighted average cost of capital, but rather to preserve financial flexibility, which is best explained in the context of the pecking order theory of the capital structure. Nevertheless, due to shocks or other random changes, firms may temporarily deviate from their optimal capital structure and then only gradually work back to the optimum. To account for these stylized facts, several researchers have adopted a dynamic model approach, where observed and target leverage may differ due to the presence of adjustment costs. For example, Fischer et al. (1989) study the difference between a firm's maximum and minimum debt ratios over time and identify characteristics of firms with larger swings in their capital structures. They use the observed debt ratio range of a firm as an empirical measure of the target capital structure. Their results are consistent with capital structure choice in the presence of adjustment costs in a dynamic setting. 3 In an even earlier paper, Jalilvand and Harris (1984) document that a firm's financial behaviour is characterized by partial adjustment to long-run financial targets. In their setup, the speed of adjustment is affected by firm characteristics and therefore varies across companies and over time. However, the long-run financial targets towards firms partially adjust are specified exogenously. Most recently, Shyam-Sunder and Myers (1999) and Fama and French (2002) also use the historical mean debt ratio for each firm over the sample period as a proxy for the target debt ratio.
Using Spanish data, De Miguel and Pindado (2001) present a novel methodology to capture the dynamics of capital structure decisions more appropriately. They develop a target adjustment model that allows explaining a firm's leverage in terms of its debt in the previous period and its target debt level, the latter being a function of well-known firm characteristics, such as profitability, growth, and tangibility of assets. Their setup endogenizes the target debt ratio, which allows identifying the determinants of the optimal or target capital structure rather than the observed one. They specify a dynamic adjustment model with predetermined variables and apply the dynamic panel estimator suggested by Arellano and Bond (1991). It is crucial to note that De Miguel and Pindado (2001) estimate a constant adjustment coefficient. Their empirical results reveal that Spanish firms face lower adjustment costs than US firms. 4 While these papers constitute important steps towards more realistic tests of capital structure theories, they still remain silent on which factors determine the adjustment process to the target leverage ratio. Banerjee et al. (2000) were the first to simultaneously endogenize both the adjustment factor and the target debt ratio. In addition to identifying the determinants of target capital structure, their setup allows estimating the speed of adjustment towards the target capital structure and identifying the determinants of the speed of adjustment simultaneously. Using US and UK data, they hypothesize that the speed of adjustment is dependent on the absolute difference from the target debt ratio, growth opportunities, and firm size. Contrary to what they expected, their results reveal that firms with higher growth opportunities adjust slower towards the target capital structure, and that larger firms adjust to changes in capital structure more readily. However, they do not find a significant relationship between the likelihood of adjustment and the absolute difference between target leverage in time t and observed leverage at the end of period t À 1. 5 In a related paper, Lo¨o¨f (2003) compares the dynamics of capital structure adjustments across the two archetypes of financial systems, the US and UK arm's length system (mostly marketbased system) and the Swedish relation-based system (mostly bank-based system). His results reveal that although firms are frequently not at their target level, the deviation is smaller for the highly equity dependent US firms. In addition, these firms adjust faster towards their target capital structure compared to 3 Consistent with Myers' (1984) pecking order theory, Hovakimian et al. (2001) document that profitability is a predictor of observed debt ratios in the short-run. Nevertheless, firms make financing and repurchase decisions that offset these earningsdriven changes in their capital structure. This supports the static trade-off theory of the capital structure and the existence of a target debt ratio. However, the latter may not be a primary goal of firms' decision-makers (see Panno, 2003;Leary and Roberts, 2004). 4 Gaud et al. (2005) and Drobetz and Fix (2005) adopt this approach for Swiss data. 5 See Heshmati (2001) for similar results using a sample of Swedish micro and small firms. the debt dependent Swedish companies. Using similar variables to capture the speed of adjustment as Banerjee et al. (2000), he finds that the estimate for the distance variable is significantly negative for UK firms, indicating that it is less costly to adjust by relatively small amounts.
Economic intuition suggests that the position of the economy in the business cycle phase is an important determinant of default risk and, hence, of financing decisions. It is therefore an interesting research question to analyse the impact of macroeconomic factors on the speed of adjustment to the target capital structure. Lacking well-defined empirical predictions, previous studies included a set of time dummy variables to capture these effects. Recently, Hackbarth et al. (2005) develop a contingent claims model in which a firm's cash flows depend on both an idiosyncratic shock and an aggregate shock that reflects the state of the economy (e.g., boom or recession). Their model delivers state-dependent shareholders' default policies, which in turn have interesting implications for optimal leverage. First, the model predicts that leverage is counter-cyclical. Second, macroeconomic conditions determine both the pace and the size of capital structure changes. Allowing a firm to adjust its capital structure dynamically, the restructuring threshold is lower in good states than in bad states. Therefore, firms should adjust their capital structure more often and by smaller amounts in booms than in recessions. The empirical results by Korajczyk and Levy (2003) support some of these predictions. Looking at a 50 years history of the US aggregate non-financial corporate debt to asset ratio, they show that target leverage is counter-cyclical, i.e., there is a negative relation between macroeconomic variables and leverage. Note that this is consistent with the pecking order theory of the capital structure, but inconsistent with a trade-off theory. In a theoretical model, Levy (2001) argues that levered managers' wealth is reduced relative to outside shareholders in recessions, which exacerbates the agency problem. In order to realign managers' incentives with those of shareholders, the optimal amount of debt increases, implying counter-cyclical debt ratios for firms that are not financially constrained. Korajczyk and Levy (2003) document that macroeconomic conditions are important for issue choices. Firms tend to time their issues to periods of favourable macroeconomic conditions, i.e., periods when the relative pricing of the security issued is favourable. Most important, firms issue equity when the stock market experienced large run-ups and when economic prospects are good, as indicated by popular business cycle variables (e.g., term spread and default spread). 6 However, the findings are not uniform across their sample. Financially constrained firms exhibit a pro-cyclical target leverage ratio, and their issue choice is less sensitive to variations in macroeconomic conditions than unconstrained firms. Intuitively, financially constrained firms are not able to time their issues.
We investigate the adjustment process to the target capital structure using a sample of 90 Swiss firms over the 1991-2001 period. In particular, the effects of firm-specific characteristics as well as macroeconomic factors on the speed of adjustment to the target leverage are analysed. We document that faster growing firms and those that are further away from their target capital structure adjust more readily. We also demonstrate that the speed of adjustment is dependent on the stage of the business cycle. Using well-known business cycle variables, our results reveal that the speed of adjustment to the target debt ratio is faster when economic prospects are good. However, systematic differences in the adjustment speed cannot be detected when we distinguish between financially constrained and unconstrained firms. One could hypothesize that the sensitivity of the adjustment speed to business cycle variables is larger for financially unconstrained firms.
The remainder of this paper is as follows. Section II starts by developing a dynamic capital structure model and introducing familiar determinants of the target capital structure. We proceed with a discussion of firm-specific and macroeconomic determinants of the speed of adjustment to the target capital structure. Section III describes our panel of Swiss company data. Section IV contains the empirical results, and Section V concludes.

II. The Dynamic Framework
In this section the framework for the dynamic capital structure model is discussed. First, the model and the estimation methodology are presented. Second, the determinants of the target capital structure are described and finally the variables that influence the speed of adjustment to the target debt ratio will be introduced.

A. A dynamic capital structure model
Following work by Heshmati (2001), De Miguel andPindado (2001), and Hovakimian et al. (2001), a dynamic capital structure model is considered. Let the optimal or target debt level of firm i in period t, labelled as LV Ã it , be a linear function of a set of L explanatory variables, X jit (where j ¼ 1, 2, . . . , L), that have been used in past cross-sectional studies of capital structure: Note that this dynamic setup implies that the target debt ratio may vary both across firms and over time.
Without frictions, the observed leverage ratio of firm i at time t, denoted as LV it , should be equal to the target leverage ratio, i.e., LV it ¼ LV Ã it . The purpose of Equation 1 is to provide an estimate of each firm's target leverage ratio, which we define as the debt ratio that firms would choose in the absence of information asymmetries, transaction costs, and other adjustment costs (e.g., Hovakimian et al., 2001;De Miguel and Pindado, 2001). However, if adjustment is costly, firms may not fully adjust their actual debt ratio from the previous period to the current target debt ratio. The notion of partial adjustment is usually formalized as follows: where it captures the speed of adjustment to the target debt ratio, starting from previous year's leverage ratio, labelled LV itÀ1 . The existence of adjustment costs is represented by the restriction that it j j < 1, which implies that LV it ! LV Ã it as t ! 1. In contrast, if it ¼ 1, all adjustment is made instantaneously, and a firm's debt ratio is always at the target. In the presence of adjustment costs, it is expected that it < 1, hence, a firm does not fully adjust from period tÀ1 to period t. Finally, if it > 1, a firm adjusts more than would be necessary and is still not at its target debt level. 7 In other models of debt adjustment, the optimal level of debt is externally determined either in terms of historical data or by an adjustment process with lags of more than one year (e.g., Jalilvand and Harris, 1984;Shyam-Sunder and Myers, 1999). The proposed model follows De Miguel and Pindado (2001) and Hovakimian et al. (2001), where firms adjust to a target debt ratio that is not determined externally as in previous studies. Instead, the target debt ratio is included in the model as a linear function of the determining factors of capital structure, as specified in Equation 1. In the following analysis we extend this class of models and also endogenize the speed of adjustment to the target debt ratio. To explain the speed of adjustment, it is assumed that it varies over time and is itself a linear function of a constant term and some predetermined explanatory variable. A determinant variable of the speed of adjustment, which is labelled as Z it , is either a firm-specific or a macroeconomic variable (see subsection C). Specifically: To keep the estimation problem tractable, different determinants of the adjustment speed are applied separately one at a time. Therefore, Z it is not a vector as is X jit (with j ¼ 1, 2, . . . , L) in Equation 1, but rather a scalar. In the case of macroeconomic variables as determinants of adjustment speed, in particular, this procedure avoids potential multicollinearity problems. 8 When firm-specific variables are used to explain the speed of adjustment, Z it has both a timeseries and a cross-sectional dimension. In contrast, in the case of macroeconomic variables, Z it is not firm-specific and does not have a between-dimension, hence, the subscript it is replaced by t.
Rewriting the target adjustment model in Equation 2, treating target leverage, LV Ã it , as linearly dependent from the capital structure determinants as specified in Equation 1, and substituting the linear specification for adjustment speed, it , from Equation 3, yields the following expression for the leverage ratio at time t: where u it is a statistical error term with mean zero and constant variance (i.e., white noise disturbance). Multiplying Equation 4 out and bearing in mind that all estimations are carried out with panel data, Equation 5 is obtained, which is subject to our empirical investigation: 7 According to Lo¨o¨f (2003), overadjustment may reflect unanticipated changes in economic conditions. 8 See Table 3 for a correlation analysis of the respective variables.
where d t is a time-specific effect, and i is a firmspecific effect. It is assumed that firm-specific effects are unobservable but have a significant impact on leverage. They differ across firms but are fixed for a given firm over time. In contrast, the time-specific effects vary over time but are the same for all firms in a given year, capturing mainly economy-wide factors that are outside the firm's control. 9 When Equation 5 is estimated, interest is mainly in 1 , which is the coefficient on the interaction term between the determinant variable of adjustment speed, Z it , and lagged leverage, LV itÀ1 . The null hypothesis is that 1 ¼ 0, i.e., the speed of adjustment is independent from firm-specific characteristics and/or the business cycle. However, this does not mean that firms do not adjust their debt ratios at all over time; this would only be the case if (1À 0 ) was estimated insignificantly as well. 10 Using panel data, Banjeree et al. (2000) and Lo¨o¨f (2003) apply non-linear least squares to estimate the parameters in a setup that is similar to that in Equation 5. However, this estimation technique leads to biased and inconsistent estimators because the error term will be correlated with lagged leverage, LV itÀ1 . Therefore, the dynamic leverage model is estimated by controlling for fixed-effects by applying a first-difference transformation. Even if unobservable firm-specific effects are not correlated with the regressors, it is still necessary to control for them in the dynamic framework. This is because LV itÀ1 will be correlated with i that does not vary through time, and a first-difference transformation to eliminate fixed effects introduces correlation between lagged dependent and differenced errors. Therefore, ÁLV itÀ1 and Áu it will be correlated through terms LV itÀ1 and u itÀ1 , and OLS will not consistently estimate the coefficient parameters. 11 Another estimation problem, not necessarily specific to the dynamic specification, arises because the firm-specific variables are unlikely to be strictly exogenous. Shocks that affect the leverage of firms are also likely to affect some of the regressor variables, such as firm profitability and firm size. Furthermore, it is likely that some of the regressor variables are correlated with past and current values of the idiosyncratic component of disturbances.
The problems described above suggest using an instrumental variables (IV) estimation method, where the lagged dependent and endogenous regressors are instrumented. Therefore, we apply the dynamic panel data estimator suggested by Arellano and Bond (1991). They prove that Generalized Method of Moments (GMM) estimation provides consistent parameter estimates by utilizing instruments that can be obtained from orthogonality conditions that exist between the lagged values of the variables and the disturbances. Specifically, Equation 5 is estimated in first differences using GMM, whereby the levels of all right-hand side variables at the second lag are used as instruments. 12 Using instrumental variables also accounts for the problem that delays may arise between the decision to change the capital structure and the actual execution. Finally, it must be noted that Hovakimian et al. (2001), among others, estimate the target debt ratio in a first step and the speed of adjustment in a second step using the fitted values from the first step. This imposes an errors-invariable problem. In contrast, our approach allows the and coefficients in Equation 5 to be estimated simultaneously.
Several specifications concerning the endogeneity of explanatory variables are tested, but we only report the results of the model that assumes that all explanatory variables are endogenous. As suggested by Arellano and Bond (1991), their one-step GMM estimator is used for inference on the coefficients. All coefficients are adjusted for heteroscedasticity. In addition, to make sure that the target debt ratio is properly specified, a Wald test statistic is reported for the null hypothesis that all coefficients on the determinants of the target debt ratio are jointly equal to zero. Arellano and Bond (1991) further show that the coefficient estimates are only consistent if there is no second order serial correlation in the differenced residuals. Therefore, we report a test-statistic (z 2 ) for the null hypothesis of no second order serial correlation in the residuals. Because this restriction is violated in most specifications, Equation 5 is 9 However, there is one caveat. In the estimations we observe that the time-specific effects, d t , absorb most of the explanatory power of the macroeconomic determinants of adjustment speed. Therefore, when Z it denotes a macroeconomic variable (see subsection C), Equation 5 is estimated without time-specific effects. 10 Finally, it must be noted that Z it not only impacts the adjustment speed, but also the time varying target debt ratio, as can be inferred from the second summation term in Equation 5. However, these dynamics are not of interest and do not further comment on this issue. 11 An alternative to first-difference transformation is the within-transformation that is commonly used in the literature. Although this approach controls for fixed effects, it introduces correlation between the lagged dependent variables and the lagged error term, leading to biased estimates. The magnitude of this bias falls with the number of years in the sample (see Nickel, 1981). However, only ten years of observations are used here and, hence, this problem will not vanish. 12 Using first differences removes possible firm-specific effects by avoiding any correlation between unobservable firm-specific characteristics and regressor variables. See Verbeek (2004) for a textbook treatment. estimated by including the second lag of leverage, LV itÀ2 , as an additional explanatory variable. Note that the presence of this additional variable accomplishes a mere statistical requirement (i.e., to guarantee consistent parameter estimates), but a deeper economic interpretation cannot be provided. Hence, the second lag of leverage, LV itÀ2 , is not modelled in the same way as the first lag, LV itÀ1 , and we omit reporting the corresponding coefficient estimates.
Again following the recommendation by Arellano and Bond (1991), their two-step GMM estimator for inference about model specification is adopted. With respect to the validity of the instruments, a Sargan (1958) test for the null hypothesis is conducted that the overidentifying restrictions are valid. As already mentioned above, the second lag of all (endogenous and exogenous) variables (in levels) is used as instruments, and the Sargan test indicates whether these instruments are independent from the residuals. To assess the stability of the system (i.e., to guarantee convergence to a target), we check whether the test statistic defined as the estimated coefficient of the lagged dependent variable, LV itÀ1 , minus the estimate of 1 times the mean of Z it falls into the interval [0, 1]. This requirement is fulfilled in all model specifications.

B. Determinants of the capital structure
According to Harris and Raviv (1991), the consensus is that 'leverage increases with fixed assets, non-debt tax shields, investment opportunities, and firm size and decreases with volatility, advertising expenditure, the probability of bankruptcy, profitability, and uniqueness of the product.' 13 In the empirical analysis we focus on four of these variables: tangibility of assets (the ratio of fixed to total assets; TANG), firm size (SIZE), the market-to-book ratio (as a proxy for investment opportunities; GROWTH) and profitability (measured as the return on assets; ROA). These determinant variables are used to provide an estimate of each firm's target leverage ratio. In this section we provide a brief explanation for the choice of these variables in the empirical analysis. Titman and Wessels (1988), Rajan and Zingales (1995), and Fama and French (2002) argue that the ratio of fixed to total assets is an important determinant of leverage. However, the direction of influence is not ex ante clear. On the one hand, alleviating the classical bondholder-shareholder conflict (e.g., Galai and Masulis, 1976;Jensen and Meckling, 1976), with more tangible assets the creditors have an improved guarantee of repayment. Even in the worst state, firm assets retain more value in liquidation. Accordingly, the trade-off theory predicts a positive relationship between measures of leverage and the proportion of tangible assets.

Tangibility (TANG): Previous empirical studies by
On the other hand, managers of highly levered firms will be less able to consume excessive perquisites, since bondholders more closely monitor such firms (e.g., Grossman and Hart, 1982). In general, the monitoring costs will be higher for firms with less collateralizable assets, i.e., firms with less collateralizable assets may voluntarily choose higher debt levels to limit consumption of perquisites. This notion implies a negative relationship between asset tangibility and leverage.

Firm size (SIZE):
The effect of firm size on leverage is also ambiguous. On the one hand, Warner (1977) and Ang et al. (1982) document that bankruptcy costs are relatively higher for smaller firms. Similarly, Titman and Wessels (1988) argue that larger firms tend to be more diversified and fail less often. Accordingly, the trade-off theory predicts an inverse relationship between size and the probability of bankruptcy, i.e., a positive relationship between size and leverage. If diversification goes along with more stable cash flows, this prediction is also consistent with the free cash flow theory by Jensen (1986) and Easterbrook (1984). This notion implies that size has a positive impact on the supply of debt.
Alternatively, size can be regarded as a proxy for information asymmetry between firm insiders and the capital markets. Large firms are more closely observed by a large number of analysts and should be more capable of issuing informationally more sensitive equity. This leads to lower debt levels for large firms. Accordingly, the pecking order theory of the capital structure predicts a negative relationship between leverage and size, with larger firms exhibiting increasing preference for equity relative to debt.
Growth opportunities (GROWTH): It is generally acknowledged that the costs from issuing debt and the associated shareholder-bondholder conflicts are higher for firms with substantial growth opportunities. Therefore, the trade-off model predicts that firms with more investment opportunities carry less leverage, because they have stronger incentives to signal that they do not engage in underinvestment and asset substitution. This notion is strengthened by Jensen's (1986) free cash flow theory, which predicts that firms with more investment opportunities have less need 13 See Harris and Raviv (1991 p. 335). for the disciplining effect of debt payments to control free cash flows. 14 Previous empirical results are mixed. For example, Titman and Wessels (1988) find a negative relationship, while Rajan and Zingales (1995) report a positive relationship between leverage and growth opportunities. 15 In fact, the simple version of the pecking order theory supports the latter result. Debt typically grows when investment exceeds retained earnings and falls when investment is less than retained earnings. Therefore, given profitability, book leverage is predicted to be higher for firms with more investment opportunities. However, in a more complex view of the model, firms are concerned with future as well as current financing costs. Balancing current and future costs, it is possible that firms with large expected growth opportunities maintain low-risk debt capacity to avoid financing future investments with new equity offerings or even foregoing these investments. Therefore, the more complex version of the pecking order theory predicts that firms with larger expected investments have less current leverage.

Profitability (ROA):
In the trade-off theory, agency costs, taxes, and bankruptcy costs push more profitable firms towards higher book leverage. First, expected bankruptcy costs decline when profitability increases. Second, the deductibility of corporate interest payments induces more profitable firms to finance with debt. Finally, in the agency models of Jensen and Meckling (1976), Easterbrook (1984), and Jensen (1986), higher leverage helps to control agency problems by forcing managers to pay out more of a firm's excess cash. The strong commitment to pay out a larger fraction of pre-interest earnings to creditors suggests a positive relationship between book leverage and profitability. This notion is also consistent with Ross' (1977) signalling hypothesis, where higher levels of debt can be used by managers to signal an optimistic future for the firm.
In contrast, according to the pecking order model higher earnings should result in less book leverage. Firms prefer raising capital, first from retained earnings, second from debt, and third from issuing new equity. This behaviour is due to the costs associated with new equity issues in the presence of information asymmetries. Debt grows when investment exceeds retained earnings and falls when investment is less than retained earnings. Hence, the pecking order model predicts a negative relationship between book leverage and profitability. 16 Again, previous empirical evidence is mixed. Rajan and Zingales (1995) report a negative relationship between leverage and profitability (supporting the pecking order theory), while Jensen et al. (1992) find a positive one (supporting the trade-off theory).
C. Determinants of the speed of adjustment to the target capital structure Firm-specific factors: It is assumed that the speed of adjustment towards the target capital structure, denoted as it , depends on three firm-specific factors. Two of these determinant variables also affect the target debt level (GROWTH and SIZE). The third variable measures the distance between observed leverage and target leverage (DIST). All three variables can be interpreted by referring to the notion of weighting the costs of changing the capital structure against the costs associated with a particular leverage level.
Distance between observed and target leverage (DIST): If fixed costs (e.g., legal fees and investment bank fees) constitute a major portion of the total cost of changing capital structure, firms with sub-optimal leverage will change their capital structure only if they are sufficiently far away from the target capital structure. Accordingly, the likelihood of adjustment is a positive function of the absolute difference between target leverage and observed leverage. This variable is defined as where LV Ã it is the fitted value from the fixed-effect regression of the debt ratio of firm i on the capital structure determinants as of time t.
If the fixed costs of adjustments are prohibitively high, firms will avoid approaching the capital market and change dividend policy to adjust towards the target leverage. Intuitively, the costs of sub-optimal dividend policy are increasing with the magnitude of the absolute difference between target leverage and 14 Recently, Fama and French (2002) show how the predictions for book leverage carry over to market leverage. The trade-off theory predicts a negative relationship between leverage and investment opportunities. Since the market value grows at least in proportion with investment outlays, the relation between growth opportunities and market leverage is also negative. 15 These conflicting results may be due to the fact that growth measures tend to be correlated with tangibility. 16 Another question is again whether these predictions for book leverage carry over to market leverage (e.g., Fama and French (2002)). The trade-off theory predicts that leverage increases with profitability. Since the market value also increases with profitability, this positive relation does not necessarily apply for market leverage. In contrast, the pecking order theory predicts that firms with a lot of profits and few investments have little debt. Since the market value increases with profitability, the negative relationship between book leverage and profitability also holds for market leverage. observed leverage. Hence, if firms adjust internally rather than using outside financing, there should be a negative relationship between DIST and the speed of adjustment. Sorting out between the two theories is an empirical matter.
Growth opportunity (GROWTH): Growing firms may find it easier to change their capital structure by choosing among several alternative sources of financing. A no-growth firm can only change its capital structure by swapping debt against equity, or vice versa, which may induce negative signalling effects in the presence of asymmetric information and decrease firm value. In contrast, a growing firm can more easily change its capital structure by altering the composition of new issues accordingly. Even under asymmetric information, firm value may remain unchanged because of the positive effect of future growth opportunities. Accordingly, a positive relationship is expected between GROWTH and adjustment speed.
Firm size (SIZE): If changing the capital structure involves substantial fixed costs, these costs might be relatively smaller for large firms, and therefore they should more readily be able to correct deviations from the target capital structure. In addition, due to better analyst coverage more information is publicly available about large firms, implying better access to both debt and equity as well as lower anticipated costs arising from asymmetric information upon announcement. Hence, a positive relationship is expected between SIZE and the speed of adjustment.
Macroeconomic factors: In addition to the firmspecific factors introduced above, Banjeree et al. (2000) and Lo¨o¨f (2003) argue that economy-wide factors should have an impact on the speed of adjustment. They include time-specific effects to capture these factors in a simplistic way. However, these time-specific effects are hard to interpret, and therefore we employ a set of macroeconomic variables in the empirical analysis and measure their effect on adjustment speed. 17 Specifically the hypothesis proposed by Hackbarth et al. (2005) that the speed of adjustment depends on the stage of the business cycle is examined. They argue that the speed is higher in booms than in recessions. Popular business cycle variables are used, i.e., variables that are usually considered to be related to the current and/or future state of the economy, to model time-variation in the target adjustment coefficient. The following four macroeconomic factors are assumed to have an impact on the speed of adjustment: the term spread (TERM), the shortterm interest rate (ISHORT), the default spread (DEF), and the TED spread (TED).
Term spread (TERM) and short-term interest rate (ISHORT): The slope of the term structure of interest rates (TERM) is generally assumed to be a predictor of future business cycle stages. It is widely acknowledged in the literature that a high (low) term spread can be interpreted as an indicator of good (bad) economic prospects (e.g., Estrella and Hardouvelis, 1991;Harvey, 1991). Consumption smoothing drives the demand for insurance or hedging, and a natural way is to substitute bonds of different maturities. If the economy is in a growth stage, but a general slowdown is expected, investors will hedge by buying assets that deliver payoffs during the future economic downturn. For example, they could purchase long-term government bonds and simultaneously sell short-term bonds for hedging purposes. If many investors follow, the price of longterm bonds increases, implying decreasing yields. In contrast, the selling pressure for short-term bonds will drive down prices and increase yields. As a result, the term structure flattens or even becomes inverted. Chen (1991) also documents that an above average term spread forecasts that the gross natural product will continue to increase over the next four to six quarters. Following the predictions in Hackbarth et al. (2005), faster adjustment in booms than in recessions is expeced, and therefore the coefficient on the interaction term between lagged leverage and TERM in Equation 5 should be positive. In a similar vein we hypothesize a negative relationship between ISHORT and adjustment speed.
There is increasing empirical evidence that managers attempt to time both equity and debt issuances. For example, the results in Baker and Wurgler (2002) suggest that the observed capital structure is a cumulative outcome of past attempts to time the equity market. An upward sloping term structure combined with generally low interest rates are usually interpreted as indicators of economic expansion. High expected real growth implies rising stock market valuations and, according to the market timing hypothesis, more equity financing activities in an attempt to time exploit 'windows of opportunities'. There is similar evidence for debt issuances.
For example, Henderson et al. (2004) find a negative relationship between the level of interest rates and the quantity of debt issued. 18 Both long-term and shortterm debt issues are negatively related to their respective interest rates. Therefore, firms appear to time their debt issues for both long-term and shortterm bonds. These empirical results are supported by survey evidence in Graham and Harvey (2001) and Drobetz et al. (2006), where managers claim that they attempt to issue debt at times of low interest rates. 19 Firms attempt to time market interest rates and issue short-term debt when they feel that short rates are low relative to long rates. Henderson et al. (2004) further document that debt issues increase when interest rates are low mainly because firms have larger capital demands, and the substitution effect of debt for equity is of secondary importance. This is an important observation in the present context, because we do not test predictions about the amount of leverage over the business cycle, but rather the speed of adjustment to the target debt ratio.
Finally, better prospects for the real activity should lead to increasing cash flows from operations. Even if firms do not approach the capital market to raise external funds, increasing profits from operations enable them to adjust internally by changing their payout policy. All these arguments strengthen the main hypothesis that there is a positive relationship between TERM and adjustment speed, and a negative relationship between ISHORT and the speed of convergence towards the target debt ratio.
In the empirical analysis, the three-month Eurodollar deposit rate for Swiss francs is used as the short-term interest rate. The term spread is constructed as the difference between the yield on long-term Swiss government bonds (with maturities of more than five years) and the three-month Eurodollar interest rate.

Default spread (DEF) and TED spread (TED):
The default spread (DEF) is calculated as the difference between the yield on US low-grade (BAA) and highgrade (AAA) corporate bonds with the same maturity. It is assumed that this variable is a legitimate proxy for global default risk. Specifically, it can be taken as an indicator of the current health of the economy. Similarly, the TED spread (TED), defined as the difference between the three-month Eurodollar rate and the 90-day yield on the US Treasury bill, can be viewed as a 'political' risk premium that reflects either actual or anticipated barriers to international investing (e.g., Ferson and Harvey, 1993). The yield differential widens when the risk of disruption in the global financial system increases. Following the general notion in Hackbarth et al. (2005) that the speed of adjustment is higher in good states than in bad states of nature, a negative relationship is expected between the adjustment speed and the size of both the default spread and the TED spread. However, in light of the results in Baker et al. (2003), we suspect that these relationships are weaker compared to the relationship between the speed of adjustment and the term spread as well as the level of interest rates. In contrast to the latter two determinant variables, they document that both the default spread and the TED spread are unrelated to longterm debt as a fraction of total debt issues as well as future bond returns.

III. Data
In general, our sample targets all 253 firms in the Swiss Performance Index (SPI). However, several adjustments are necessary. First, the SPI consists of a great number of financial institutions. Because banks and insurances are subject to specific rules and regulations according to Swiss law, their leverage is severely affected by exogenous factors. Following Rajan and Zingales (1995), all firms categorized as 'Financials' according to the sector classification of Swiss Exchange (SWX) are excluded and we focus exclusively on non-financial firms. Second, it was not possible to collect the necessary data for many of the smaller firms in the SPI. These adjustments leave an unbalanced panel of 90 firms over the 1991-2001 period. 20 All data is taken from the Worldscope database.
Apart from the many competing theories of capital structure, there is not even a clear-cut definition of 'leverage' in the academic literature. The specific 18 See Baker et al. (2003) for similar evidence. They document that firms borrow long when debt market conditions suggest that the relative cost of long-term debt is low. 19 The evidence is not conclusive yet as to whether managers are in fact successful timers. Baker et al. (2003) report that firms tend to borrow long when excess bond returns are predictably low. Long-term debt issues predict lower excess bond returns, and short-term debt issues predict higher excess bond returns. In contrast, Henderson et al. (2004) only find weak evidence that the aggregate level of the quantity of new debt issued predicts future changes in interest rates. 20 Drobetz and Fix (2005) look at a larger (balanced) panel of Swiss firm over the shorter period from 1997 to 2001. However, to give a better picture of the adjustment process, which is the main focus of the present analysis, a longer, albeit smaller, sample of firms is used. choice depends on the objective of the analysis, and Bevan and Danbolt (2002) document that the determinants of leverage vary significantly, depending upon which component of debt is being analysed. Following Rajan and Zingales (1995), two alternative definitions of leverage are applied. The first and broadest definition of leverage is the ratio of total (non-equity) liabilities to total assets, denoted as LVLTA. This measure can be viewed as a proxy of what is left for shareholders in case of liquidation. However it is not without problems. First, it does not provide a good indication of whether the firm is at risk of default in the near future. Second, since total liabilities also include items like accounts payable, which are used for transaction purposes rather than for financing, it is likely to overstate the amount of leverage. Finally, this measure of leverage is potentially affected by provisions and reserves, such as pension liabilities. 21 An alternative, and possibly more appropriate, definition of leverage is the ratio of interest bearing debt to capital, where capital is defined as total debt plus equity, denoted as LVDC. This measure of leverage looks at the capital employed and, therefore, best represents the effects of past financing decisions. It most directly relates to the agency problems associated with debt, as suggested by Jensen and Meckling (1976) and Myers (1977).
An additional issue is whether leverage should be computed as the ratio of the book or the market value of equity. Fama and French (2002) argue that most of the theoretical predictions apply to book leverage. Similarly, Thies and Klock (1992) suggest that book ratios better reflect management's target debt ratios. The market value of equity is dependent on a number of factors which are out of direct control for the firm. Therefore, using market values may not reflect the underlying alterations initiated by a firm's decision makers. In fact, corporate treasurers often explicitly claim to use book ratios to avoid distortions in their financial planning caused by the volatility of market prices. A similar rationale is often heard from rating agencies. From a more pragmatic point of view, the market value of debt is not readily available. Bowman (1980) documents a high correlation between market and book values of leverage. It should therefore come as no surprise that most previous literature relates to the book value of leverage. Nevertheless, we also report quasi-market leverage, where the book value of equity is replaced by the market value of equity, but debt is valued at its book value. Table 1 shows the data description for different definitions of leverage over the sample period from 1991 to 2001. Both the mean and median leverage ratios for each year are reported together with the cross-sectional standard deviation. There are three important observations. First, independent of the definition of leverage, book leverage declines. This might be explained by an attempt to increase the marginal debt capacity during the prosperous decade of the 1990s. Second, market leverage has increased recently. For example, the mean ratio of debt to capital increased from 24.80% to 31.81% between 2000 and 2001. Of course, this can be explained by the sharp decline in stock market capitalization, which strengthens the notion that market leverage is not directly under control of the firm. Finally, leverage ratios of Swiss firms are similar to the figures reported by Rajan and Zingales (1995) for US firms. 22 Panel A in Table 2 presents summary statistics for the determinants of the target capital structure. The exact definitions of the variables are as follows. First, TANG is the ratio of fixed assets to total assets. The more direct approach using intangible assets in the nominator cannot be applied due to a lack of data. Second, following Titman and Wessels (1988), SIZE is the natural logarithm of net sales. The logarithmic transformation accounts for the conjecture that small firms are particularly affected by a size effect. 23 Third, GROWTH is measured as the ratio of market-to-book equity. Simple cash flow valuation models suggest that this is a forward-looking measure. Unfortunately, we do not have research and development expenditures for most firms in the sample available. Alternatively, past growth rates of total assets could be used. However, this measure may not be appropriate because historical growth is not necessarily linked to future growth (e.g., Chan et al. (2003)). Finally, following Titman and Wessels (1988), the ratio of operating income over total assets, or return on assets (ROA), is used as our profitability measure. Panel B in Table 2 shows the summary statistics for the macroeconomic variables and the distance measure used as determinants of the speed of adjustment. Since the business cycle variables are well known and widely used in the literature, we omit a detailed discussion. 21 In Switzerland this should not be important because pension liabilities need not be expensed in the balance sheet. In contrast to most other continental European countries, pension money is managed in separated entities. 22 For a more detailed discussion and international comparison of Swiss data see Drobetz and Fix (2005). 23 Alternatively, total assets could be used, but possibly net sales is a better proxy for size, because many firms attempt to keep their reported size of asset as small as possible, e.g., by using lease contracts. Table 3 reports the correlation coefficients between the leverage variables (both LVLTA and LVDC) and all explanatory variables. The relationship between leverage and asset tangibility (TANG) as well as firm size (SIZE) is positive, and there is a negative relationship between leverage and growth opportunities (GROWTH) as well as profitability (ROA). These results confirm previous evidence (e.g., Rajan and Zingales, 1995;Panno, 2003) and indicate that the determinant variables are appropriate to model a time-varying target debt ratio. This proposition will be furthered using fixed effects panel regressions in Section IV. In addition, the table reveals that the correlation coefficients between macroeconomic determinants of the adjustment speed are very high. Most important, the correlation between TERM and ISHORT is À0.94, and the correlation between TERM and DEF is À0.77. To avoid multicollinearity problems when estimating Equation 5 in the presence of these high correlations, the determinants of the adjustment speed are used separately in the estimations one at a time.

Determinants of target leverage
The estimation of the dynamic model in Equation 5 crucially depends on the correct specification of the LVLTAB is the ratio of total (non-equity) liabilities to total assets, and LVDC is the ratio of total debt to capital, where capital is defined as total debt plus equity. For the market values of leverage the book value of equity is replaced by the market value of equity. All numbers are expressed in %. TANG is defined as the ratio of fixed assets to total assets, SIZE is the natural logarithm of net sales, GROWTH is the ratio of market-to-book equity, and ROA is the ratio of operating income over total assets. DIST is the difference between the target and the current debt ratio, where the target debt ratio is constructed as the fitted value of the fixed-effect regression of the debt ratio on the four capital structure determinants TANG, SIZE, GROWTH, and ROA. TERM is the term spread, defined as the difference between the yield on long-term Swiss government bonds (with maturities of more than five years) and the three-month Eurodollar interest rate, ISHORT is the three month Eurodollar deposit rate for Swiss francs, DEF is the difference between the yield on US low-grade (BAA) and high-grade (AAA) corporate bonds, and TED is the difference between the three-month Eurodollar rate for US dollars and the 90-day yield on US Treasury bills. target capital structure. Although the correlation analysis in Table 3 has already provided preliminary evidence, the basic specification of the target debt ratio in Equation 1 is further tested by running fixed effects regressions. Fixed effects regressions preserve the time series variation in leverage, but ignore most of the cross-sectional differences across firms. There is one caveat to mention, which is that leverage is sticky. A firm with higher-than-predicted leverage in one year is likely to have higher-than-predicted leverage in the next year. This stickiness in financial policy may lead to inflated t-statistics. Therefore, a dummy variable is added for each year to estimate a combined time fixed and entity fixed effects regression model. The additional dummies control for variables that are constant across entities (firms) but evolve over time. This combined time and firm fixed effects model eliminates an omitted variables bias arising both from unobserved variables that are constant over time and from unobserved variables that are constant across firms. Estimation results are displayed in Table 4. The results reveal that tangibility (TANG) is always positively correlated with leverage, and all coefficients are significant at the 5% level of significance. This supports the prediction of the trade-off theory that the debt-capacity increases with the proportion of tangible assets on the balance sheet. Size (SIZE) is also positively related to leverage, indicating that size is a proxy for a low probability of default, as suggested by the trade-off theory. The estimated coefficients are again significant at the 5% level. For Germany, where firms tend to be liquidated more easily than in Anglo-Saxon countries, Rajan and Zingales (1995) report that large firms have substantially less debt than small firms. Given that Swiss company law is very similar to the German regulation, our results for Switzerland are interpreted as size being a proxy for low expected costs of financial distress, and where small firms are especially wary of debt. Another result is that firms with high market-to-book ratios (GROWTH) tend to have lower leverage than firms with low marketto-book ratios. This result is consistent with both the trade-off theory and an extended version of the pecking order theory (e.g., Fama and French, 2002). However, the relationship is only significant for market values of leverage. Finally, profitability  TANG  SIZE  GROWTH ROA  TERM  ISHORT DEF  TED  DIST LVLTA Book À0.01 0.27*** À0.26*** À0.16*** À0.05* 0.08** À0.01 0.07** 0.09*** LVLTA Market 0.24*** 0.10*** À0.62*** À0.33*** À0.18*** 0.25*** 0.06* À0.24*** 0.10*** LCDC Book 0.18*** 0.07** À0.22*** À0.15*** À0.06* 0.09*** À0.01 0.09** 0.03 LVDC Market 0.38*** À0.03 À0.46*** À0.27*** À0.20*** 0.25*** 0.07** À0.24*** 0.24*** TANG 1.00 SIZE À0.30*** 1.00 GROWTH À0.30*** À0.08** 1.00 ROA À0.03 0.20*** 0.28*** 1.00 TERM 0.02 0.03 0.14*** 0.07* 1.00 ISHORT 0.01 À0.05 À0.19*** À0.06* À0.94*** 1.00 DEF À0.06* À0.01 À0.03 À0.05* À0.77*** 0.61*** 1.00 TED À0.04 0.04 0.23*** 0.10*** 0.46*** À0.61*** À0.06* 1.00 DIST LVLTA_Book 0.01 À0.25*** À0.08** À0.18*** À0.19*** 0.24*** 0.01 À0.20*** 1.00 DIST LVLTA_Market À0.01 À0.22*** À0.13*** À0.35*** À0.21*** 0.24*** 0.05 À0.21*** 1.00 DIST LVDC_Book À0.01 À0.21*** À0.10*** À0.12*** À0.14*** 0.20*** À0.02 À0.19*** 1.00 DIST LVLDC_Market 0.20*** À0.28*** À0.23*** À0.35*** À0.22*** 0.27*** 0.01 À0.24*** 1.00 Notes: The table reports the correlation coefficients between the leverage variables and all explanatory variables. LVLTAB is the ratio of total (non-equity) liabilities to total assets, and LVDC is the ratio of total debt to capital, where capital is defined as total debt plus equity. For the market values of leverage the book value of equity is replaced by the market value of equity. TANG is defined as the ratio of fixed assets to total assets, SIZE is the natural logarithm of net sales, GROWTH is the ratio of market-to-book equity, and ROA is the ratio of operating income over total assets. DIST is the difference between the target and the current debt ratio, where the target debt ratio is constructed as fitted values from a fixed effects regression of the debt ratio on the four capital structure determinants TANG, SIZE, GROWTH, and ROA. TERM is the term spread, defined as the difference between the yield on long-term Swiss government bonds (with maturities of more than five years) and the threemonth Eurodollar interest rate, ISHORT is the three-month Eurodollar deposit rate for Swiss francs, DEF is the difference between the yield on US low-grade (BAA) and high-grade (AAA) corporate bonds, and TED is the difference between the three-month Eurodollar rate for US dollars and the 90-day yield on US Treasury bills. Coefficients of correlation that are significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and *, respectively.
(PROF) is negatively correlated with leverage, both for book and market leverage. This result supports the predictions of the pecking order theory. All coefficients are statistically significant. A Wald test rejects the null hypothesis that all explanatory variables are simultaneously equal to zero. In addition, the test statistic of a Hausman (1978) test is reported, where the null hypothesis is that the fixed effects estimator and random effects estimator are equivalent. As stated at the bottom of Table 4, the Hausman test statistic rejects this null hypothesis, which is usually interpreted in favour of the fixed effects estimator. Overall, these preliminary results are comparable to those in Drobetz and Fix (2005), and they indicate that the capital structure variables are appropriate to model the timevarying target debt ratio in a dynamic adjustment model. These results strongly confirm the preliminary evidence from our (univariate) correlation analysis in Table 3. This is crucial, because most of the information contained in our panel data set is crosssectional. However, this cross-sectional information is ignored when introducing firm specific effects, as only deviations from means (or differences in time) remain.

Determinants of adjustment speed
In this section we report the dynamic panel estimation results from Equation 5, which compounds the main hypotheses regarding time-variation in adjustment speed. Dynamic panel estimation using GMM (see Arellano and Bond, 1991) allows estimation of all coefficients in Equation 5 simultaneously. We do not use the fitted values from estimating the target debt ratio in Equation 1 and Table 4, but rather estimate all and parameters in one single step. The main focus lies on the estimate for 1 , which is the coefficient on the interaction term between a determinant variable of adjustment speed, Z it , and lagged leverage, LV itÀ1 . As explained in subsection IIA above, the determinant variables of adjustment speed are used in the estimations only separately one at a time to keep the estimation problem tractable and to avoid multicollinearity. Pairwise correlation analysis in Table 3 indicates that the latter problem is particularly important for the macroeconomic variables. Therefore, for each dynamic panel regression specification in Tables 5 and 6, we test the null hypothesis that 1 ¼ 0, i.e., the speed of adjustment is constant and independent from a particular firm Notes: The table reports the results from fixed effects panel regressions of the leverage ratio on firm-specific capital structure determinants. LVLTAB is the ratio of total (non-equity) liabilities to total assets, and LVDC is the ratio of total debt to capital, where capital is defined as total debt plus equity. For the market values of leverage the book value of equity is replaced by the market value of equity. The capital structure determinants are as follows: TANG is defined as the ratio of fixed assets to total assets, SIZE is the natural logarithm of net sales, GROWTH is the ratio of market-to-book-equity, and ROA is the ratio of operating income over total assets. Firm-specific and time-specific effects are included. Coefficients that are significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and *, respectively. Robust standard errors are in brackets. Numbers in brackets for the Wald test and the Hausman test denote the degrees of freedom.
characteristic or a business cycle variable. In addition to the specification tests laid out in subsection IIA, the coefficient on this interaction term, 1 , together with the coefficient on the lagged debt ratio, denoted as (1À 0 ) are reported. Note that Equation 5 specifies a negative sign on 1 , and therefore the signs of the estimated coefficients on the respective interaction terms must be interpreted accordingly. Table 5 summarizes the impact of firm-specific factors on the speed of adjustment. Most important, in contrast to Lo¨o¨f (2003) and Banerjee et al. (2000), a statistically weak positive relationship is revealed between the speed of adjustment and the distance variable (DIST). This result lends support to the hypothesis that the fixed costs of adjustment are significant, and firms with sub-optimal leverage will change their capital structure only if they are sufficiently far away from the target debt ratio.
Second, the estimated coefficient on the interaction term with growth opportunities (GROWTH) indicates that firms with higher growth opportunities adjust faster towards their target leverage. This result confirms the hypothesis that a growing firm may find it easier to change its capital structure by altering the composition of new issuances. Finally, the results for the impact of firm size (SIZE) on the adjustment speed are mixed and do not allow further interpretations. Specifically, the finding in Banerjee et al. (2000) and Lo¨o¨f (2003) that larger firms are more concerned about capital structure decisions than smaller firms cannot be confirmed. Table 6 contains the results for the impact of the macroeconomic variables on adjustment speed. Consistent with our hypotheses, the estimated coefficients on the interaction terms related to TERM and ISHORT are positive and negative, respectively,  Arellano and Bond (1991). Variations in sample size are due to data limitations. LVLTAB is the ratio of total (non-equity) liabilities to total assets, and LVDC is the ratio of total debt to capital, where capital is defined as total debt plus equity. For the market values of leverage the book value of equity is replaced by the market value of equity. The determinants of the speed of adjustment are as follows: SIZE it is the natural logarithm of net sales; GROWTH it is the ratio of market-to-book equity; and DIST it is constructed as the fitted values from a fixed-effect regression of the debt ratio on the four capital structure determinants TANG it , SIZE it , GROWTH it , and ROA it . The table shows the coefficients on the lagged leverage ratio and on the interaction term of the determinant of adjustment speed with the lagged debt ratio. Coefficients that are significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and *, respectively. Robust standard errors are in brackets. The Wald test statistic refers to the null hypothesis that all coefficients on the determinants of the target debt ratio are jointly equal to zero. The test statistic z 2 tests the null hypothesis of no second order correlation in the residuals. The Sargan test statistic for the null hypothesis that the overidentifying restrictions are valid uses the Arellano-Bond two-step estimator.
albeit not always statistically significant. Because a large term spread and a low short-term interest rate indicate that economic prospects are good, this result confirms the notion proposed by Hackbarth et al. (2005) that the speed of adjustment is higher in booms than in recessions. In contrast, the positive coefficients on the interactions terms related to DEF and TED contrast with our exante intuition and are harder to interpret. An alternative hypothesis that is compatible with this result is that that firms are forced to correct deviations from the target capital structure more readily in times of high uncertainty, i.e., they cannot afford to stay off the optimum. It is also noteworthy that in most cases a Wald test rejects the null hypothesis that the coefficients on the determinants of the target debt ratio are jointly zero, indicating that the target is properly specified.
To further explore this issue, we refer to the observation in Korajczyk and Levy (2003) that their results differ for subsamples of financially constrained and unconstrained firms. Given that a firm's access to financial markets is expected to  Arellano and Bond (1991). Variations in sample size are due to data limitations. LVLTAB is the ratio of total (non-equity) liabilities to total assets, and LVDC is the ratio of total debt to capital, where capital is defined as total debt plus equity. For the market values of leverage the book value of equity is replaced by the market value of equity. The determinants of the speed of adjustment are as follows: TERM t is the term spread, defined as the difference between the yield on long-term Swiss government bonds (with maturities of more than five years) and the three-month Eurodollar interest rate, ISHORT t is the three-month Eurodollar deposit rate for Swiss francs, DEF t is the difference between the yield on US low-grade (BAA) and high-grade (AAA) corporate bonds, and TED t is the difference between the three-month Eurodollar rate for US dollars and the 90-day yield on US Treasury bills. The table shows the coefficients on the lagged leverage ratio and on the interaction term of the determinant of adjustment speed with the lagged debt ratio. Coefficients that are significantly different from zero at the 1%, 5%, and 10% level are marked with ***, **, and *, respectively. Robust standard errors are in brackets. The Wald test statistic refers to the null hypothesis that all coefficients on the determinants of the target debt ratio are jointly equal to zero. The test statistic z 2 tests the null hypothesis of no second order correlation in the residuals. The Sargan test statistic for the null hypothesis that the overidentifying restrictions are valid uses the Arellano-Bond two-step estimator. affect its capital structure choice, and financial constraints clearly have a macroeconomic dimension, the sample is split into two categories, depending on whether a firm is financially constrained or unconstrained. 24 Financially constrained firms cannot postpone adjustment in either state of high or low uncertainty and, hence, they cannot time issues. Accordingly, there should not be any relationship between adjustment speed and both the default spread and the TED spread. At least, the sensitivity of the adjustment speed to the business cycle variables should be higher for financially unconstrained firms. Korajczyk and Levy (2003) define a firm as financially constrained if it does not have sufficient cash to undertake investment opportunities and if it faces severe agency costs when accessing financial markets. We experimented with different criteria to classify a firm as financially constrained or unconstrained. In a first approach, a firm's retention rate together with the existence of investment opportunities is used: a firm-event window is considered as financially constrained if in a given year a firm has (i) a dividend yield of zero and (ii) a Tobin's Q greater than one. All firm-events that are considered as financially constrained constitute the first subsample, and all other firm-events fall into the financially unconstrained subsample. Accordingly, each firm can be financially constrained in one year, but unconstrained in other years. This approach is similar to Korajczyk and Levy (2003). Second, instead of using both the dividend yield and Tobin's Q, we attempt to define a firm-event window to be financially constrained if only the dividend yield is zero. Third, instead of looking at yearly events, we follow Fazzari et al. (1988) and define a firm as financially constrained if its dividend yield is zero during five (not necessarily consecutive) years over the 1991-2001 sample period. As a stability test, we also look at different numbers of years with zero dividend payments. Finally, an attempt is made to combine the dividend yield criteria with the requirement of having a Tobin's Q larger than one, on average, over the sample period. Note that the last two sample split-ups classify firms as either financially constrained or unconstrained over the entire sample period.
Overall, we feel that the empirical results for the subsample tests are unsatisfactory and hardly allow meaningful interpretations. Detailed results are omitted, because we cannot detect any systematic differences in the magnitude of the coefficients on the interaction term between constrained and unconstrained firms. 25 In addition, the GMM estimates are inconsistent for several model specifications, as indicated by significant z 2 test statistics for the null hypothesis of no second order serial correlation in the residuals. The limited size of our panel data set may be one explanation why it is not possible to find evidence in this direction, but more future research is clearly necessary.

V. Conclusions
Capital structure is a key issue for financial decision makers. Empirical evidence as well as evidence from surveys indicates that firms seek a target debt-equity ratio. The dependence of firms' leverage ratios on well-known firm characteristics has usually been interpreted in favour of one or the other standard capital structure models, e.g., the trade-off theory or the pecking order theory. However, these models remain silent about the nature of the adjustment process towards the target debt ratio. Because of random events or other changes, firms may temporarily deviate from their target capital structure and then only gradually work back to the optimum. In fact, in the presence of adjustment costs, it might be cheaper for firms not to fully adjust to their targets even if they recognize that their existing leverage ratios are not optimal. Nevertheless, there is surprisingly little empirical evidence on the determinants of a time-varying adjustment speed and about the influence of macroeconomic variables on the adjustment process, in particular.
We present a simple model that endogenizes both the target leverage ratio and the speed of adjustment. Using a dynamic adjustment model and panel methodology for a sample of 90 Swiss firms over the 1991-2001 period, it is possible to shed new light (i) on the determinants of the target capital structure rather than the observed capital structure and (ii) on the determinants of the adjustment speed. Most important, the effects of firm-specific characteristics as well as macroeconomic factors on the speed of adjustment to the target leverage are analysed. We document that faster growing firms and those that are further away from their target capital structure adjust more readily. The results also reveal interesting interrelations between the adjustment speed and well-known business cycle variables. Most important, 24 Note that a sample split-up allows that all the coefficients in the model are different between the two subsamples, whereas a dummy variable approach usually only allows that selected coefficients differ between the subsamples. 25 Detailed results are available upon requests from the authors. the speed of adjustment is higher when the term spread is higher and when economic prospects are good.
However, our work has clear limitations. Possibly due to the small sample size, it is not possible to identify differences in the speed of adjustment between financially constrained and unconstrained firms. Given that a firm's access to financial markets is expected to affect its capital structure choice, and financial constraints clearly have a macroeconomic dimension, this remains an interesting open research question.