How Ownership Structure Affects Capital Structure and Firm Performance? Recent evidence from East Asia

Transcription

1 How Ownership Structure Affects Capital Structure and Firm Performance? Recent evidence from East Asia Nigel Driffield, Aston Business School Vidya Mahambare Cardiff Business School Sarmistha Pal Brunel University * 14 March 2005 Abstract There is a good deal of anecdotal evidence suggesting that the lack of corporate governance was significant in generating a deep and long-lasting crisis in the South East Asian economies in the late 1990s, though this remains largely hitherto untested. This paper thus examines the effects of corporate governance structures on capital structure and performance of south East Asian firms in the period leading up to the crisis. Previous work in this area largely ignores the bias generated by simultaneity between capital structure and firm performance, and we show that this can generate misleading results. There is evidence of non-entrenchment dilution effects so that higher voting rights give rise to higher leverage in both countries though higher voting rights may increase or decrease profit margin depending on the level of concentration in ownership. JEL: G32, L25, Keywords: Asian Crisis,, Capital structure, Firm performance, Simultaneity bias. * Corresponding author: Department of Economics and Finance, Brunel University, Uxbridge UB8 3PH, UK. Tel ; Fax The research is funded by the ESRC grant number RES Sarmistha Pal is much grateful to Professor Stijn Claessens for providing her the ownership data. We are solely responsible for any errors. 1

2 How Ownership Structure Affects Capital Structure and Firm Performance? Recent evidence from East Asia 1. INTRODUCTION The Asian Crisis of the late 1990s has highlighted the problems of corporate governance among South East Asian corporations. While recent literature confirms aspects of concentrated ownership, dominance of controlling shareholders, separation of voting and cash flow rights and limited protection of minority rights in many of these countries badly affected by the Crisis (Claessens et al., 2000; 2002), an understanding of the effects of ownership structure on capital structure and firm performance remains much unexplored. While Claessens et al. (2000) examine the pattern of ownership, Claessens et al. (2002) analysed the effects of ownership structure on firm valuation. Lemmon and Lins (2003) in addition link ownership structure to stock returns in these countries, but largely ignoring the effect of ownership structure on capital structure and firm performance in the worst affected countries. The purpose of this paper is to fill in this gap of the literature and examine how ownership structure may affect capital structure and firm performance. In doing so, we not only allow for the possible non-linearity in these relationships, but also correct for the simultaneity bias, if any, between capital structure and firm performance, often ignored in the literature. The relationship between ownership structure, capital structure and firm performance is far from being unambiguous. Traditional literature highlights that agency problems between managers and shareholders may reduce the leverage ratio below the optimum level, in an attempt to ensure the continued viability of the firm. Jensen and Meckling, (1976) however argue that introduction of managerial share ownership may reduce these agency problems, thus aligning the interests of managers and shareholders. 1 Brailsford et al.(2002) have gone further to suggest that the 1 Demsetz (1983), Demsetz and Lehn (1985) went further to claim that the level of optimal managerial ownership is firm-specific and endogenous to expected performance. 2

3 relationship between managerial share ownership and leverage may in fact be nonlinear. There is also evidence that concentration of ownership may improve (e.g., Shleifer and Vishny, 1986) or even deteriorate firm performance, depending on the level of concentration (e.g., see Morck et al., 1988). Much of this literature is however based on the functioning of the US firms and as such these models assume a much wider variation in ownership structure than one finds in SE Asian countries. Ownership pattern among the East Asian corporations is not only concentrated (often dominated by family ownership), but often characterised by the presence of a CEO, Board Chairman or Vice Chairman who is also a controlling shareholder of the company. Presence of a controlling manager shareholder may however have mixed effects on firm performance, depending on the level of concentration. This is not all; ownership is also characterised by separation of voting rights from cash flow rights where control rights (or voting rights) of the largest owners were often generally greater than the corresponding cash flow rights prior to the crisis (Claessens et al., 2000). In traditional literature ownership refers to cash-flow rights, i.e., the right to claim dividends. Voting right refers to the degree of control of a firm, i.e. the right of a shareholder to vote in person or by proxy for members of the board of directors and other corporate policies. Higher voting rights may give rise to serious agency problems due to deviation from one-share-one-vote, pyramiding schemes and/or crossholding. When a large shareholder keeps significant control rights with relatively small cash flow rights, s/he may be averse to increasing outside equity financing because the latter may threaten the dominance of the controlling shareholder (often labelled as non-dilution of entrenchment effects, e.g., see Claessens et al., 2002). In addition, separation of voting rights from cash flow rights may enhance the incentives of controlling shareholder to engage in expropriation, which in turn is likely to adversely affect the performance of the firm, and in turn its value, especially during crisis time. In fact, in the East Asian it is often difficult to dissociate concentrated ownership from separation of cash flow and voting rights. 2 Ownership concentration is also likely to be closely related to the presence of controlling manager shareholder (see further discussion in section 2). These characteristics emphasize the complex nature of the inter-relations between ownership 2 Within a theoretical model, Bebchuk (1999) demonstrates that the two go hand in hand. 3

4 structure, capital structure and firm performance in East Asia that we try to disentangle in the present paper. Our work is distinctive in a number of ways. (a) The theoretical basis of any link between ownership structure on the one hand and capital structure and firm performance on the other in our work allows for both moral hazard and adverse selection (e.g., see Bajaj, Chan and Dasgupta, 1998). This allows us to determine indicators of capital structure and firm performance jointly in terms of ownership and the degree of monitoring 3, both assumed to be given exogenously. (b) In doing so, we also recognise the problems of simultaneity between capital structure and firm performance, often ignored in the literature, but recently highlighted by Berger and di Patti (2003). (c) In view of evidence of non-linearity between ownership concentration and capital structure in our samples, we also try to incorporate the possible non-linearity in these relationships. Our analysis is based on the Worldscope firm-level panel data for the period We study two of the countries most deeply affected by the crisis, Indonesia and Korea. These countries provide an interesting contrast, given the different corporate histories in the two countries, and the different levels of development of their capital markets (for further discussion on this see Chelley- Steeley, 2004). These differences could yield significant differences in the effects of ownership on capital structure and firm performance in our analysis. The paper is developed as follows. Section 2 presents the data and its characteristic features, highlighting the relationships between ownership structure, capital structure and firm performance. Section 3 explains the analytical framework to explain the observed pattern in our sample while section 4 outlines the econometric issues and discusses the empirical results. Finally section 5 concludes. 2. DATA AND PRELIMINARY OBSERVATIONS 3 While ownership concentration variable is directly observable, we use some proxies to capture the degree of monitoring including family ownership and separation of voting rights from cash flow rights (see further discussion in section 3). 4

5 Data used for the analysis in this paper come from two sources: (a) firm-level accounting data comes from Worldscope (b) Firm-level ownership data comes from Claessens et al. (2002). The final data set is then constructed by matching the Worldscope company accounting data with the Claessens ownership structure data. Since ownership structure is rather stable over time (La Porta et al. 1999), we employ 1996 ownership data to examine the relationships between ownership structure, capital structure and corporate leverage during Ownership Structure The differences in ownership structures among firms in these countries are illustrated in Table 1A and Table 1B. While 75% of firms in Indonesia were family owned, the proportion is even higher in Korea (79%). Secondly, in nearly 70% of firms the CEO, Board Chairman or Vice-chairman was also a controlling owner in both countries, labelled as Cronyman here. Our data also reveals a strikingly close association between family ownership and the presence of a Cronyman in both countries. For example, as high as 90% of family owned firms in Indonesia is characterised by the presence of a Cronyman; the corresponding proportion is about 77% for the Korean family owned firms. In contrast, presence of Cronyman is rather low among the non-family owned firms in both countries, especially in Indonesia. The separation of voting rights from cash flow rights is another important feature of East Asian corporations. In particular, voting rights are higher in more than half the sample Indonesian firms (the corresponding proportion is about 25% among Korean firms). More interestingly, there is a close association between presence of Cronyman and higher voting rights in both countries: as high as 90% or more firms with Cronyman is also associated with higher voting rights in these countries. This initial analysis clearly reveals the complex nature of ownership structure in the selected countries. This needs to be taken care of in our final analysis as to how ownership structure affects capital structure and firm performance, generally overlooked in the existing literature. 5

6 2.2. Capital Structure Most firms in our sample tend to use both debt and equity finance. Less than 10% firms in our sample countries use no debt. Leverage is a measure of capital structure in our analysis and is measured by debt-equity ratio, defined as total debt divided by book value of common equity. In some cases however debt-equity ratio could be negative if the equity value is negative though the corresponding debt may be high. In order to avoid this problem of negative debt equity ratio, we shall make use of the absolute value of the debt-equity ratio in our analysis. Table 2A shows the share of low debt-equity firms (firm relying more on equity financing) in Korea and Indonesia, for a base year (1993), the period in the run up to the crisis ( ) and the crisis period ( ). For comparison, we also consider the corresponding proportion of low leverage firms in Singapore, a country that remained least affected by the crisis. In comparison to 22% Korean and 59% Indonesian firms, as high as 84% of firms in Singapore relied more on equity financing during (Table 2). 4 Demirguc-kunt and Maksimovic (1995) suggest that the over-reliance on debt in the worst affected countries, especially Korea, can be partially explained by the relatively low levels of stock market development in the country Ownership and capital structure In this section, we explore the nexus between ownership structure and capital structure among Indonesian and Korean corporations in our samples. First, Table 3 summarises the average leverages for different types of ownership structure in the two countries. Clearly, for any category of ownership structure, average levels of leverage are lower in Indonesia in the pre-crisis period, though it moved up radically in the post-crisis period. While average leverage levels were higher among the Korean firms in the post-crisis period, the difference in the levels of leverage before and during the crisis was much limited in the Korean case. While highest level of concentration (>50%) in Korea has been associated with highest leverage in the pre-crisis period,, highest average level of leverage has been noted for Indonesian firms with medium 4 These figures contrast with Singapore, one of the least affected countries, which relies far more heavily on equity finance. 6

7 range of concentration. Presence of a cronyman is however associated with higher level of leverage in both countries while higher voting rights does not necessarily give rise to higher leverage in our samples (seems to hold only for Indonesia). In addition, the nonparametric Kernel scatter plot (see Figures 1, 2) reveals some degree of non-linearity in the relationship between level of ownership concentration and capital structure in both countries, though particularly among Indonesian firms. For example, there seems to be a u-shaped relationship for Indonesian firms for both sub-periods and 97-98: thus at a lower level of concentration leverage may fall (and more outside equity may be used) possibly because existing shareholders are not concerned about the dilution of their dominance. But at a higher level of concentration, leverage increases (the trend is more obvious if we exclude the outlier firm with concentration level of 73%) because of the fear of dilution of dominance of large controlling shareholders. The u-shaped relationship is however not so pronounced for Korea at lower level of concentration, especially for the period; during this period, leverage level does not change much with increase in level of concentration (below 45% level). Similar trend is observed for period at comparable level of concentration. But beyond 45% level of ownership concentration, one can identify a kind of u-shaped relationship among the Korean firms as well for the period 94-96; non-linearity is much less obvious during the crisis period where clearly leverage falls with increase in concentration beyond 45% level Ownership and firm performance Table 3 also shows the average levels of profit margin associated with different types of ownership structure in our samples. In this respect, we focus on the pre-crisis years (94-96) as during the crisis years (97-98) there has been a general deterioration of firm performance. For this period average profit margin increases slowly with levels of concentration in Korea, though the effect seems to be just opposite in Indonesia, though only marginally. We also do not observe any significant difference in profit margin for Indonesian firms with/without family ownership, cronyman or higher/lower voting rights while the difference is only marginal among Korean firms such that profit margin is slightly lower among Korean firms with family ownership, cronyman or higher voting rights. 7

8 We also make use of the nonparametric kernel scatter plots (see the middle panels in Figures 1, 2). While there is no obvious non-linearity in this respect for Indonesia (more or less uniform performance with higher levels of concentration), one can observe some degree of non-linearity in the relationship for Korean firms, especially noted for the crisis period. In particular, it appears that compared to firms with lowest and highest levels of concentration, firm performance is lower for the firms with the medium levels of concentration (above 30% and below 50%); similar trend is noted for both sub-periods in Korea. Having done the initial analysis, we shall now move onto develop an analytical framework where we incorporate the complex nature of the interrelationship between ownership structure, capital structure and firm performance, as noted in this section. 3. ANALYTICAL FRAMEWORK This section outlines the analytical framework to study the effects of ownership structure both on capital structure and firm performance in East Asia. In doing so, we take account of the possible simultaneity between capital structure and firm performance, which may bias the results otherwise. We also take account of the possible non-linearity between (a) ownership concentration and capital structure, (b) ownership concentration and firm performance and (b) capital structure and firm performance, as evident in our samples (see discussion in section 2). An understanding of the conflict of interests between managers and owners, i.e., agency problems, remains central to an understanding of how ownership structure may affect capital structure and firm performance (e.g., Leland and Pyle, 1977; Jensen and Meckling, 1976). In a recent attempt Bajaj, Chan and Dasgupta (1998) extend Leland and Pyle (1977) and develop a signalling model to show how both adverse selection and moral hazard may interact to determine firms financing decisions and performance measures in terms of ownership structure. Our empirical work is primarily based on the framework suggested by Bajaj et al. (1998). 8

9 3.1. A General Model The simple cross-sectional implications of Bajaj et al. s work (1998) are pertinent for our analysis. Denoting indices of capital structure and firm performance by Y 1 and Y 2 respectively, one can write the following: Y Y 1i 2i = f ( α, γ ) = g( α, γ ) i i i i (1) Thus each endogenous variable Y ki, k=1,2, for the i-th firm, i=1,,n j for the j-th country depends on indices of ownership (α i ) as well as degree of outside monitoring (γ i ). Bajaj et al. (1998) argued that ownership 5 is positively correlated with indices of firm performance and also with various measures of the debt-equity ratio 6 ; ownership is however negatively correlated with perquisite consumption per unit of investment. 7 The latter is labelled as a measure for the degree of moral hazard. The agency view would imply that firms with greater degree of moral hazard should have more debt, which commits managers to paying out residual cash flow a la Jensen, (1986). If however outside monitoring is less effective, managers have an incentive to under-lever the firm to avoid bankruptcy risk (e.g., see Mehran, 1992) Specification of the Empirical Relationships Our discussion in sections 1 and 2 highlights the essential differences of the ownership structure in East Asia. While much of the existing literature assumes ownership is widely dispersed, La Porta et al. (1999) suggest that ownership may be highly concentrated in many countries including the countries of our interest (e.g., 5 Here ownership is defined as managerial shareholding. 6 Zhang (1998) too argues that a controlling large shareholder is more averse to risky projects (due to under-diversification, which is also the opportunity cost of concentrated ownership) than shareholders whose portfolios are fully diversified. The latter may result in under-investment by rejecting projects preferred by the minority shareholders. This under-investment problem can however be mitigated by issuing debt since the risk-shifting effect of debt offsets the under-investment incentive of the underdiversified owner. Thus a firm s leverage increases with concentrated ownership, and this relation becomes stronger the more risk-averse the controlling shareholder is. 7 It is assumed here that the manager s compensation is affected not only by the fraction of equity they retain, but also by their ability to divert cash flows for perquisite consumption. The latter captures how moral hazard too can affect the relationships of our interest. 9

10 Claessens et al. 2002). Secondly, our concentration variable relates to overall ownership concentration; we do not however have any continuous information on managerial shareholding; the Cronyman variable is arguably the closest proxy for managerial shareholding in our data-set. Thirdly, in view of the observed non-linearity especially between ownership and capital structure, we need to take account of this non-linearity, that has not been discussed by Bajaj et al. (1998). Finally, unlike most existing studies, we allow for the possible simultaneity between capital structure and firm performance and also the non-linearity between capital structure and firm performance noted in our sample. Last, but not the least, we need to take account of the high degree of correlation between different ownership variables in our samples and thus carefully choose the best and the most parsimonious specification. All these considerations necessitate us to modify the set of equations (1). This is explained below Ownership and Capital Structure While ownership structure is directly observable, certain clarifications are in order. First of all we observe the cash flow rights of top five shareholders; the latter constitutes our measure of ownership concentration. Although we do not observe managerial shareholding, we observe if the CEO, board chairman or Vice Chairman is also a controlling owner (variable labelled as Cronyman in our data-set). These two variables constitute our indices of ownership structure. However, we need to be careful here as there is a high degree of correlation between levels of concentration and Cronyman in our samples (see discussion in section 2). This necessitates that we either include the level of concentration or the Cronyman variable in our analysis, but try to avoid including both in one equation. A more difficult problem is to find an appropriate measure of the degree of monitoring. Various proxies have been used in the existing literature, e.g., percentage of outside directors (Mehran, 1992), shareholder voting rights (Lippert and Moore) or control potential (e.g., measured by institutional ownership, as in Mehran, 1995). Given the limited ownership information at our disposal, we could possibly use two indices to instrument the degree of monitoring in our model; first, if control rights are greater than the cash flow rights and also if the largest share holder is a family (family 10

11 ownership). When a large shareholder keeps significant control rights with relatively small cash flow rights, s/he has little stake in firm value and can get away despite taking reckless policies undermining the interests of the company. Similar problem may arise with a family ownership. Thus in these cases market forces such as the product market (Hart 1983) or the corporate control market (Stulz 1988) may fail to discipline the controlling shareholder towards firm value maximisation. In addition, Zhang (1998) suggested that higher concentration of ownership in the hands of a few holders may lead to slower response to changing market conditions due to a lack of professional monitoring mechanism. Secondly, a higher level of ownership concentration may be an indication of an environment where it is costly to conduct control-related activities. In other words, our concentration variables including Cronyman would indirectly account for the lack of monitoring of the activities of minority of controlling shareholders. Level of leverage among firms in our samples is however contingent on the level of concentration, and we observe a kind of non-linear relationship in this respect. This is evident in the non-parametric Kernel scatter plots (see Figures 1, 2 and discussion in section 2). There is a kind of u-shaped relationship for Indonesian firms in particular such that at lower level of concentration, shareholders may make use more of outside equity (resulting in a lower leverage) since they would not be concerned about the dilution of their dominance. The relationship however seems to change as we move to higher level of concentration when leverage level increases with further increases in levels of concentration possibly because of the non-dilution of the entrenchment effect. Similar effect is also noted among Korean firms though it remains less pronounced. Finally, we need to allow for the fact that we have a sample of panel nature where we observe firms over a period of five year, though we assume ownership structure to be stable over this period. This in turn means that our ownership variables do not vary over time (as is indicated by Bajaj et al, 1998), though most other firm-level variables tend to vary over time. Taken together, the relationship between ownership structure and capital structure (DE) for firm i in year t can be expressed as follows: 11

12 DE + it α 4 = α α + ( <= 25%) i+ (25% < <= 50%) i+ ( > 50%) i i + α Famowni + Votingi + X + u 5 α 6 α 7 1it 1it Cronyman Concen α Concen α Concen where X 1it refer to other possible control variables (see discussion later in this section) and the residual error term is u 1it. The binary variable Famown takes a value 1 if the largest owner of the i-th firm is a family and zero otherwise. Voting is a binary variable taking a value 1 if voting rights of the largest shareholder is higher than the cash flow rights. As argued above, inclusion of Famown and Voting is expected to control for the variation in the degree of monitoring. However, given the high degree of correlations between these ownership variables, we find it difficult to include all the variables in the estimated equation (2). We systematically run a series of estimates to see the nature of these estimates and after careful inspection find that two specifications seem to dominate the others: (a) concen and voting along with other control variables X 1it and (b) Famown and voting along with X 1it ( see further discussion in section 4). Assuming ownership structure of the i-th firm to be given exogenously, we experimented with a series of ownership variables including the top five-shareholder s concentration ratio. Thus Equation (2) shows the most general specification that we have in mind. In order to obtain the best estimates, we however need to experiment with different combinations of ownership variables and also with different cut-off points for the ownership concentration variables (to capture non-linearity; see further discussion in section 4). (2) Ownership and Firm Performance The link between ownership structure and firm performance has been subject to an on-going debate going back to Berle and Means (1932), who suggested that firms with a wide dispersal of shares tend to under-perform. In general, a positive relation between ownership concentration and firm efficiency is predicted and many studies (Shleifer & Vishny, 1986; Short, 1994; Gedajlovic & Shapiro, 1998; Thomsen & Pederson, 2000; Gorton & Schmidt, 1996; Kang and Shivadasani, 1995) have empirically confirmed this positive relationship between level of concentration and firm efficiency. Some studies have however contradicted this general finding (see, for 12

13 example, Demsetz and Lehn, 1985 and Morck et al. 1988), much of which appears to be explained not only by difficulties in obtaining a uniform measure of firm performance but also by a lack of appropriate control for ownership structure and also whether ownership structure is treated as endogenous. Firm performance/efficiency PFT in our analysis is measured by the pre tax profit margin. 8 In constructing a standard model of firm performance based on the industrial economics literature, one would include numerous variables relating to market structure, such as industry concentration, in order to allow for inter-firm variation in profits generated through inter-industry variation. However, such data that can be matched in with these data are not available, so it is necessary to remove the industry level variation from the data. We therefore calculate the firm level deviation of firm profit (pre and post tax) from the corresponding industry mean 9 and specify the most general profit equation (3) as follows: PFT + β it 4 = β 0 + β Cronyman i ( Concen <= 25%) β 5 Famown i + i β 2 β 6 (25% < Concen <= 50%) + Voting i + β X + u 7 1it 2it i β 3 ( Concen > 50%) Here X 2it captures all other possible factors influencing this relationship (3). Other variables included in equation (3) are rather similar to equation (2). We include three levels of ownership concentration variable to capture the possible non-linearity, if any. In the absence of continuous information on managerial shareholding, we also include the binary variable Cronyman to indicate the dominant role of managers in the board of director. In addition to these ownership concentration variables, inclusion of Famown and Voting is expected to control for the degree of outside monitoring on firm performance. (3) i 3.3. Capital Structure and Firm Performance Most existing literature, however, tends to ignore the simultaneity between firm performance and capital structure. If firm performance affects the choice of capital structure and vice versa, then the failure to take this into account may result in serious 8 Please note that we also tried using post-tax profit margin and obtained similar results. 9 In section 4 we present the estimates using profit margin in deviation form. We however find that estimates using profit margin with industry dummies are very similar to those using profit margin as deviation from industry mean. 13

14 simultaneity bias, with important implications for pattern of firm financing and performance. Berger and di Patti (2003) offer two hypotheses for the reverse causation. First, more efficient firms choose lower equity ratios than others, all else equal, because higher efficiency reduces the expected costs of bankruptcy and financial distress. The second hypothesis focuses on the income effect of the economic rents generated by efficiency (as an indicator of performance) on the choice of leverage. Thus more efficient firms choose higher equity capital ratios, all else equal, to protect the rents or franchise value associated with high efficiency from the possibility of liquidation. Prior evidence supports the notion that firms hold additional equity capital to protect franchise value (e.g., Keeley, 1990). In the light of the two-way relationship between capital structure and firm efficiency, one needs to allow for simultaneity between equations (2) and (3). Thus the modified equations of interest will be as follows: DE + it α 4 PFT + β it 4 = α 0 + α 1 Cronyman = β 0 + Cronyman ( Concen <= 25%) + i β i 1 + α 5 5 Famown Famown i i + i α 2 α 6 ( Concen <= 25%) + + β + i β 2 β 6 (25% < Concen <= 50%) + Voting Voting i i + + α 7 7 PFT DE it it + α 8 (25% < Concen <= 50%) + β +β i α 3 ( Concen > 50%) 2 PFT u it X + +α 1 9 1it (2 ) i β 3 X + u 7 1it 2it ( Concen > 50%) As argued above firms with higher efficiency (measured by higher profit margin) may substitute outside equity capital for debt so that α 7 >0 in equation (2 ). On the other hand it may also be true that more efficient firms try to protect the value of their high income by holding more equity capital so that α 7 < 0. The estimated value of α 7 would capture the net value of these two possible and opposite effects. One may also expect some non-linearity in the effects of firm efficiency on capital structure so that firms at a higher level of efficiency may behave differently from those at a lower level. Since we are not sure about the nature of this nonlinearity, we experiment with a few alternatives, namely, (a) inclusion of an additional square term of efficiency measure in equation (2 ); (b) replacing efficiency measure (3 ) i it i 14

15 by its log (natural) and (c) inclusion of an additional inverse term of the efficiency measure. The agency cost hypothesis would predict that an increase in leverage raises efficiency, i.e., β 7 > 0. Some may however argue that there is a possible non-linearity in the effects of leverage on profit margin as a measure of firm efficiency. In particular, when leverage is sufficiently high, further increases may result in lower efficiency because the benefits in terms of reduced agency costs of outside equity are overcome by greater agency costs of debt. Our discussion in section 2 however suggests that there is not any evidence of non-linearity as reflected in the nonparametric scatter plots (Figure 1, 2); hence we do not allow for any non-linearity in the effects of capital structure on profit margin. This is an important difference between equations (2 ) and (3 ) Other Explanatory Variables In addition to indicators of ownership pattern, leverage and firm performance, a number of other control variables are included in equations (2 ) and (3 ). Firm size: Firm size is measured by the log of total sales. Firm size may be positively Friend and Lang, 1988; Marsh, 1982) or negatively (Rajan and Zingales, 1995) related to leverage. Large firms may exercise economies of scale, have better knowledge of markets and are able to employ better managers. Large size may enable greater specialisation. It may also measure a firm's market power or the level of concentration in the industry. On the other hand, however, relatively large firms can be less efficient than smaller ones, because of the loss of control by top managers over strategic and operational activities (Himmelberg et. al 1999, Williamson 1967). Also as Jensen (1986) notes professional managers of a firm (who are not the owners) derive personal benefits from expanding beyond the optimal size of the firm by their desire to have, among others, power and status. The latter may increase leverage and lower firm efficiency. 15

16 Tobin s Q value: This is a proxy for growth opportunities. The trade-off theory predicts that firms with more opportunities carry less leverage. The traditional version of the pecking order theory predicts the opposite result. Debt typically grows when retained earnings are less than investment requirement and vice-versa. Hence, for a given level of profitability, leverage is likely to be higher for firms with more growth/investment opportunities. Age of the firm: Firm performance may depend on the accumulated knowledge about the market, experience and firm s reputation. Hence, one would expect a positive relationship between age and profit margin. Old firms however, may be less open to new technology as well as more rigid in terms of style and effectiveness of managerial governance. This may result in a negative relation between the age and performance of the firm. As for capital structure, old firms, particularly in East Asian countries, are likely to have developed close links with their lenders and hence may be able to acquire debt more easily and at a cheaper rate, resulting in a positive relationship between the age and leverage of the firm. Diversification: A firm is classified as diversified if it operates in more than three market segments, each accounting for more than 10% of the total revenue of the firm. Diversified firms may enjoy higher profits as a result of combining activities such as production, distribution, marketing and research. The transaction cost theory (Williamson 1975) and imperfect external capital markets provide a rationale for firms to diversify. A different strand of this literature, however, argues that diversification has a negative effect on firm performance since diversified firm is prone to cross-subsidise investments poor growth opportunities (Berger and Ofek 1995) and the distortions in investment decisions can occur in the presence of managerial power struggle among the firm's various diversified divisions (Rajan, Servaes, and Zingales 2000). Empirically diversified firms do not appear to perform better and the causation tends to run from low performance resulting in a diversification of a firm. Inconclusive empirical evidence on this issue also suggests that managers may have objectives other than maximising profits, such as the growth of revenue, that lead firms to become diversified. As for capital structure, Lewellen (1971) argues that diversified firms enjoy greater debt capacity. Also if diversified firms have more stable cash flows, this is likely to have a positive impact on the supply of debt. 16

17 17

18 4. EMPIRICAL RESULTS Section 4.1 discusses the estimation issues while section 4.2 presents and analyses the empirical estimates Estimation Issues Given that ownership information is available only for the year 1996, we could construct a cross-section data-set for the period This would mean that there will be a single observation for each firm such that leverage and firm performance relate to the average values of these variables for the period while all other variables correspond to the initial year There are at least two disadvantages with this data-set. First, the single cross-section data cannot capture the aspect of time variation for a particular firm, if any. For one thing, the relationship between capital structure and firm performance is more pertinent for a given firm over time rather than among the cross-section of the firms. Secondly, period could be quite destabilising for the corporate sector in these countries when the crisis was in full fledge. Thus by focusing on the crisis period only, we may lose sight of some significant behavioural patterns among these Asian corporations. Accordingly, we make use of the annual panel data-set for the period , which we believe would capture the behavioural transition of these corporations better. In doing so we however need to assume that ownership structure is relatively stable over time, without much loss of generality (La Porta et al. 1999). In this respect, we also need to clarify the issue regarding the potential endogeneity of ownership as argued by Demsetz (1983) though empirical evidence does not corroborate this. For example, Demsetz and Lehn (1985) used two stage least square (treating ownership as potentially endogenous) to find that ownership fails to explain variations in firm performance, which is further confirmed by Hermalin and Weisbach (1988) and Cho (1998). On the other hand, Morck et al. (1988) and other studies ignored the issue of endogeneity of ownership structure and produce evidence of a statistically significant effect of ownership structure on performance. Thus without much loss of generality, we treat ownership structure to be exogenously given. In any case, given that our ownership information is available only for 1996, following La Porta et al. (1999) we assume ownership structure to be rather stable in 18

19 our sample until before the post-crisis restructuring started. This allows us to focus directly on the issues of our interest, i.e., to reinvestigate the relationship between capital structure and firm performance in terms of ownership structure, among other things, allowing for the simultaneity and non-linearity between capital structure and firm performance. This however means we cannot apply the standard within panel data determination of capital structure and firm performance,. While it is trivial to correct for the potential endogeneity with instrumental variables estimation, a preferred strategy is to jointly estimate equations (2 ) and (3 ), allowing for simultaneity between capital structure and firm efficiency. While the use of panel data to estimate systems of simultaneous equations is well understood, this generally involves converting the data to differences and estimate the system by either three stage least squares (3SLS) or generalised methods of moments (GMM) using lagged values as instruments to generate orthogonality conditions on differenced data. The latter is a different approach to the now well-understood dynamic panel estimation (Arrellano and Bond (1988, 1991) or the more recent Blundell and Bond (1998) GMM systems estimator. This is a more straightforward simultaneous equations estimator following Holtz-Eakin et al (1988) or Cornwell et al (1992), which allows for individual effects both within individual equations and in the covariance matrix between the equations. It still relies on employing lags as instruments, so with short panels of unbalanced data such estimation reduces the number of observations dramatically. However, the essential problem here is that the data contain time-invariant variables. As such, one cannot adopt one of these approaches, as differencing the data becomes infeasible. We therefore adopt the 3SLS within estimation with error components suggested by Baltagi and Li (1992), based on Baltagi (1981). In practice this involves estimating equations (1 ) and (2 ) separately using a standard within estimator 10, and then calculating the covariance matrix between the equations using the errors. The data are the transformed by dividing through by the square root of the covariance, and finally equations (1 ) and (2 ) are estimated by 3SLS employing the transformed data. 10 For both equations for both countries, random the random effects estimator rejects the restriction of fixed effects. 19

20 4.2. Results We have experimented with various specifications, especially those relating to indicators of ownership structure and forms of non-linearity between leverage and profit margin. Given the high degree of correlations between and among various ownership variables of our interest, we started with the individual effects of these ownership variables, taking one at a time. These results are summarised in Appendix Table A1 for the two sample countries. We then tried different combinations of ownership variables (after controlling for all other factors) and after some experimentation, we end up with two sets of most parsimonious 3SLS estimates, which appear to be superior to others in terms of t-statistics and also other diagnostics. These two specifications are labelled here as specification 1 where include Famown and Voting among others (see Table 5A) 11 and specification 2 that includes Concentration (with some non-linear control for the leverage equation) and voting, among others (see Table 5B). We also compare our results from these two specifications with the corresponding single equation estimates (summarised in Table 6A and 6B for these two specifications) Although there are some similarities, differences in these results between Korea and Indonesia are more pronounced. Let us first examine this with respect to the nature of the simultaneity between capital structure and firm performance in the sample countries. For example, higher profit margin raises relative debt levels in Indonesia though the effect is just opposite in Korea. There is also some evidence of non-linearity observed 12 in this respect and this holds for both the sample countries. In particular, while the substitution effect (against equity capital) is greater for more efficient firms in Indonesia, the income effect is relatively greater for Korean firms (see discussion in section 3). Considering the reverse causation, we however do not find any evidence of non-linearity in the effect of capital structure on profit margin. Hence, we do not include the non-linear term in the final estimation. As before, we find opposite effects for the two sample countries here. Higher the absolute level of debt, higher is the profit margin among 11 Appendix Table A2 also shows the GMM estimates for specification There is also some difference in the nature of nonlinearity. The result is saying that leverage declines at a more than linear rate in Korea as profit increases. This is different from Indonesia, where a turning point can be identified, i.e., the two profit terms have opposite signs. 20

21 Indonesian firms and lower is the profit margin among Korean firms in our sample. While the former confirms the agency cost hypothesis, the latter seems to contradict it. Some may however argue that the average level of leverage is significantly higher in Korea (see Table 3) so that further increases in debt may result in lower efficiency because the benefits in terms of reduced agency costs of outside equity are outweighed by greater agency costs of debt. Effects of ownership structure We start with three sets of ownership variables in both equations (2 ) and (3 ), pertaining to concentration (Concen), higher voting rights (Voting) and family ownership (Famown). In both cases the indicator of concentration turns out to be insignificant. Hence in the final set of estimates shown in Tables 5A and 5B, we only include Voting and Famown. Separation of voting rights from cash flow rights has a very pronounced and significant effect in both countries. The effect is similar in case of leverage in both Korea and Indonesia: higher voting rights among the largest shareholders result in relatively higher debt among firms in both countries. This result is supportive of the non-dilution entrenchment effect so that when a controlling large shareholder keeps large control rights with relatively small cash flow rights, s/he can be averse to increasing outside equity financing because the latter may threaten the shareholding dominance of the existing controlling shareholder. The effect of family ownership on capital structure is however significant in Indonesia only such that the prevalence of family ownership is significantly associated with lower debt levels among Indonesian firms. The latter however contradicts the general belief that family owners with close links to financial corporations are more likely to have higher debt. Effects of separation of voting rights from cash flow rights on profit margin are however different among Indonesian and Korean firms in our sample. Higher voting rights lower profit margin among Indonesian firms but enhance it among Korean firms. While the Korean result generally confirms our expectations that higher leverage (if voting rights are higher than cash flow rights) is associated with higher efficiency, Indonesian result seems to oppose it. The latter may be a result of a significantly higher level of concentration in Indonesia as compared to Korea (see Table 3). In addition to the entrenchment effect, high level of concentration may cause 21

22 a slower response to changing market conditions due to a lack of professional monitoring mechanism. A higher level of ownership concentration may also be an indication of an environment where it is costly to conduct control-related activities, thus adversely affecting firm performance. Family ownership, however, seems to lower profit margin among corporations in both sample countries (note that the effect is insignificant in Indonesia). The latter may be in line with the more risk-averse nature of family firms undertaking less risky investment, thus lowering profit margin. The importance of allowing for simultaneity between performance and capital structure is illustrated by the single equation estimates of (1 ) and (2 ) in Tables 6A and 6B. These are standard random effects estimates, and so do not allow for simultaneity are endogeneity. While in general these models perform well, what is clear is that the endogenous variables dominate the ownership variables, such that family ownership and voting structures appear insignificant. As table 5 illustrates however, when one allows for endogeneity in performance, then ownership and voting patterns become important determinants of capital structure and profitability. This is particularly important when one considers the high concentrations of ownership, and high levels of debt that were prevalent in these economies leading up to the crisis. Failing to allow for such simultaneity would therefore generate highly misleading results in terms of the importance of corporate governance in Indonesia and Korea. Effects of other variables Among other included variables, the coefficient of firm size is negative for both countries in the determination of leverage though the effect is not significant for Korean firms. Thus only larger Indonesian firms have significantly lower leverage (absolute). Effects of firm size on profit margin are however opposite in Korea and Indonesia. In particular, larger Indonesian firms are associated with higher profit margin while the larger Korean firms experience lower profit margin. In general, we also note that older firms have lower relative debt as well as profit margin. Q-value are however insignificant in the determination of leverage, though it has significant but opposite effects on profit margin in the two countries. The effect is favourable in Indonesia, but not so in Korea. Assuming q to be a measure of corporate growth 22

23 opportunities, trade-off theory would argue that Korean firms with more opportunities have less leverage and therefore lower profit margin. The opposite (as in the case of Indonesian firms) would support the prediction of pecking order theory so that debt would typically grow if there are more investment opportunities than met by internal funds. Finally, effects of diversification on leverage are negative for firms in both sample countries, perhaps reflecting various kinds of inefficiencies/distortions involving cross-subsidisation of poor investment projects, especially in the presence of managerial power struggle. 7. CONCLUDING COMMENTS While many recent studies have highlighted the role of corporate governance on the recent Asian crisis (e.g., Classens et al., 2000, 2002), effects of corporate governance (as reflected in the ownership structure) of these Asian corporations on capital structure and firm performance remains much unexplored. The present paper departs from this literature not only by examining the effects of ownership structure on capital structure and firm performance, but in doing so it also takes account of the possible simultaneity and non-linearity between capital structure and firm performance. Results obtained from panel data drawn from a sample of Indonesian and Korean firms are supportive of a significant simultaneity between capital structure and firm performance, though these results differ somewhat between these two samples. These results confirm the case of non-entrenchment dilution effects so that higher voting rights give rise to higher leverage in both countries though higher voting rights may increase or decrease profit margin depending on the level of concentration in ownership. 23