Strategic Interaction of Capital Structures: A Spatial Econometric Approach Zhengyu Zhang Center for Econometric Study Shanghai Academy of Social Sciences [email protected] This Version: December 2011 Abstract In this article, we suggest an alternative setting for empirically examining firms’ strategic interaction in choosing their capital structure. Following Lyandres (2006)’s theoretical model, this article explicitly focuses on how the competitive interaction in output market may induce a firm to take the rival firms’ capital structure into account in deciding its own capital structure. It is also shown that the direction of such strategic response depends on whether the output market competition is in strategic substitutes or in strategic complements. A spatial regression model is introduced to test the relationship between firms’ financial choices and their product market strategies. The empirical evidence suggests that inclusion of the spatially lagged term of a firm’s leverage could be empirically significant in explaining the optimal choice of a firm’s financial structure. Key words: Capital structure; Output market competition; Spatial regression; JEL: C21; G32; L13; 1. Introduction A relatively new literature, dating back to the seminal work of Brander and Lewis (1986), is examining the effects of the interaction among firms in output market on their financial and operating decisions. Generally, this strand of research argues that debt allows the firm to commit to a certain strategy to its rivals in the output market. Almost all models in these studies can be boiled down to two main theories. One theory, developed by Brander and Lewis (1986) and Maksimovic (1988), establishes that due to the effect of limited liability, the leverage will cause firm to behave aggressively, which makes competition tougher. The other class of predatory model, represented by Bolton and Scharfstein (1990), argues that where leveraged firms decrease output, the unleveraged rival has the incentive to increase output or cut prices to drive the leveraged firm out of market and as a result, debt commits leveraged firms to behave less aggressively, which makes competition softer. One undesirable feature of the limited liability effect models is that the theoretical predictions they derive may heavily rely on the specific form of product market competition. For example, the models that are based on the choice of debt in the first stage as a commitment to the subsequent product market strategy in the second stage, have traditionally been developed in a Cournot type quantity competition setting (Brander 1 and Lewis, 1986; Maksimovic, 1988; Glazer, 1994). In this competition setting, debt commits firms to a choice of higher quantity than in the absence of debt benefits. However, these results could change dramatically if alternative frameworks of oligopolistic competition are adopted. Showalter (1995) finds that the optimal strategic debt choice of Bertrand competitors depends on the type of uncertainty of the output market. When costs are uncertain, price competing firms will not use debt and the use of debt is strategically advantageous only if demand conditions are uncertain. Interestingly, Schuhmacher (2002) comes up with the opposite results. He shows that the capacity-price competitors, by committing to capacity levels prior to setting prices, will use no debt when there is demand uncertainty and will use high debt levels when there is cost uncertainty. As an attempt to mitigate the weakness that the derived results rely on certain competition type, Lyandres (2006) develops a model where firms are not necessarily required to compete against each other by quantity or by price. Lyandres shows that, regardless of whether firms’ product market choices are strategic substitutes or strategic complements, 1the stronger the extent of competitive interaction among rival firms in output market, the larger the strategic benefit of debt. Although the theoretical link between firms’ financial choices and product market strategies has been investigated by numerous aforementioned models, none of them explicitly answers the following question: as a strategic response to the rival firms increasing leverage to commit a more aggressive market strategy, will a firm have the incentive to raise or reduce its own debt accordingly? If so, how the direction and magnitude of such strategic adjustment is determined? This paper extends the existing literature by theoretically and empirically examining how the competition among firms in output market may induce strategic interaction among these firms’ financial structures. Specifically, the current study contributes to the exiting literature from the two dimensions that follow: Theoretically, by revisiting Lyandres (2006)’s model, we explicitly show how the competitive interaction in output markets may induce a firm to take its rival firms’ capital structures into calculus in deciding its own one. Although the basic setting of our theoretical analysis looks similar to Lyandres’, it should be pointed out that our major interest and the consequent insight focus on the strategic financial decision-making among competing firms instead of the relationship between the extent of competition and the optimal leverage. Most research on determination of corporate capital structure mainly builds on traditional hypotheses, such as tradeoff or pecking order theory, which largely ignore interactions among firms' financial policies. However, as it has been well recognized by the literature that interactions in the product markets can generate interactions among financial policies, it might be expected that firms do not make financing decisions in isolation of one another. Being interested in off-equilibrium According to Bulow et al. (1985), strategic substitutes and strategic complements are defined by whether a firm’s more aggressive strategy lowers or raises its rivals’ marginal profits. 1 2 outcome by focusing on firms' best response in their financial decision making instead of equilibrium itself, our analysis provides an initial and intuitive bridge between the corporate finance of capital structure and the industrial organization structure in which the firm operates. Consequently, such off-equilibrium analysis contributes to further understanding the factors that could be responsible for determining a firm's leverage ratio. Recall that in Lyandres (2006), the debt level is shown to be positively correlated to the extent of competition regardless of whether firms’ product market choices are in strategic substitutes or complements. In other words, the explicit role played by the firms’ product strategy type remains silent there. By contrast, the main theoretical insight here is that the direction of such strategic leverage adjustment depends on whether the output market competition is in strategic substitutes or in strategic complements. Furthermore, it can be shown that the magnitude of such adjustment monotonically relies on the extent of competition in output market. Empirically, we relate the theoretical predictions to a spatial autoregressive model and show that the resulting empirical implications can be tested within this spatial econometric framework. We run a series of spatial panel data model regressions to estimate the response slope of a firm’s leverage ratio with respect to its rivals within the same industry for a broad sample of manufacturing firms. The empirical results suggest that inclusion of the spatially lagged term of a firm’s leverage could be empirically significant in explaining the optimal choice of a firm’s financial structure. Many existing studies aimed to empirically examining the link between the extent of competition and a firm’s financial decisions such as IPO timing (Chod and Lyandres, 2011), leverage choice (Lyandres, 2006), payout policy (Grullon and Michaely, 2006) and repurchasing (Massa et al., 2007) often require proxies for the degree of competitive interaction among product market rivals, given the degree of product market competition not being straightforwardly observed. The proxies usually employed in these studies include the number of rival firms in the same industry, estimated effect of firms’ actions on their rivals’ marginal profits, Herfindahl index and among others. These empirical studies, however, share two common features that should be noted. First, the proxies, in entering the regression equation, are treated as one of the exogenous determinants of optimal leverage like firm size and profitability, among others. Second, the regression model is often estimated by a OLS-based method. Intuitively, the proxying method is appropriate only if the number of competing firms is substantially large and the influence of any one firm is too limited to affect the entire competition structure in output market. Only under this circumstance should the proxies for the extent of competition be regarded approximately exogenous. Otherwise, when the number of competing firms is moderate and a single firm’s operating strategy may have significant impact on the rest, exogeneity of the proxying variable and justifiability of the resulting OLS-based estimation method seems to be debatable. By contrast, the spatial econometric framework manages to circumvent the dilemma like this: on the one hand, it accommodates possible endogeneity 3 of intra-industry competition explicitly in the model specification; on the other hand, there have already existed numerous well-developed estimation methods to cope with such endogeneity. In this sense, this paper supplements the existing literature by providing an alternative empirical approach to examining the relation between firms’ financial choices and output market competition. The remainder of this paper is organized as follows: Section 2 revisits Lyandres (2006)’s model and presents our theoretical predictions. Section 3 lays down the empirical framework and summarizes the empirical implications of the model. Section 4 includes the empirical evidence in support of the theoretical results and Section 5 concludes. Technical details are relegated to the Appendix. 2. Theoretical Motivations While Lyandres (2006) concentrates on the equilibrium relation between the degree of competitive interaction on output market and the debt levels chosen by its participants, our subsequent analysis explicitly highlights the strategic interaction of firms’ financial choices induced by their competition in output market. To save space while keep our expositions complete, the rest of this section will be organized like this: we briefly re-introduce the key elements of Lyandres’ model in Section 2.1. Our main theoretical prediction is summarized in a proposition with some discussions coming in order in Section 2.2. The proof is relegated to the Appendix. 2.1 The Model Lyandres (2006)’s model has two stages. In the first stage, enter the market and simultaneously choose their debt level firms, , ’s. In the second stage, information concerning the financial structures of these firms is disclosed and shareholders of each firm choose their aggressiveness of market strategy ’s to compete against each other in the product market. A firm’s market strategy is assumed to affect its expected value from three aspects. First, a more aggressive strategy, corresponding to greater , generates higher cash flows once it succeeds. We assume that the cash flows generated by the project are proportional to the aggressiveness chosen by the firm through a parameter aggressiveness .1 Second, the greater is, the more likely the project is to fail. Third, the probability of project success relies not only on the operating decisions made by the firm itself but also on the strategies chosen by its rivals. All the three aspects are incorporated into the following shareholders’ expected payoff function: (1) where , is the expected value of firm ’s equity, and denote the aggressiveness In Lyandres’ model, the marginal benefit of aggressiveness is standardized to one. It is relaxed in Eqn.(1) since we intend to emphasize that the strategic response among firms’ financial choices is shown to rely only on the extent of market competition while independent of this parameter . See Proposition 1 in Section 2.2. 1 4 and debt level chosen by firm , reflects the marginal cash flow generated by an aggressive strategy. The parameters determine the extent of impact of a , then firm’s strategy upon the probability of another firm’s project’s success. If and firms strategy compete in strategic substitute. Increasing the aggressiveness of firm reduces the marginal profit of firm default. For , firms aggressiveness of firm and ’s strategy ’s and also increases its probability of compete in strategic complements. Increasing the raises the marginal profit of firm , firms reduces its probability of default. In the marginal case and and also operate independently in output market and firm ’s strategy has no impact on the cash flows the firm receives. Consequently, is the probability of success for firm and this probability decreases with its own aggressiveness and is related to other firms’ operating strategies through , . The term reflects a firm’s equity value if the project succeeds. It is the residual cash flows generated by a successful project after the debt obligations are met. 2.2 Theoretical Prediction and Discussion Similar to Lyandres (2006), we can solve the game using backward induction. To keep the analysis manageable and to highlight the essential characteristics, we make the following simplifications: Assumption 1. (i) ’s, , are equal to each other, namely, ,and they satisfy (ii) . Assumption 1 is implicitly maintained in Lyandres (2006)’s model. Assumption 1-(i) requires the extent of competition is homogenous among firms in output market and excludes the coexistence of competition both in strategic substitutes and strategic complements among the firms considered. (ii) implies for any , , which requires that the total extent of competition from all the rival firms should have an upper bound. As the number of firms becomes larger, the extent of competition from a single rival firm becomes smaller. Without (ii), the system may have no equilibrium solution as goes to infinity. Our main theoretical motivation then can be summarized by the following proposition. Proposition 1. (i) The slope of the debt level reaction function depends on extent of market competition operating aggressiveness and is independent of the marginal benefit of . (ii) Under Assumption 1, for any firms compete in strategic substitutes, namely, , then compete in strategic complements, namely, , then Assumption 1, decreases with as and , if the ,; if the firms . (iii) Under . Proposition 1 highlights some new features of the relationships between firms’ financial choices and their product market competition, and thus supplements the existing literature in several aspects. First, it explicitly relates how and to what extent a firm strategically adjusts its financial structure in response to the change of its rival firms’ to the parameters of competition structure . Lyandres(2006) derives the 5 relationship between extent of competition in product market and firm’s optimal leverage, but he does not derive the explicit form of firm’s leverage function in response to other firms’ leverage change. Second, it clarifies the exact role played by the firms’ product strategy type, i.e. strategic substitutes or strategic complements, in determining the direction of strategic interaction of firm’s financial structures. Proposition 1-(ii) should be interpreted carefully. Essentially, it characterizes how a firm will reacts instantaneously towards its rivals’ deviation from the equilibrium leverage. Intuitively, consider the firms compete in strategic substitutes and at time 0, all of them stayed at their equilibrium debt levels. Assume that at time 1, one firm developed or acquired a new technique. This positive external shock may induce this firm to raise its debt level and undertake a more aggressive market strategy, e.g. increase the output. The Proposition then says that for the rest of the firms, their optimal reactions towards such change are to instantaneously reduce their leverages, which is consistent with the conclusion of Brander and Lewis (1986). Third, the monotonic relationship between the extent of market competition and the magnitude of the strategic adjustment is useful. It allows me in the subsequent empirical analysis to indirectly compare the extents of market competition across different industries through the estimated magnitude of strategic adjustment of financial structures. To get a visual impression of Proposition 1-(iii), we plot the function graphs of for in Figure 1. The most important feature of Figure 1 is that both and always decreases with and by looking into increases with for . Figure 1. Graphs of with Different ’s. 3. Empirical Framework In this section, we relate the theoretical results derived in Section 2 to a class of spatial autoregressive models and establish the framework that can be used to test the theory. According to Proposition 1, we are interested in both the magnitude and the sign of . In this position, we argue that spatial econometrics provide us an appropriate approach to estimating the reaction slope in practice. Specifically, we can estimate the 6 following regression equation to learn about : (2) , . Eqn. (2) is termed as spatial autoregressive (SAR) model. As usual, each , contains a set of exogenous determinants of firm leverage , such as firm , size and profitability, among others. Generally, the most important component of SAR model is the spatial weights , which determine the structure of cross sectional dependence among cross sectional agents. Parameterized SAR model requires specification of these spatial weights in line with certain economic theories or practical principles by the applied researchers.1 The parameter of most interest in model (2) is the spatial autoregressive coefficient , whose magnitude measures the extent of the relevant cross-sectional dependence and whose sign suggests the direction of such dependence. For our problem, as the cross sectional dependence of firms’ financial structures is caused by their competition in product market, one may expect the relevant spatial weights can be defined from the perspective of product market competition. See Section 4.3 for a detailed description of spatial weighting schemes for this paper. Within the empirical setting of Eqn. (2), the empirical implications of theoretical predictions can be summarized as follows: First, the sign of should be determined by firms’ product market competition strategy. If they compete in strategic substitutes, one has . Second, by Proposition 1-(iii), the absolute value of is positively related to the extent of firms’ competition in product market. As one feature of this SAR model (2), the spatially lagged term correlated with the error term is , and such endogeneity then precludes any OLS-based estimation methods. To get a consistent estimate of the coefficients, a number of estimation procedures have been developed for SAR model, including the method of moments by Kelejian and Prucha (1999, 2010), the method of quasi-maximum likelihood estimation by Lee (2004), the method of two-stage least squares by Kelejian and Prucha (1998), Lee (2003), Zhang and Zhu (2010) and the generalized method of moments by Lee (2007), Lin and Lee (2010), and Liu et al. (2010). It is instructive to compare the spatial econometric framework with the empirical setting in the prior literature. The close relationship between firms’ financial choices and their competition in output market has been recognized by many financial economists. Empirical test as such requires proxies for the degree of competitive interaction among product market rivals, given the degree of product market competition not being straightforwardly observed. The proxies usually employed in these studies include the number of rival firms in the same industry, estimated effect of firms’ actions on their rivals’ marginal profits, Herfindahl index and among others. For example, Lyandres(2006) As one of the recent advances, spatial econometrics has developed models which relax parameterization of the spatial weights and allow for nonparametrically specified spatial weights (Pinkse and Slade 2002). But the issue of nonparametric spatial weights will not be explored in this paper. 1 7 regresses the leverage on these proxies and other exogenous determinants of optimal leverage to empirically explore the relationship between optimal leverage and extent of competition. In addition, such tactics has been used by Chod and Lyandres (2011) to examine the relationship between product market competition and firms’ IPO timing and by Grullon and Michaely (2006) and Massa et al. (2007) to empirically investigate the relationship between extent of competition and firms’ payout policies and repurchasing decisions, respectively. These empirical studies, however, share two common features that should be noted. First, the proxies, in entering the regression equation, are treated as one of the exogenous determinants of optimal leverage like firm size and profitability, among others. Second, the regression model is often estimated by the OLS-based method. Arguably, capital structure may influence a firm's willingness or ability to compete with its rivals, i.e., its potential entry/exit decisions, and a firm's position in its competitive environment, measured by price, output and market share, e.g., the industry's equilibrium Herfindahl index. Phillips (1995) and Chevalier (1995a, 1995b) empirically investigate the interaction between product market outcomes and capital structure by examining competitive responses to sharp increases in leverage. A subsequent group of studies by Khanna and Tice (2000, 2005) and Campello (2003) analyze shocks to competitive environments, exploring how differences in ex-ante capital structure are associated with differential responses and competitive outcomes. Considering possible influence that a firm’s capital structure may have upon the consequent competitive environment, the proxying method is appropriate only if the number of competing firms is substantially large and the influence of any one firm is too limited to affect the entire competition structure in output market. In other words, only under this circumstance should the proxies for the extent of competition be regarded approximately exogenous. Otherwise, when the number of competing firms is moderate and a single firm’s operating strategy may have significant impact on the rest, exogeneity of the proxying variable and the resulting OLS-based estimation method seems to be debatable. In other words, the prevalent tactics of proxying is vulnerable to such a dilemma: On the one hand, applied researchers expect their choice of proxies to be indisputably exogenous. Such expectation is evidenced by the usual OLS-based econometric analysis that follows. On the other hand, reasoning above suggests that the extent of competition might be endogenous. By contrast, the spatial econometric framework manages to circumvent the dilemma like this: on the one hand, it accommodates possible endogeneity of intra-industry competition explicitly in the model specification; on the other hand, there have already existed numerous well-developed estimation methods (mentioned in the 3rd paragraph of this section) to cope with such endogeneity. In this sense, this paper supplements the existing literature by providing an alternative empirical approach to examining the relation between firms’ financial choices and output market competition. 8 Since our spatial model essentially regresses leverage on the weighted industry mean leverage and other control variables, this framework looks very similar to the framework of Frank and Goyal (2009), who regress leverage on the industry median leverage and other control variables. The differences between the two settings, however, can be argued as follows. First, the two regression frameworks are differently motivated. Frank and Goyal (2009)’s motivation for the inclusion of industry median leverage is to employ it as a proxy for some factors that are common to all firms in the industry but are not captured by firm-specific variables. By contrast, inclusion of a spatially lagged term in our regression is to explicitly capture firms’ strategic interaction in their financial decision making. Second, as will become clear in Section 4, we run each regression on the firms belonging to the same manufacturing industries but not pooling them together from different industries. This arrangement makes our framework not conflict with Frank and Goyal (2009)’s because for the firms in the same industry, they share an identical median leverage and this common effect will be absorbed into the intercept term or the year dummy variable. As a result, the spatial coefficient, once the common factors are controlled for by the intercept or year dummy, exactly captures the strategic response as desired. Third, we have pointed out that the spatially lagged term (weighted mean of dependent variables) in the SAR model (2) is endogenous in nature, implying that any OLS-based estimator will generally produce an inconsistent result. Similar argument applies to Frank and Goyal (2009)’s regression which assumes the dependent variable (leverage) responds to the median of its neighboring values (median industry leverage). In this sense, Frank and Goyal’s estimate remains vulnerable to the dilemma discussed above, that is, given the fact that the industry median leverage can not be indisputably justified to be exogenous, usual OLS-based econometric analysis is probably unreliable. 4. Empirical Evidence Our empirical implementation essentially builds on the setting of Lyandres (2006) and Frank and Goyal (2009). But while there already have been a large body of empirical literature examining various factors that determine a firm’s financial structure, or the relation between firms’ capital structure and their product market strategies, the subsequent empirical exercise we will do differs from the previous studies in the three aspects that follow. First, our empirical exercise serves to explicitly test the theoretical predictions concerning the strategic interaction of firms’ capital structures derived in Section 2 while the past researches are directed to examine other aspects of the capital structure theory. Second, unlike all its past studies, we use the spatial econometric setting to do the empirical work. This framework, as introduced in Section 3, is developed specifically for testing the theoretical results derived in this paper and surely, allows us to learn about something that can not be revealed using the variable proxying method. Third, unlike the past studies, we run each regression for the sample of the firms from the same manufacturing industries but not for the firms pooled from different industries. The rationale for this is twofold. On the one hand, only the firms in the same manufacturing 9 industry are expected to compete against each other in the product market. Regression by pooling the firms from different firms does not make any financial sense. On the other hand, since one may expect the capital structure determination takes on some industry-specific characteristics, the extent of strategic interaction of firms’ financial choices may also vary from industry to industry. Although pooling the firms from different industries poses no technical difficulty for estimation, such pooling has nothing to contribute but blur the industry-related heterogeneities. 4.1 Data and Econometric Model The data source used in the empirical test is Compustat Annual Industrial Files. To make our empirical findings robust and credible, the empirical exercises are conducted with the companies drawn from ten industry groups (composed of firms with identical SIC codes up to the 3rd or 4th digit) of the manufacturing division. Table 1 lists these industry groups considered in our empirical test. The sample consists all Compustat firms belonging to these industries and having a complete record on the variables used in the analysis from 1998 to 2009. Table 1:Summary of Industry Groups Description of Industry SIC(First 3 or 4 Digits) Total Observations Beverages 208x 369 Women’s, Misses’ and Juniors’ Outerwear 233x 179 Household Furniture 251x 201 Paper Mills 262x 182 Motor Vehicles And Motor Vehicle Equipment 371x 922 Pharmaceutical Preparations 2834 2652 In Vitro and In Vivo Diagnostic Substances 2835 835 Biological Products, Except Diagnostic Substances 2836 2205 Printed Circuit Boards 3672 259 Semiconductors and Related Devices 3674 1730 We choose these manufacturing industries for empirical analysis for several reasons. First, the manufacturing division contains a relatively large number of major groups and industry groups. Second, each industry group, particularly those which we choose in Table 1, has the number of firms ranging from twenty to over two hundred. Considering the firms’ number in an industry is expected to be negatively related to the intensity of market competition, such variation allows us to explore the debt reaction slope within diverse types of market competition. Third, manufacturing enterprises are expected to intensely compete against each other in the product market and this competitive feature makes the sample typical for testing the empirical relevance of the theory. With a double index for each observation, the regression model (2) can be written as 10 (3) , where firm has observations, is the number of total observations in a given industry. Choice of control variables and specification of spatial weights for model (3) will be discussed in detail below. 4.2 Definition of Variables Leverage:We use market leverage ratios in the empirical test. A firm’s market leverage is defined as the ratio of the book value of its debt to the sum of the market value of its equity and book value of its debt. We also run the regression using book value of leverage ratio, defined as the ratio of the book value of its debt to the book value of assets and obtain similar empirical results. A large body of literature on capital structure has explored a large number of determinants of firms’ leverage choices. In a recent study, however, Frank and Goyal (2009) show that only a small number of factors are really empirically robust and financially significant. Particularly, they identify seven factors that seem to be the most important in explaining firms’ leverage choices. In line with Lyandres (2006), we use five of them, namely, Mix of growth options and assets in place: measured as the ratio of the sum of the market value of its equity and the book value of its debt to the book value of its assets. Collateral: measured as the ratio of the sum of net fixed assets and inventories to assets. Profitability: measured as the ratio of operating income to assets. Dividends: a 0-1 dummy, which equals one if the firm has paid cash dividend in a given year. Size: defined as the logarithm of its assets. Year dummy: used to control for macroeconomic changes over time. See Lyandres (2006) for a brief review of related theories on how these factors explain the cross sectional variation in firms’ capital structures. The other two factors that were found by Frank and Goyal (2009) to be influential in explaining firms’ leverage choice are the median industry leverage and expected inflation. However, since our regression is run on firms belonging to the same industry, the two factors will be absorbed into the intercept term and year dummy respectively, and will not appear in the control variables. The descriptive statistics of the dependent and control variables of several selected industries are summarized in Table 2. 4.3 Spatial Weighting Schemes In any empirical implementation using spatial econometric model, choice of appropriate spatial weights scheme is fairly important. For the current study, specifying the spatial weights are motivated by the following three considerations. First, as the firms in the same industry manufacture the products with similar characteristics, any firm is expected to compete with, in other words, be spatially correlated with the rest of the industry. Second, as argued, such interdependence of firms’ financial choices, if any, is 11 induced by their competition in product market. Then the extent of the correlation between two firms is expected to rely on their relative market shares or operating incomes. Table 2:Summary of Statistics of Firm Characteristics Industry Market Stat SIC M/B Profitability Collateral Dividend Size Leverage Ratio Dummy Mean 0.215 4.729 1.169 0.145 0.461 0.456 SD 0.210 3.485 1.471 0.422 0.237 0.435 Max 1 10.79 74.37 0.723 0.987 1 Min 0 -5.809 0.163 -4.342 0 0 Mean 0.245 6.234 2.342 0.214 0.431 0.571 SD 0.145 4.231 1.843 0.634 0.214 0.345 Max 1 13.34 56.34 0.832 0.891 1 Min 0 -2.635 0.496 -5.345 0 0 Mean 0.278 4.576 3.234 0.174 0.573 0543 SD 0.261 2.346 2.242 0.564 0.424 0.442 Max 1 9.42 65.73 0.674 0.967 1 Min 0 -1.451 0.345 -2.342 0 0 Mean 0.381 5.325 4.578 0.262 0.427 0.527 SD 0.264 3.432 2.563 0.632 0.372 0.362 Max 1 8.272 85.57 0.894 0.941 1 Min 0 -4.562 1.527 -3.458 0 0 Mean 0.344 6.344 2.169 0.184 0.452 0.436 SD 0.242 3.905 1.548 0.274 0.354 0453 Max 1 7.36 102.45 0.824 0.923 1 Min 0 -6.238 0.546 -2.345 0 0 208x 233x 251x 262x 371x Note: The sample consists all Compustat firms belonging to these industries and having a complete record on the variables used in the analysis from 1998 to 2009. A firm’s market leverage is defined as the ratio of the book value of its debt to the sum of the market value of its equity and book value of its debt. M/B ration is measured as the ratio of the sum of the market value of its equity and the book value of its debt to the book value of its assets. Collateral is the ratio of the sum of net fixed assets and inventories to assets. Profitability is the ratio of operating income to assets. Dividend indicator is a 0-1 dummy, which equals one if the firm has paid cash dividend in a given year. Size: defined as the logarithm of its assets. Third, as a matter of fact, prior to the current paper, spatial econometrics has already been applied to a number of branches of economics such as regional, urban and environmental economics. Therefore, it is useful to refer to some practical criteria of specifying spatial weights that have already been well recognized in other application fields of spatial econometrics. For example, Fredriksson and Millimet(2002), in analyzing the strategic environmental policymaking of U.S. states, define the following population 12 weighted spatial weights matrix , with , (4) denotes the population of state , where is the set of states that share a common boundary with state . Moreover, Case et al. (1993) in examining the budget spillover and fiscal interdependence among U.S. states, argue that states may regard as neighbors other states that are similar to them economically or demographically, regardless of geographical proximity. For example, they construct the following spatial weights based on states’ economic similarity: , (5) where is the income per capital of state averaged over the sample period. All these considerations suggest that we explore the following two criteria in specifying the spatial weights matrix for the current study. Let upon firm at year . (a) The spatial weight be the spatial weight put by firm is equal to the proportion of firm total sales to the sum of total sales of all other firms in the industry than firm ’s at year . Equivalently, we have (6) , where or is the total sales of firm , let at year . If firm has no observation at year , . Since for each industry in our sample, there are always more than two firms observed at a given year, a weighting scheme like (6) is always feasible. Moreover, unlike operating incomes and net profits, total sales are always nonnegative, thus avoiding some unnecessary troubles in constructing the weights and interpreting the results. (b) Following Case et al. (1993), for any ,firm ’s weight given by firm is proportional to the reciprocal of the absolute value of their sales difference at year , namely, (7) , . Both schemes introduced above imply different interaction patterns among firms. By definition, scheme (a) assumes a firm is more likely to be influenced by firms with larger sales than with smaller ones in the same market while scheme (b) assumes that a firm is likely to interact with peer firms with comparable sales. A close look into both further suggests a significant difference between them. The row elements of summed to be one, but it is not so for are always . Actually, there are two merits for such row normalization. First, row normalization facilitates interpretation of spatially lagged term of cross sectional unit as weighted average of its neighbors. Second, it can make different spatial autoregressive parameters comparable.1By row normalization, the elements of can be redefined as As the magnitude of the spatial interaction parameters are to be compared across industries, the second merit is of much importance to the subsequent analysis. 1 13 (8) . , 4.4 Empirical Results We employ the quasi-maximum likelihood method (Lee 2004) and two-stage least squares (2SLS, Kelejian and Prucha 1998) to estimate model (3) under the weighting schemes as defined above. The 2SLS estimator employs as the instrument for the endogenous spatially lagged term while the maximum likelihood program can be downloaded from http://www.spatial-econometrics.com. According to Lee (2007), for a spatial autoregressive model like (3), both MLE and 2SLSE are consistent but the former will be more efficient than the latter. The regression results using market leverage as the dependent variable are reported in Table 3. 1 The main findings from Table 3 are summarized as follows. Table 3:Regression Results Industry Div Size M/B Profitability Adjusted Collateral SIC Dummy MLE 0.127* 0.027** -0.014 -0.021* 0.268** -0.083** 0.346 2SLS 0.144* 0.026** -0.010 -0.026* 0.246** -0.080** 0.325 MLE 0.099* 0.025* -0.015 -0.030* 0.233** -0.077** 0.330 2SLS 0.104* 0.028** -0.018 -0.027* 0.242** -0.079** 0.321 MLE 0.146** 0.019* -0.008 -0.016* 0.366** -0.066* 0.425 2SLS 0.158** 0.016* -0.006 -0.017* 0.347** -0.054* 0.419 MLE 0.117* 0.015* -0.006 -0.012* 0.329** -0.074** 0.410 2SLS 0.123* 0.018* -0.007 -0.016* 0.384** -0.076** 0.402 MLE 0.091* 0.024* -0.010 -0.033* 0.336** -0.098** 0.376 2SLS 0.078* 0.022* -0.008 -0.036* 0.314** -0.100** 0.367 MLE 0.066 0.019* -0.009 -0.037* 0.366** -0.097** 0.343 2SLS 0.068 0.022* -0.009 -0.035* 0.345** -0.098** 0.338 MLE 0.218** 0.037** -0.024* -0.024* 0.375** -0.086** 0.465 2SLS 0.222** 0.038** -0.022* -0.025* 0.386** -0.085** 0.456 MLE 0.229** 0.034** -0.021* -0.020* 0.401** -0.086** 0.451 2SLS 0.236** 0.032** -0.026* -0.026* 0.392** -0.087** 0457 MLE 0.157** 0.045** -0.021* -0.032* 0.328** -0.106** 0.376 2SLS 0.156** 0.046** -0.022* -0.033* 0.314** -0.132** 0.334 MLE 0.131* 0.043** -0.022* -0.030* 0.324** -0.118** 0.379 2SLS 0.127* 0.047** -0.019* -0.034* 0.313** -0.114** 0.340 (a) 208x (b) (a) 233x (b) (a) 251x (b) (a) 262x (b) (a) 371x (b) Note: The regression results are computed using market leverage ratio as the dependent variable. Double asterisk (**) 1 Regression results using book leverage ratio as the dependent variable are available upon request. 14 indicates statistical significance at 95% level while single asterisk (*) indicates statistical significant at 90% level. The spatial weighting schemes (a) and (b) are defined according to (6) and (8), respectively. Table 3:(Continued) Industry Div Size M/B Profitability Adjusted Collateral SIC Dummy MLE 0.069 0.031** -0.027* -0.032* 0.311** -0.075* 0.255 2SLSE 0.068 0.032** -0.030* -0.034* 0.312** -0.078** 0.244 MLE 0.066 0.036* -0.026* -0.033* 0.316** -0.074* 0.247 2SLSE 0.065 0.037** -0.028* -0.032* 0.312** -0.078** 0.241 MLE 0.112* 0.025* -0.012 -0.025* 0.280** -0.095** 0.326 2SLS 0.108* 0.024* -0.010 -0.026* 0.273** -0.099** 0.318 MLE 0.088* 0.027* -0.008 -0.025* 0.277** -0.092** 0.313 2SLS 0.080 0.028* -0.011 -0.024* 0.283** -0.096** 0.310 MLE 0.127* 0.035** -0.037** -0.023* 0.425** -0.087** 0.395 2SLS 0.126* 0.038** -0.036** -0.025* 0.427** -0.085** 0.397 MLE 0.113* 0.033** -0.032** -0.022* 0.473** -0.088** 0.387 2SLS 0.098* 0.037** -0.038** -0.024* 0478** -0.084** 0.379 MLE 0.264** 0.042** -0.015 -0.033* 0.395** -0.095** 0.347 2SLE 0.280** 0.045** -0.013 -0.034* 0.394** -0.099** 0.348 MLE 0.229** 0.046** -0.012 -0.030* 0.384** -0.096** 0.339 2SLS 0.238** 0.048** -0.011 -0.036* 0.389** -0.101** 0.336 MLE 0.188** 0.039** -0.011 -0.040* 0.366** -0.115** 0.327 2SLS 0.199** 0.041** -0.010 -0.040* 0.365** -0.116* 0.325 MLE 0.148* 0.037** -0.009 -0.045* 0.349** -0.118** 0.318 2SLS 0.135* 0.042** -0.008 -0.043* 0.359** -0.115** 0.316 (a) 2834 (b) (a) 2835 (b) (a) 2836 (b) (a) 3672 (b) (a) 3674 (b) Note: The regression results are computed using market leverage ratio as the dependent variable. Double asterisk (**) indicates statistical significance at 95% level while single asterisk (*) indicates statistical significant at 90% level. The spatial weighting schemes (a) and (b) are defined according to (6) and (8), respectively. First, in most cases, both MLE and 2SLSE of the spatial autoregressive parameter have significantly positive signs across the industries based on both spatial weighting schemes, implying that firms are engaged in strategic interaction with their rivals in choosing financial structures. Moreover, positive the spatial coefficient suggests that firms tend to react in the same direction to leverage adjustment of their rivals and they compete against each in strategic substitutes. Second, we see significant variation of the extent of strategic interaction across the industries. Among the selected industries, the extent of strategic interaction varies from 0.06 to 0.28, implying certain heterogeneity of competition intensity from one product market to another. Among them, firms in the industries of paper mills or printed circuit 15 boards are expected to adjust their financial structures by a larger margin in response to rivals’ change, suggesting more intense competition in these markets. Furthermore, Lyandres (2006) show that the extent of market competition is expected to be negatively related to the firms’ number in the market. Our estimation results seem to be consistent with this prediction. Third, in most cases, the estimates associated with weighting scheme (b) have smaller magnitude than those with scheme (a). It seems to us that scheme (a) would be a more appropriate criterion relative to scheme (b) in characterizing firms’ cross-sectional correlation in their output market in terms of the magnitude of both the estimates and the adjusted R square. Considering the different patterns implied by two weighting schemes, this may indicate that a firm would be likely to mimic the larger firms or just follow the industry leaders rather than interact with its comparable peers. Fourth, coefficient signs of the control variables are consistent with prior literature. For example, firm size and collateral are positively correlated with the leverage while M/B ratio, dividend dummy and profit margin are negatively correlated with the leverage. Finally, estimated coefficients of these controls are robust to the choice of spatial weighting scheme, suggesting that firm-level characteristics are not strongly correlated with rivals' leverage ratios. 4.5 Robustness Check To check the robustness, first we run a pooled regression on the sample from several industries which have been estimated to have relatively significant extent of strategic interaction. That is, we put the firms from seven industries (excluding SIC 251x, 2834 and 2835) together and run the regression (3) by adding an industry dummy. It should be noted that in this regression the spatial matrix will be a block-diagonal matrix, i.e., , where are the spatial weight matrices for , each industry in the previous regressions. The regression results are summarized in Table 4. We find that the results have the same sign and comparable magnitude to those reported in Table 3. Table 4:Robustness Check: Pooled Regression Div Size M/B Profitability Adjusted Collateral Dummy MLE 0.117* 0.033** -0.024* -0.030* 0.380** -0.098** 0.247 2SLSE 0.105* 0.032** -0.022* -0.034* 0.357** -0.096** 0.239 MLE 0.094* 0.035** -0.023* -0.032* 0.371** -0.099** 0.227 2SLSE 0.096* 0.033** -0.023* -0.033* 0379** -0.094** 0.225 (a) (b) Note: The regression pools the firms from seven industries (excluding SIC 251x, 2834 and 2835) together and run the regression (3) by adding an industry results. Double asterisk (**) indicates statistical significance at 95% level while single asterisk (*) indicates statistical significant at 90% level. Although Table 3 has indicated certain robustness of our results to the choice of spatial 16 weighting schemes, a naïve weighting scheme is used here to check the robustness further. We assign equal weight to each firm in the same industry, regardless of their heterogeneous market shares, that is, we let (9) , where , , is the number of firms observed at year . This weighting scheme makes our regression resemble Frank and Goyal (2009)’s regression setting where the industry median leverage is replaced by the sample mean leverage here, although both are differently motivated. The estimation results are reported in Table 5. Under equally weighting scheme, the spatial coefficient is less significant than under scheme (a) and (b), which might be regarded as evidence for asymmetry of the strategic interaction among firms. Such asymmetry is also consistent with the observation in Section 4.4 that the data are in favor of weighting scheme (a) (that puts more weight on larger firms) relative to scheme (b) (that puts more weight on comparable peers). Since the fundamental purpose of this paper is to empirically test for the strategic effect, exploring the exact strategic pattern and distinguishing between several conflicting patterns of interaction seem to be beyond the scope of this paper. It may be the topic for future research. Table 5:Robustness Check: Equally Weighting Scheme Industry Div Size M/B Profitability Adjusted Collateral SIC Dummy MLE 0.054 0.026** -0.012 -0.023* 0.263** -0.083** 0.308 2SLSE 0.050 0.024** -0.011 -0.027* 0.248** -0.078** 0.299 MLE 0.078* 0.019* -0.009 -0.015* 0.358** -0.065* 0.378 2SLSE 0.075 0.020* -0.006 -0.013* 0.343** -0.055* 0.372 MLE 0.043 0.025* -0.011 -0.036* 0.337** -0.094** 0.316 2SLSE 0.042 0.023* -0.009 -0.032* 0.323** -0.094** 0.320 MLE 0.104* 0.035** -0.025* -0.021* 0.367** -0.087** 0.419 2SLSE 0.115* 0.036** -0.025* -0.023* 0.373** -0.085** 0.423 MLE 0.082* 0.042** -0.023* -0.030* 0.330** -0.102** 0.353 2SLSE 0.083* 0.040** -0.024* -0.032* 0.315** -0.126** 0.348 MLE 0.035 0.028** -0.026* -0.037* 0.307** -0.077* 0.230 2SLSE 0.038 0.029** -0.029* -0.036* 0.312** -0.078** 0.228 MLE 0.062 0.025* -0.015 -0.024* 0.274** -0.090** 0.286 2SLSE 0.066 0.027* -0.012 -0.022* 0.268** -0.095** 0.291 MLE 0.074* 0.037** -0.035** -0.025* 0.428** -0.087** 0.367 2SLSE 0.072* 0.034** -0.032** -0.027* 0.430** -0.084** 0.361 MLE 0.135* 0.040** -0.013 -0.034* 0.396** -0.094** 0.319 2SLSE 0.132* 0.041** -0.012 -0.032* 0.391** -0.100** 0.312 MLE 0.094* 0.043** -0.013 -0.040* 0.367** -0.114** 0297 2SLSE 0.098* 0.045** -0.010 -0.043* 0.365** -0.117* 0.285 208x 233x 251x 262x 371x 2834 2835 2836 3672 3674 17 Note: The regression uses an equally weighting scheme, as defined by (9). Double asterisk (**) indicates statistical significance at 95% level while single asterisk (*) indicates statistical significant at 90% level. Finally, instead of using the first order spatial lag of the exogenous regressors as the instrument for the endogenous part, we can use second order or higher order spatial lags as the instruments. The regression results are quite similar to the 2SLSE in Table 3 and are not reported here. 4.6 Further Analysis Although our prior analysis seems to support the strategic interaction hypothesis, we provide further evidence in this subsection. To distinguish the interpretation of our empirical result from other competing theoretical alternatives, particularly Frank and Goyal (2009)’s, we regress the time changes in firms leverage on time changes of weighted average leverage of other firms in the same industry and other controls, that is, (10) , where 。The basic intuition behind such regression , is that even if the industry median leverage in Frank and Goyal (2009)'s is empirically significant in explaining a firm’s leverage, it will be eliminated by first-order differencing while our weighted mean leverage is not. Consequently, the spatial coefficient remains identifiable from (10). Although regression (10) can eliminate the industry-specific common effect, it is by no means the only way to achieve this. We have also explored two alternative specifications for robustness consideration. One regresses the time in firms leverage on time changes changes of weighted average leverage of other firms in the same industry and other controls, namely, (11) . Note that the interpretation of the two regressions (10)-(11) differs in that the latter indicates a simultaneous strategic reaction of firm’s leverage adjustment while the former implies that such strategic effect is one period lagged. The third regression then is aimed to nesting both above, that is, we regress the time time and changes in firms leverage on both changes of weighted average leverage of other firms in the same industry and other controls, namely, (12) . We use the two-stage least squares method (Kelejian and Prucha, 1998) to estimate Eqn. (10)-(12) under weighting scheme (a)-(b) and both the spatial coefficients and R squared are presented in Table 6. The major finding from Table 6 is that the spatial coefficients for regression (10)-(11) are estimated to be significantly positive across most industries and these coefficients have comparable magnitude to those obtained in prior regressions, suggesting that the 18 hypothesized strategic effect is empirically distinguishable from Frank and Goyal (2009)’s common effect hypothesis. Moreover, it is also noted that the coefficient for lagged strategic interaction is essentially insignificant given the presence of simultaneous strategic effect. These results indicate that such interaction is likely to take place simultaneously instead of having a dynamic pattern. Table 6:Further Analysis: Regression by Differencing Eqn. (10) Industry SIC Eqn.(11) Adjusted Eqn.(12) Adjusted Adjusted (a) 0.125* 0.298 0.130* 0.317 0.047 0.117* 0.326 (b) 0.094* 0.289 0.103* 0.312 0.039 0.092* 0.325 (a) 0.136* 0.395 0.151** 0.405 0.059 0.138* 0.417 (b) 0.107* 0.374 0.120* 0.389 0.043 0.105* 0.396 (a) 0.068 0.337 0.080 0.359 0.032 0.073 0.372 (b) 0.043 0.329 0.051 0.336 0.029 0.038 0.345 (a) 0.198** 0.427 0.217** 0.447 0.094* 0.187** 0.457 (b) 0.189** 0.414 0.203** 0.441 0.086* 0.174** 0.449 (a) 0.136* 0.319 0.150** 0.332 0.059 0.128* 0.339 (b) 0.121* 0.328 0.128* 0.353 0.052 0.124* 0.359 (a) 0.058 0.229 0.065 0.247 0.034 0.052 0.258 (b) 0.051 0.212 0.058 0.236 0.028 0.045 0.241 (a) 0.094* 0.302 0.101* 0.324 0.046 0.083* 0.338 (b) 0.079* 0.287 0.085* 0.305 0.042 0.073 0.326 (a) 0.103* 0.363 0.124* 0.384 0.049 0.094* 0.393 (b) 0.094* 0.349 0.104* 0.363 0.041 0.089* 0.375 (a) 0.227** 0.328 0.253** 0.345 0.102* 0.202** 0.367 (b) 0.219** 0.314 0.236** 0.332 0.094* 0.198** 0.343 (a) 0.162** 0.305 0.162** 0.328 0.072 0.136* 0.335 (b) 0.136* 0.289 0.136* 0.314 0.058 0.118* 0.324 208x 233x 251x 262x 371x 2834 2835 2836 3672 3674 Note: The regression results are computed using the two-stage least squares method. Double asterisk (**) indicates statistical significance at 95% level while single asterisk (*) indicates statistical significant at 90% level. The spatial weighting schemes (a) and (b) are defined according to (6) and (8), respectively. 5. Conclusion Modern capital structure theory is marked by systematically examining the interaction between firms’ financial decisions in capital market and their industrial competition in output market. Focusing on the strategic interaction of firms’ financial choices induced by their competition in product market, this paper provides several new results that have not appeared or have not been explicitly investigated in the past literature: 19 Theoretically, we show how the competition in output market may induce a firm to take its rival firms’ capital structures into calculus in deciding its own one. Furthermore, the direction of such strategic interaction is related to firms’ competitive strategy type and a monotonic correspondence is established between the magnitude of such strategic adjustment and the extent of competition in output market. Empirically, a spatial econometric model is introduced to test the relation between firms’ financial choices and their market strategies. We lay down the empirical framework, interpret the empirical relevance of the parameters and evaluate the proposed setting in contrast with variable proxying method employed by the prior literature. We also elaborate on the specification of spatial weights by borrowing from some practical criteria that have already been well recognized in other application fields of spatial econometrics. Based on a board sample of U.S. firms from several manufacturing industries, the empirical facts tell us that inclusion of the spatially lagged term of a firm’s leverage could be significant in explaining the optimal choice of firm’s financial structure, at least empirically. Appendix: Proof of Proposition 1. Proof of (i): We solve the game by backward induction. In the second stage of the game, the management chooses the operating strategy to maximize the expected wealth of the firm’s shareholders, conditional on firms’ financial structures disclosed after the first stage. Differentiate the right-hand side of Eqn (1) with respect to , it gives . (A.1) with Introduce the following notations, , , , . Then Eqn. (A.1) can be written as , (A.2) where is an identity matrix. For the moment assume Then firm ’s aggressiveness leverage , can be represented as the linear combination of firms’ , namely, . (A.3) By defining (A.4) is invertible. , the following identities always hold, , , . Let’s return to the first stage of the game. In the first stage, firms choose their debt levels, with the objective of maximizing their ex ante values of both debtholders and shareholders, given by (A.5) . Substituting Eqn. (A.3) into (A.5) and using the identities in (A.4), we have (A.6) 20 , where and row of and denote the column vector composed of the elements in the th , respectively. Each is column vector with th component being one and others being zero. The first order condition for the firms to choose their optimal debt levels is then given by (A.7) , to be the unique solution to the For now denote . equations in (A.7). For our purpose, it is of interest to learn something about how a firm’s equilibrium debt level changes in response to the change of the other firm’s equilibrium debt level, that is, . To this end, it is unnecessary to solve explicitly the in attempt to work out the response slope as a function of equations for . Instead, in view of implicitly determined by Eqn. (A.7), we can with some given differentiate the right-hand side of (A.7) with respect to , which gives . (A.8) Then Proposition 1-(i) follows since the right-hand side of Eqn. (A.8) is independent of . Proof as of (ii): Starting from Eqn.(A.4), one . For any -dimensional column vector may write nonsingular matrix and some , applying the formula , where gives for . Then one has and . It suffices to verify that under Assumption 1, there and holds always has contrary sign to follows from both follows from both and . If and . If , , . Whenever , there always holds or , , (b) because (a) (c) , and (d) . Proof of (iii): The right-side hand of Eqn. (A.8) can be represented as , where and , , . Then it suffices to verify the first order derivative of with reference to first order derivative of has negative sign. It can be verified as follows: the has the same sign as 21 . By some computation, one has , and , because , , . References [1] Bolton, P. and Scharfstein D., 1990, A theory of predation based on agency problems in financial contracting, American Economic Review, 80, 93–106. [2] Brander, J. and Lewis, T., 1986, Oligopoly and financial structure: The limited liability effect, American Economic Review, 76, 956–70. [3] Bulow, J. John, G. and Paul, K., 1985, Multi-market oligopoly: strategic substitutes and complements. Journal of Political Economy, 93, 488-511. [4] Case, A. C. , Rosen, H. S. and Hines , J.R., 1993, Budget spillovers and fiscal policy interdependence: evidence from the states, Journal of Public Economics, 52, 285 -307. [5] Campello, M., 2003, Capital structure and product markets interactions: evidence from Business cycles, Journal of Financial Economics, 68, 353-378. [6] Chevalier, J., 1995a, Capital structure and product market competition: empirical evidence from the supermarket industry, Journal of Finance, 50, 1095-1112. [7] Chevalier, J., 1995b, Do LBO supermarkets charge more? An empirical analysis of the effects of LBOs on supermarket pricing, American Economic Review, 85, 415-435. [8] Chod, J. and Lyrandres, E., 2011, Strategic IPOs and product market competition, Journal of Financial Economics, 100, 45-67. [9] Frank, M. and Goyal, V., 2009, Capital structure decisions: Which factors are reliably important, Financial Management, 38, 1-37. [10] Fredriksson, P. G., and D. L. Millimet, “Strategic Interaction and the Determinants of Environmental Policy across US States”, Journal of Urban Economics, 2002, 51, 101-122. [11] Glazer, Jacob, “The strategic effects of long-term debt in imperfect competition”, Journal of Economic Theory, 1994, 62, 428–443. [12] Grullon G. and Michaely, R., 2008, Corporate payout policy and product market competition, Working paper,Rice University and Cornell University and IDC. [13] Kelejian, H.H. and Prucha, I.R., 1999, A generalized moments estimator for the autoregressive parameter in a spatial model, International Economic Review, 40, 509-533. [14] Kelejian, H. H. and Prucha, I. R., 2010, Specification and estimation of spatial autoregressive models with autoregressive and heteroscedastic disturbances, Journal of Econometrics, 157, 53-67. [15] Khanna, N. and Tice, S., 2000, Strategic responses of incumbents to new entry: the effect of new ownership structure, capital structure and focus, Review of Financial Studies, 22 13, 749-779. [16] Khanna, N. and Tice, S., 2005, Pricing, exit and location decisions of firms: evidence on the role of debt and operating efficiency, Journal of Financial Economics, 75, 397-427. [17] Lee. L.F., 2003, Best spatial two stage least squares estimators for a spatial autoregressive model with autoregressive disturbances, Econometric Reviews, 22, 307-335. [18] Lee, L.F., 2004, Asymptotic distribution of quasi-maximum likelihood estimators for spatial autoregressive models, Econometrica, 72, 1899-1925. [19] Lee, L.F., 2007, GMM and 2SLS estimation of mixed regressive, spatial autoregressive models, Journal of Econometrics, 137, 489-514. [20] Lin, X. and Lee, L.F., 2010, GMM estimation of spatial autoregressive models with unknown heteroscedasticity, Journal of Econometrics, 157, 34-52. [21] Lyandres, E., 2006, Capital structure and interaction among firms in output markets: theory and evidence, Journal of Business, 79, 2381–2421. [22] Maksimovic, V., 1988, Capital structure in repeated oligopolies, RAND Journal of Economics, 19, 389–407. [23] Massa, M., Rehman, Z. and Vermaelen, T., 2007, Mimicking repurchase, Journal of Financial Economics, 84, 624-666. [24] Phillips, G., 1995, Increased debt and industry product markets: an empirical analysis, Journal of Financial Economics, 37, 189-238. [25] Pinkse, J., Slade, M. and Brett, C., 2002, Spatial price competition: a semiparametric approach, Econometrica, 70, 1111–1153. [26] Showalter, D., 1995, Oligopoly and financial structure: comment, American Economic Review, 85, 647–53. [27] Schuhmacher, F., 2002, Capacity-price competition and financial structure, Working paper, Aachen University of Technology, Aachen. [28] Zhang, Z. and Zhu, P., 2010, A more efficient best three stage least squares estimator of spatial autoregressive models, Annals of Economics and Finance, 11, 155-184. 23
© Copyright 2026 Paperzz