The Relationship Between Ownership Structure and Performance in Listed Australian Companies

Transcription

1 The Relationship Between Ownership Structure and Performance in Listed Australian Companies by Emma Welch Abstract: This paper examines the relationship between ownership structure and corporate performance in Australian listed companies. The study applies the models advanced by Demsetz and Villalonga (2001), examining the relationship between ownership and performance when ownership is modelled as a multi-dimensional endogenously determined variable. OLS results suggest that ownership is significant in explaining performance. However, when endogeneity is taken into account, ownership is not statistically dependent on the performance measure. Finally, previous research by authors including Morck, Schleifer and Vishny (1988) suggests that the relationship between ownership and performance is nonlinear. We fit a generalised nonlinear model that nests models advanced previously. Results provide limited evidence of a nonlinear relationship between managerial share ownership and firm performance. Keywords: OWNERSHIP STRUCTURE; PERFORMANCE; ENDOGENEITY OF OWNERSHIP STRUCTURE; NONLINEARITY. School of Finance and Applied Statistics, Australian National University, Canberra, ACT, 0200, AUSTRALIA; Emma.Welch@anu.edu.au Comments and suggestions made by Richard Heaney, Michael Martin and Tom Smith are gratefully acknowledged. Australian Journal of Management, Vol. 28,. 3 December 2003, The Australian Graduate School of Management 287

2 AUSTRALIAN JOURNAL OF MANAGEMENT December Introduction T he relationship between ownership structure and corporate performance is one that has received considerable attention in the finance literature. However, a notable feature of this body of literature is its failure to reach a consensus regarding the nature of the relationship. Demsetz and Villalonga (2001) posit that the conflicting results may stem from differences with respect to the measurement of variables, sample period, estimating technique and whether or not the research explicitly accounts for the endogeneity of a firm s ownership structure, that is documented by Demsetz (1983) and Demsetz and Lehn (1985), among others. Demsetz and Villalonga (2001, p. 211) stress that not only should the endogeneity of ownership structure be accounted for, ownership should be modelled simultaneously, as an amalgam of shareholdings owned by persons with difference interests. In particular, the fractions of shares owned by outside shareholders and by management should be measured separately. The failure of previous research to examine two dimensions of this structure likely to represent conflicting interests 1 provides a motivation for Demsetz and Villalonga (2001) to re-examine the relationship between ownership structure and corporate performance. The current paper revisits the work of authors including Morck, Schleifer and Vishny (1988) and Demsetz and Villalonga (2001), applying similar models to Australian listed companies. The study seeks to add to the limited evidence regarding this relationship in the Australian context, where extant studies have failed to account for the endogeneity of ownership structure, and consider whether results are consistent with those found for American companies. The remainder of this paper is organised as follows. Section 2 reviews the previous studies that examine the relationship between ownership structure and corporate performance, including research that has been conducted using Australian data. Section 3 provides formal model specifications to be fit using the data outlined in section 4. Section 5 describes results as well and discusses their robustness to the use of alternate performance measures. These results are summarised in section Literature Review Berle and Means (1932) are among the first to consider the relationship between a firm s ownership structure and its performance. They assert that as the diffuseness of ownership increases, shareholders become powerless to control professional managers. Further, they argue that, given the interests of management and shareholders are not generally aligned, corporate resources are not used efficiently in maximising corporate profit. Therefore, Berle and Means (1932) suggest that the relationship between ownership concentration and performance should be a negative one. However, Demsetz (1983, p. 386) argues it is unreasonable to suppose that diffuse ownership has destroyed profit maximisation as a guide to resource allocation. Instead, he asserts that a firm s ownership structure is an endogenous outcome of a maximising process in which more is at stake than just accommodating to the shirking problem : 2 1. Demsetz and Villalonga (2001, p 211). 2. Demsetz (1983, p. 387). 288

3 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE When scale requirements are large, especially when the survival of the firm requires a rapid attainment of large scale, then there will be economic pressure to satisfy the consequent need for sizable equity capital by turning to a diffuse ownership structure. The greater monitoring cost that might arise from such an ownership structure may be more than offset by the reduction in risk-associated capital cost, so that maximisation of the value of the assets of the firm actually requires a diffuse ownership structure. single ownership structure is suitable for all situations if the value of the firm s assets is to be maximised. A summary of empirical studies examining the nature of the relationship between ownership and performance is presented in table 1. Despite the conceptual and empirical support for the endogeneity of ownership structure, many studies have failed to take this endogeneity into account when estimating the effect of ownership structure on performance (see Morck, Shleifer & Vishny 1988; McConnell & Servaes 1990; and, in the Australian context, Craswell, Taylor & Saywell 1997). The importance of accounting for the endogeneity of ownership is further emphasised by the work of Demsetz and Villalonga (2001). Modelling ownership as a multi-dimensional variable that separately reflects the fraction of shares owned by outsiders and management and performance, Demsetz and Villalonga (2001) find ordinary least squares testing suggests that firm performance is always dependent on at least one measure of ownership structure. However, when testing is performed using a 2-stage least squares approach, which accounts for the possible endogeneity of ownership structure, neither measure of ownership structure is statistically significant in explaining variation in performance. Demsetz and Villalonga (2001, p. 227) argue these results are consistent with the view that ownership structure is chosen so as to maximise firm performance, and that the greater diffuseness in ownership, although it makes the agency problem more severe, conveys compensating advantages on firms that choose to rely on a diffuse ownership structure. Demsetz and Villalonga (2001) note other fundamental differences in previous studies, including differences in variable measurement, as well as the failure of many researchers to acknowledge that ownership is an amalgam of shareholdings owned by people with potentially divergent interests. These variations, which are clearly evidenced in table 1, may provide an explanation for the failure of the extant literature to reach a consensus regarding the nature of the relationship between ownership and corporate performance. 3. Model Specification The review of extant literature in section 2 shows the considerable variation in ownership and performance measures employed in studies examining the relationship between the two variables. This study will adopt the variable definitions used by Demsetz and Villalonga (2001). More specifically, two ownership variables are considered in the paper: the shareholdings of the firm s 5 289

4 290 Table 1 Summary of Previous Studies Examining the Relationship Between Ownership Structure and Corporate Performance Authors Ownership Measure/s Performance Measure/s Demsetz and Lehn (1985) Morck, Shleifer and Vishny (1988) McConnell and Servaes (1990) Hermalin and Weisbach (1991) Loderer and Martin (1997) Craswell, Taylor and Saywell (1997) 1. % of shares held by top 5 shareholders 2. % of shares held by top 20 shareholders 3. Herfindahl measure of ownership concentration 4. % of shares controlled by top 5 families and individuals 5. % of shares controlled by institutional investors Post-Tax Accounting Profit / Book Value of Equity % of shares held by company directors 1. Tobin s Q 2. Accounting Profit Rate 1. % of shares held by insiders Tobin s Q 2. % of shares held by blockholders 3. % of shares held by institutional investors % of shares held by the current CEO and past CEOs still on the board Methodology Ordinary Least Squares Regression Ownership Endogenous? Results Significant Relationship Piecewise Linear Regression Significant n-monotonic Relationship Ordinary Least Squares Regression Significant Curvilinear Relationship Tobin s Q Piecewise Linear Regression Significant non-monotonic relationship % of shares held by officers and directors Tobin s Q Simultaneous Equations Ownership doesn t predict performance, but performance is a negative predictor of ownership. 1. % of shares held by company directors 2. % of shares owned by institutional investors Proxy Tobin s Q (market value of equity / book value of net 1. Linear Regression 2. Curvilinear Regression 3. Piecewise Regression Cho (1998) % of shares held by company directors Tobin s Q 1. Piecewise Linear Regression 2. System of 3 Equations Himmelberg, Hubbard and Palia (1999) Holderness, Kroszner and Sheehan (1999) Demsetz and Villalonga (2001) % of shares held by insiders managers and directors Tobin s Q 1. Quadratic Piecewise Model 2. Piecewise Linear Model Weak curvilinear relationship. Firm performance affects ownership structure, but not vice versa. Quadratic form of ownership effect on performance % of shares held by officers and company directors Tobin s Q Piecewise Linear Regression Significant non-monotonic relationship % of shares held by top management, the CEO and company directors 1. Tobin s Q 2. Accounting Profit Rate 1. Ordinary Least Squares Regression 2. 2-Stage Least Squares significant relationship AUSTRALIAN JOURNAL OF MANAGEMENT December

5 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE largest shareholders ( TOP5 ); and, the shareholdings of the firm s top management and board of directors ( MSO ). Both ownership variables are defined in greater detail as part of the formal model specification below. As noted by Demsetz and Villalonga (2001), it is important to differentiate between ownership by top shareholders and the board as these two measurements provide an indication of the relative power held by two parties with potentially diverging interests. Moreover, the percentage of shares owned by a company s top 5 shareholders indicates the ability of outside shareholders to control the actions of management. Conversely, the level of board ownership indicates the ability of directors to ignore the wishes of other shareholders. Therefore, these two measures capture the interests of two groups with arguably the most divergent interests. In addition to the ownership variables described above, two measures of performance are collected: Tobin s Q ( Q ); and, average accounting profit rate ( PRATE ). Both performance measures are more comprehensively defined as part of the formal model specifications below. While Tobin s Q is the most common measure that has been used to date in modelling the relationship between ownership structure and corporate performance, it is important to test the robustness of reported results to the use of an alternate performance measure. It is for this reason that modelling is also performed using accounting profit rate as the performance measure. Finally, when modelling the relationship between ownership structure and corporate performance, it is also necessary to control for firm-specific characteristics. The inclusion of such variables allows for the possibility that a number of factors jointly affect ownership structure or corporate performance and therefore induce spurious correlation between them. In considering which control variables to include in the models fitted, previous research in the area was consulted. Table 2 shows control variables used by researchers that have examined the relationship between ownership structure and corporate performance previously. Control variables used in the current study are those employed by Demsetz and Villalonga (2001). Firstly, it is necessary to account for the intangible assets that we could expect to impact on Tobin s Q. We may expect the existence of intangible assets to distort Tobin s Q given the total book value of assets, which represents the denominator of Tobin s Q, may not include the value of all intangibles. Observable measures of these intangible assets include a firm s research and development and property plant and equipment expenditures. Previous studies have also included advertising expenditure as an additional measure of intangible assets that may distort Tobin s Q (e.g. Morck, Schleifer & Vishny 1988; Demsetz & Villalonga 2001, among others). However, given Australian accounting standards, which do not require the disclosure of this expenditure, a similar variable was not included in the current study. 291

6 292 Authors Demsetz and Lehn (1985) Morck, Shleifer and Vishny (1988) McConnell and Servaes (1990) Hermalin and Weisbach (1991) Loderer and Martin (1997) a Craswell, Taylor and Saywell (1997) Cho (1998) Table 2 Control Variables Used by Previous Studies in Modelling the Relationship Between Ownership Structure and Corporate Performance* Firm Size (Av. MV common equity/bv h (RC (RC (ln(rc ) (ln(sales)) (ln(bv ) (log(rc of ) Financial Leverage (MV longterm debt / RC (Debt / RC Liquidity Total Risk of the Firm (σ of stock returns and σ of. Acc ROR) Market Risk Firm-Specific Risk (std. error from market model) Advertising Expenditure (Sales / advertising) (Advertising / RC (Advertising / RC (Advertising weighted by RC (σ and σ 2 of stock returns) (BV debt / BV (MV longterm debt / RC R & D Expenditure (Sales / R&D) (R&D / RC (R&D / RC (R&D weighted by RC Capital Expenditure (Sales / capex) (R&D / BV (Cashflows b /RC of (σ of changes in (profit f / RC (R&D / RC (Capex/RC Industry Membership (Dummies for utility and media firms) (Dummies based on SIC codes) (Dummies based on ASX classes) (Dummies based on SIC codes) AUSTRALIAN JOURNAL OF MANAGEMENT December

7 293 Authors Himmelberg, Hubbard and Palia (1999) d Holderness, Kroszner and Sheehan (1999) Demsetz and Villalonga (2001) Table 2 Cont. Control Variables Used by Previous Studies in Modelling the Relationship Between Ownership Structure and Corporate Performance* Firm Size (ln(sales) and (ln(sales)) 2 ) (Total (Av. BV h Financial Leverage (Debt / (Av. debt / BV 14 Liquidity Total Risk of the Firm Market Risk Firm-Specific Risk (σ of idiosyncratic stock price risk) e (std. error from market model). (β from regression of stock returns on market returns) (std. error of market risk measure) Advertising Expenditure f (Advertising / PP&E) R & D Expenditure f (R&D / PP&E) Capital Expenditure g (Capex / PP&E expenditure) Industry Membership (Dummies based on SIC codes) (Av. (advertising / sales)) h (Av. (R&D/sales)) h (Av. (PP&E expenditure / sales)) h (Dummies for utilities, media and financial firms) te: * The table provides a summary of control variables used in various tests during each paper. In many instances, only a fraction of all variables listed for a paper were fitted in any given model. netheless, the table provides a useful summary of variables considered in modelling the relationship between ownership structure and corporate performance. NB: RC = Replacement cost; BV = Book value; and, MV = Market value; SIC Codes = Standard Industrial Classification Codes; Capex = Capital expenditure; a. Loderer and Martin s (1997) study uses acquisition data to examine the relationship between ownership structure and corporate performance. Therefore, in addition to the variables listed, the authors also include a dummy variable indicating whether acquisitions are financed by stock. b. Cash flows in the Cho (1998) study are defined as after-tax income + depreciation + amortisation. c. Profit is defined as profit before extraordinary items. d. Himmelberg et al. (1999) also include the (operating income / sales) in modelling as a proxy for a firm s market power as well as a measure of cash flows stemming from operations. e. This value was calculated as the standard error of the residuals from a CAPM model estimated using daily data. f. Himmelberg et al. (1999) include dummy variables to indicate whether firms separately disclosed their R&D and advertising expenditures. g. Himmelberg et al. (1999) also include (PP&E / Sales) and (PP&E / Sales) 2 to measure reduction in agency costs resulting from the fact these assets are easily monitored. h. Averages are calculated using 5 years of data. Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE 293

8 AUSTRALIAN JOURNAL OF MANAGEMENT December 2003 Further, modelling should control for firm leverage, as pecking order theory predicts a negative correlation between a firm s debt levels and corporate performance. 3 Also, firm leverage provides a measure of monitoring provided by credit providers, which may reduce the need for additional monitoring provided by concentrated ownership. The inclusion of a measure of firm-specific risk takes account of the fact that there are different levels of risk associated with investing in different companies. It is possible that ownership concentration may vary in line with the level of firmspecific risk. Also, as Demsetz and Villalonga (2001, p. 222) note higher levels of market and firm-specific risk indicate better prospects to profit form the use of inside information, and, therefore, a stronger causation effect that runs from variations in expected firm performance to variations in management shareholdings. In modelling the relationship between ownership structure and corporate performance, it is also necessary to control for firm size to account for the possibility that performance and ownership are related through the size of the firm. More specifically, it is quite possible that either top shareholders or insiders are able to obtain larger fractions of shares in smaller firms, but firm size and Tobin s Q could be expected to be negatively related to one another. The inclusion of utility and finance industry dummy variables is important, as they control for the possibility of spurious correlation between ownership structure and corporate performance that stems from industry effects (e.g. Demsetz & Lehn 1985). This correlation may result from the relative advantage to investors of obtaining large stakes in a firm caused by regulations within a given industry that may serve to limit how shareholders can use the firm s assets. Finally, the inclusion of a media dummy variable reflects the fact it is an industry that displays what Demsetz and Lehn (1985) refer to as amenity potential, or investor utility additional to that generated by profitability. They argue that given the existence of this amenity potential, they would expect ownership within media firms to be more concentrated, all other factors constant. The econometric model developed by Demsetz and Villalonga (2001) comprises two equations. These equations are also tested in the current paper, and are formally presented below: Q = β + β MSO + β TOP5 + β RDSALE + β FIXSALE + β DEBTASSET β UTIL + β MED + β FIN (1) where: Q = the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(year-end book value of debt + yearend market value of equity) / year-end book value of assets]; 3. Morck, Schleifer and Vishny (1988) explain that leverage also captures the value of corporate tax shields that could result in higher values of performance indicators, including Tobin s Q. However, as a result of the imputation tax credit system introduced into Australia in 1987, corporate tax effectively a prepayment of personal taxes and, therefore, the value of corporate leverage-induced tax shields is zero in the Australian context. 294

9 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE MSO = log (RAWMSO / (100 RAWMSO)), where RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; TOP5 = log (RAWTOP5 / (100 RAWTOP5)), where RAWTOP5 is the percentage of ordinary shares owned by the top 5 shareholders in 2000; RDSALE = the average ratio of annual research and development expenditure to annual sales, calculated over 2 years (1999 and 2000); FIXSALE = the average ratio of the change in gross fixed assets to annual sales, calculated over 2 years (1999 and 2000); DEBTASSET= the average ratio of year-end debt to the year-end book value of assets, calculated over 2 years (1999 and 2000); UTIL = a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise) 4 ; MED = a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise) 4 ; and FIN = a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise). 4 MSO = β 0 + β 1 DEBTASSET + β 2 UTIL + β 3 MED + β 4 FIN + β 5 Q + β 6 MKTRISK + β 7 FIRMRISK + β 8 ASSET (2) where: MSO = log (RAWMSO / (100 RAWMSO)), where RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; DEBTASSET=the average ratio of year-end debt to the year-end book value of assets, calculated over 2 years (1999 and 2000); UTIL = a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise) 5 ; MED = a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise) 5 ; FIN = a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise) 5 ; Q = the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(year-end book value of debt + yearend market value of equity) / year-end book value of assets]; 4. Industry classifications were based on the Global Industry Classification Standard ( GICS ) as at 11th October Ibid. 295

10 AUSTRALIAN JOURNAL OF MANAGEMENT December 2003 MKTRISK = the beta coefficient obtained from a regression of monthly stock returns on monthly market returns using stock price data from July 1998 to June 2002 inclusive 6 ; FIRMRISK = the standard error obtained from the regression used to estimate MKTRISK; and ASSET = the average year-end book value of assets calculated across 1999 and Consistent with Demsetz and Villalonga (2001), the maximum likelihood estimates of β 1, β 2,, β 8 for both equations (1) and (2) are obtained through both ordinary least squares and, in order to account for endogeneity, 2-stage least squares approaches. Results are reported in sections 5.1 and 5.2. These sections also consider the robustness of these results to the use of average accounting profit rate ( PRATE ), which is calculated as the average across 1999 and 2000 of annual net income to the year-end book value of equity, instead of average Tobin s Q. Examination of table 1 shows that previous research has also found some support for a non-linear relationship between ownership structure and corporate performance. Given this, a general non-linear model specification that nests the linear, quadratic and piecewise models fitted in previous research is considered. The models employed previously are considered in detail in tables 1 and 2. The added benefit of the model fit in the current paper is that it affords continuity and fits a very general non-linear specification. The model is formally defined as follows, with results reported in section 5.3. While Bayesian variable selection could be employed to select the knot points included in the regression 7, the inclusion of knots used in earlier research allows us to nest their specifications in our model. The model is also estimated using accounting rate of return as an alternate performance measure: Q = β + β RDSALE + β FIXSALE + β DEBTASSET + β UTIL + β MED β FIN + β RAWMSO + β RAWMSO + β RAWMSO β KNOT5 + β KNOT (3) where: Q = the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(year-end book value of debt + yearend market value of equity) / year-end book value of assets]; RDSALE = the average ratio of annual research and development expenditure to annual sales, calculated over 2 years (1999 and 2000); FIXSALE = the average ratio of the change in gross fixed assets to annual sales, calculated over 2 years (1999 and 2000); DEBTASSET= the average ratio of year-end debt to the year-end book value of assets, calculated over 2 years (1999 and 2000); 6. Beta estimates were calculated using 4 years of monthly return figures from July 1998 to June The Risk Management Service in the Centre for Research in Finance at the AGSM provides two types of beta estimates: one calculated using a standard ordinary least squares approach; and, the other calculated using a Scholes and Williams (1977) approach to account for the existence of thin trading. As betas were calculated using monthly observations, it was decided that it was not necessary to use estimates that had been adjusted for thin trading. 7. Smith and Kohn (1996). 296

11 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE UTIL = a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise) 8 ; MED = a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise) 8 ; and, FIN = a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise) 8 ; RAWMSO = the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; RAWMSO 2 = RAWMSO squared; RAWMSO 3 = RAWMSO cubed; KNOT5 3 = max (RAWMSO k 1, 0) 3, where k 1 represents a knot point at 5%. The 5% knot point is taken directly from the work of Morck, Schleifer and Vishny (1988), among others; and, KNOT25 3 = max (RAWMSO k 2, 0) 3, where k 2 represents a knot point at 25%. The 25% knot point is also based on the work of authors including Morck, Schleifer and Vishny (1988). 4. Data The sample utilised in this study comprises data for 114 public companies listed on the Australian Stock Exchange. Data was collected for these companies in respect of the period 1999 to 2000 inclusive, with the sample including every company for which all data was available. Accounting information was collected from Compustat, ownership data was obtained from Connect 4, share price information was downloaded from Datastream, and market risk measures as well as information used to calculated measures of firm-specific risk were obtained from the Risk Measurement Service within the Centre for Research in Finance at the Australian Graduate School of Management. Table 3 provides an overview of characteristics of companies forming part of the final sample, and shows that the sample includes companies of different size and industry membership. Summary statistics for variables collected in respect of each company forming part of the final sample are presented in table 4, below. Table 5 provides a summary of correlations between variables collected for each company. Examination of this table reveals nothing of concern. 5. Results 5.1 OLS Results Table 6 presents the results for the Demsetz and Villalonga (2001) replication employing an ordinary least squares methodology. Most notably, results show that performance, as measured by Q, is not statistically dependent on either measure of ownership. Further, ownership is not statistically dependent on Q. However, when accounting rate of return is employed as the performance measure, performance is 8. Industry classifications were based on the Global Industry Classification Standard ( GICS ) for Australia as at 11th October

12 AUSTRALIAN JOURNAL OF MANAGEMENT December 2003 dependent on the level of managerial share ownership and vice versa. The positive coefficient on the managerial share ownership coefficient in the regression where performance is measured by accounting rate of return is consistent with the idea that, as managerial share ownership increases, management has greater incentive to maximise firm performance. These results are somewhat inconsistent with Demsetz and Villalonga (2001) who report that both measures of performance are always statistically dependent on at least one of the two ownership measures and ownership, but the reverse is not true. Table 3 Characteristics of Companies Forming Part of the Final Sample Panel A: Firm Size a < $50 million $50 million $100 million $100 million $500 million $500 million $1 billion $1 billion $5 billion > $5 billion Panel B: Industry Membership b Energy Materials c Industrials d Consumer Discretionary e Consumer Staple f Health Financials h Information Care g Technology i Telecommm unications Utilities te: a Firm size is based on average market capitalisation at firms reporting dates in 1999 and b Sector classification is based on the Global Industry Classification Standard ( GICS ). c Materials includes firms in the following classes: chemicals; construction materials; diversified metals and mining; gold; precious metals and minerals; steel and aluminium; and, paper, forest products and packaging. d Industrials includes firms in the following classes: building products; construction and engineering; machinery; conglomerate and other capital goods; commercial services and supplies; and, transportation. e Consumer discretionary includes firms in the following classes: automobile and components; consumer durables and apparel; hotels, restaurants and leisure; media; and, retailing. f Consumer staple includes firms in the following classes: food and drug retailing; beverages; and, food, other products and tobacco. g Health care includes firms in the following classes: health care equipment and supplies; health care providers and services; and, pharmaceuticals and biotechnology. h Financials includes firms in the following classes: banks; diversified financials; insurance; real estate investment trusts; and, real estate management and development. i Information technology includes firms in the following classes: internet software and services; IT consulting and services; software; and, technology hardware and equipment. Examination of table 6 also reveals that, irrespective of the performance measure employed, the ratio of debt to assets is a negative and significant predictor of firm performance. This is consistent with the idea that, as debt rises, so too do the costs associated with servicing it and, as a result, firm performance declines. Further, when performance is measured by Q, the ratio of research and development expenditure to sales is a statistically significant positive predictor of performance. 298

13 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE Conversely, Q is negatively statistically dependent the ratio of the change in fixed gross assets to sales. Results for equation (2) also suggest that firm industry is important in explaining variation in corporate ownership structures, with both MED and FIN coefficients statistically significant. The positive sign on both these industry coefficients suggests that, all else constant, ownership is more concentrated in media and financial firms relative to companies operating in other industries. These results are consistent with the idea of amenity potential advanced by Demsetz and Lehn (1985). Table 4 Summary Statistics for Variables Used in Modelling the Relationship Between Ownership Structure and Corporate Performance The notation used in the table below is defined as follows: Q is the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(book value of debt + market value of equity)/book value of assets]; PRATE is the average of annual profit rates for 1999 and Annual profit rates are calculated as (net income / book value of equity); RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; MSO is the log of (RAWMSO / (100 RAWMSO)); RAWTOP5 is the percentage of ordinary shares owned by the top 5 shareholders in 2000; TOP5 is the log of (RAWTOP5 / (100 RAWTOP5)); RDSALE is the average ratio of annual research and development expenditure to annual sales, and is calculated as the average of the 1999 and 2000 ratios; FIXSALE is the average ratio of the change in gross fixed assets to annual sales, and is calculated as the average of the 1999 and 2000 ratios; DEBTASSET is the average ratio of debt to the book value of assets, and is calculated as the average of the 1999 and 2000 ratios; MKTRISK is the β coefficient obtained from a regression of monthly stock returns on monthly market returns; FIRMRISK is the standard error of the β estimate obtained as the measure of MKTRISK; and, ASSET is the average book values of assets across 1999 and Variable Mean Standard Deviation 25 th Percentile Median 75 th Percentile Q PRATE RAWMSO MSO RAWTOP TOP RDSALE FIXSALE DEBTASSET MKTRISK FIRMRISK ASSET ($ million)

14 300 Q Table 5 Correlation Matrix The notation used in the table below is defined as follows: Q is the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(book value of debt + market value of equity)/book value of assets]; PRATE is the average of annual profit rates for 1999 and Annual profit rates are calculated as (net income / book value of equity); RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; MSO is the log of (RAWMSO / (100 RAWMSO)); RAWTOP5 is the percentage of ordinary shares owned by the top 5 shareholders in 2000; TOP5 is the log of (RAWTOP5 / (100 RAWTOP5)); RDSALE is the average ratio of annual research and development expenditure to annual sales, and is calculated as the average of the 1999 and 2000 ratios; FIXSALE is the average ratio of the change in gross fixed assets to annual sales, and is calculated as the average of the 1999 and 2000 ratios; DEBTASSET is the average ratio of debt to the book value of assets, and is calculated as the average of the 1999 and 2000 ratios; MKTRISK is the β coefficient obtained from a regression of monthly stock returns on monthly market returns; FIRMRISK is the standard error of the β estimate obtained as the measure of MKTRISK; and, ASSET is the average book values of assets across 1999 and Q 1.00 PRATE RAWMSO PRATE RAWMSO MSO RAWTOP5 TOP5 RDSALE FIXSALE DEBTASSET MKTRISK FIRMRISK MSO RAWTOP TOP RDSALE FIXSALE DEBTASSET MKTRISK FIRMRISK ASSET AUSTRALIAN JOURNAL OF MANAGEMENT December

15 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE Table 6 Regression Results for the Demsetz and Villaonga (2001) Replication Where Performance is Measured by Average Tobin s Q 9 The table below presents results for regressions (1) through (2) using Q as the measure of firm performance. Q is defined as the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(yearend book value of debt + year-end market value of equity) / year-end book value of assets]. Results are included for ordinary least squares ( OLS ) and 2-stage ordinary least squares ( 2-SLS ) model fits. t-statistics are included in parentheses below coefficient values. The notation is defined as follows: Q is the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(book value of debt + market value of equity)/book value of assets]; MSO is log (RAWMSO / (100 RAWMSO)), where RAWMSO is the average yearend percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; TOP5 is log (RAWTOP5 / (100 RAWTOP5)), where RAWTOP5 is the percentage of ordinary shares owned by the top 5 shareholders in 2000; RDSALE is the average ratio of annual research and development expenditure to annual sales, and is calculated as the average of the 1999 and 2000 ratios; FIXSALE is the average ratio of the change in gross fixed assets to annual sales, and is calculated as the average of the 1999 and 2000 ratios; DEBTASSET is the average ratio of debt to the book value of assets, and is calculated as the average of the 1999 and 2000 ratios; UTIL is a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise); MED a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise); FIN is a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise); MKTRISK is the β coefficient obtained from a regression of monthly stock returns on monthly market returns; FIRMRISK is the standard error of the β estimate obtained as the measure of MKTRISK; and, ASSET is the average book values of assets across 1999 and Performance Ownership OLS 2SLS OLS 2SLS (Intercept) (4.6685***) (1.9839**) (Intercept) ( ***) ( ***) MSO (0.4995) (0.1277) DEBTASSET ( *) ( *) TOP (0.0338) (0.0308) UTIL (1.1687) (1.1661) RDSALE (3.4825***) (3.4774***) MED (1.7347*) (1.7327*) FIXSALE ( ***) ( ***) FIN (1.7351*) (1.7265*) DEBTASSET ( *) ( ) Q (0.5417) (0.2073) UTIL (0.3965) (0.3591) MKTRISK ( ) ( ) MED (0.0913) (0.0730) FIRMRISK (0.1186) (0.1186) FIN ( ) ( ) ASSET ( **) ( **) F-stat *** *** F-stat ** ** Multiple R Multiple R Adjusted R Adjusted R te: * significant at the 10% level; ** significant at the 5% level; and *** significant at the 1% level. 9. Regression analysis revealed several leverage points of concern. Models were refit excluding these observations, however, results did not differ substantially. 301

16 AUSTRALIAN JOURNAL OF MANAGEMENT December 2003 Finally, irrespective of the performance measure employed, results suggest that ownership is statistically dependent on firm size. The negative firm size coefficient is consistent with the idea that, as firms become larger, ownership concentration decreases, as shareholders have to invest greater amounts to obtain a given level of shareholdings. 5.2 Two-Stage Least Squares Results The results for the Demsetz and Villalonga (2001) replication using 2-stage least squares are reported in table 7. Results show that, once the endogeneity of ownership structure has been accounted for, ownership concentration is not significant in explaining firm performance. This outcome is consistent with the findings of Demsetz and Villalonga (2001), and provides evidence of the endogeneity of ownership structure within the Australian context. With the exception of these changes, results of the 2-stage least square testing are similar to those obtained for ordinary least squares. 5.3 Results of General n-linear Model Fit Results for the general non-linear model specification (equation (3)) are reported in Table 8. Examination of the table shows that the results are similar to those for the ordinary least squares fit of equation (1). The only notable difference is that the newly included variable accounting for a knot point at managerial ownership of 25% is negative and statistically significant. This result is inconsistent with the findings of Morck, Schleifer and Vishny (1988) who report a statistically significant positive relationship for managerial share ownership in excess of 25% and firm performance. However, similar to Morck, Schleifer and Vishny (1988), our results are not robust to the use of accounting profit rate as an alternative measure of firm performance. When PRATE is employed as the performance proxy, neither knot point variable is statistically significant. Instead, the ratio of debt to assets is a statistically significant negative predictor of performance. This is consistent with the idea that the increase in interests costs associated with higher levels of debt leads to reduced corporate profitability. 6. Conclusion The current study was primarily motivated by a lack of evidence regarding the relationship between ownership structure and corporate performance in Australian companies. More specifically, previous research has not considered the relationship between these two variables when ownership is modelled as multi-dimensional and endogenously determined. Results suggest that the endogeneity problem documented in the American context is also present in Australia, with ordinaryleast-squares testing finding that performance is statistically dependent on managerial ownership. However, when endogeneity is taken into account, performance exhibits no statistical dependence on either ownership measure. Finally, previous research has provided evidence that the relationship between managerial share ownership and firm performance may be a non-linear one. To test 302

17 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE Table 7 Regression Results for the Demsetz and Villaonga (2001) Replication Where Performance is Measured by Average Profit Rate The table below presents results for regressions (1) through (2) using PRATE as the measure of firm performance. PRATE is defined as the average of annual profit rates for 1999 and Annual profit rates are calculated as (net income / book value of equity). Results are included for ordinary least squares ( OLS ) and 2-stage ordinary least squares ( 2-SLS ) model fits. t-statistics are included in parentheses below coefficient values. The notation is defined as follows: PRATE is the average of annual profit rates for 1999 and Annual profit rates are calculated as (net income / book value of equity); MSO is log (RAWMSO / (100 RAWMSO)), where RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; TOP5 is log (RAWTOP5 / (100 RAWTOP5)), where RAWTOP5 is the percentage of ordinary shares owned by the top 5 shareholders in 2000; RDSALE is the average ratio of annual research and development expenditure to annual sales, and is calculated as the average of the 1999 and 2000 ratios; FIXSALE is the average ratio of the change in gross fixed assets to annual sales, and is calculated as the average of the 1999 and 2000 ratios; DEBTASSET is the average ratio of debt to the book value of assets, and is calculated as the average of the 1999 and 2000 ratios; UTIL is a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise); MED a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise); FIN is a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise); MKTRISK is the β coefficient obtained from a regression of monthly stock returns on monthly market returns; FIRMRISK is the standard error of the β estimate obtained as the measure of MKTRISK; and, ASSET is the average book values of assets across 1999 and (Intercept) (4.3601***) MSO (1.9122*) TOP (0.1367) RDSALE ( ) FIXSALE ( ) DEBTASSET ( ***) UTIL (0.5786) MED (0.3331) FIN ( ) Performance Ownership OLS 2SLS OLS 2SLS ( ) ( ) ( ) ( ) ( *) ( ***) (1.3981) (1.4445) (0.2111) (Intercept) ( ***) DEBTASSET ( ) UTIL (1.2463) MED (2.0291**) FIN (2.2497**) PRATE (2.7872***) MKTRISK ( ) FIRMRISK (1.6249) ASSET ( **) ( ) ( ) (1.1471) (1.5292) (1.2361) ( ) ( ) ( ) ( **) F-stat ** ** F-stat *** ** Multiple R Multiple R Adjusted R Adjusted R te: * significant at the 10% level; ** significant at the 5% level; and *** significant at the 1% level. 303

18 AUSTRALIAN JOURNAL OF MANAGEMENT December 2003 Table 8 Results of the General n-linear Model Specification The table below presents results for regression (3) using Q and PRATE as alternate measures of firm performance. Q is defined as the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(year-end book value of debt + year-end market value of equity) / year-end book value of assets]. PRATE is defined as the average of annual profit rates for 1999 and Annual profit rates are calculated as (net income / book value of equity). t-statistics are included in parentheses below coefficient values. The notation is defined as follows: Q is the average of annual Tobin s Q values for 1999 and Annual Tobin s Qs are calculated as [(book value of debt + market value of equity)/book value of assets]; RDSALE is the average ratio of annual research and development expenditure to annual sales, and is calculated as the average of the 1999 and 2000 ratios; FIXSALE is the average ratio of the change in gross fixed assets to annual sales, and is calculated as the average of the 1999 and 2000 ratios; DEBTASSET is the average ratio of debt to the book value of assets, and is calculated as the average of the 1999 and 2000 ratios; UTIL is a utility company indicator variable (which is equal to 1 if the firm operates in the utility industry, and 0 otherwise); MED a media company indicator variable (which is equal to 1 if the firm operates in the media industry, and 0 otherwise); FIN is a finance company indicator variable (which is equal to 1 if the firm operates in the finance industry, and 0 otherwise); RAWMSO is the average year-end percentage of ordinary shares owned by top management and the board calculated over 1999 and 2000; RAWMSO 2 is equal to RAWMSO squared; RAWMSO 3 is equal to RAWMSO cubed; KNOT5 3 is equal to max (RAWMSO 5,0) 3 ; and, KNOT25 3 is equal to max (RAWMSO 25,0) 3. Q PRATE (Intercept) (4.0285***) (2.4505**) RDSALE (3.4794***) ( ) FIXSALE ( ***) ( ) DEBTASSET ( *) ( ***) UTIL (0.3494) (0.5786) MED (0.0145) (0.5807) FIN (0.2171) ( ) RAWMSO ( ) (0.4022) RAWMSO (1.1624) ( ) RAWMSO ( ) (0.3564) KNOT (1.2677) ( ) KNOT ( **) (0.5246) F-stat *** Multiple R Adjusted R te: * significant at the 10% level; ** significant at the 5% level; and *** significant at the 1% level. 304

19 Vol. 28,. 3 Welch: OWNERSHIP STRUCTURE AND PERFORMANCE this result, we fit a generalised nonlinear model that nests models advanced in earlier research. Results provide limited evidence of a nonlinear relationship between managerial share ownership and firm performance. (Date of receipt of final transcript: May 29, Accepted by Garry Twite & Doug Foster, Area Editors.) References Berle, A. & Means, G. 1932, The Modern Corporation and Private Property, The Macmillan Company, New York. Cho, M-H. 1998, Ownership structure, investment and the corporate value: An empirical analysis, Journal of Financial Economics, vol. 47, pp Craswell, A.T., Taylor, S.L. & Saywell, R.A. 1997, Ownership structure and corporate performance: Australian evidence, Pacific-Basin Finance Journal, vol. 5, pp Demsetz, H. 1983, The structure of ownership and the theory of the firm, Journal of Law and Economics, vol. 26, pp Demsetz, H. & Lehn, K. 1985, The structure of corporate ownership: Causes and consequences, Journal of Political Economy, vol. 93, pp Demsetz, H. & Villalonga, B. 2001, Ownership structure and corporate performance, Journal of Corporate Finance, vol. 7, pp Hermalin, B. & Weisbach, M. 1991, The effects of board composition and direct incentives on firm performance, Financial Management, vol. 20, pp Himmelberg, C., Hubbard, R.G. & Palia, D. 1999, Understanding the determinants of managerial ownership and the link between ownership and performance, Journal of Financial Economics, vol. 53, pp Holderness, C., Kroszner, R. & Sheehan, D. 1999, Were the good old days that good? Evolution of managerial stock ownership and corporate governance since the great depression, Journal of Finance, vol. 54, pp Loderer, C. & Martin, K. 1997, Executive stock ownership and performance: Tracking faint traces, Journal of Financial Economics, vol. 45, pp McConnell, J. & Servaes, H. 1990, Additional evidence on equity ownership and corporate Value, Journal of Financial Economics, vol. 27, pp Morck, R., Shleifer, A. & Vishny, R. 1988, Management ownership and market valuation: An empirical analysis, Journal of Financial Economics, vol. 20, pp Scholes, M. & Williams, J. 1977, Estimating betas from non-synchronous data, Journal of Financial Economics, vol. 5, pp Smith, M. & Kohn, R. 1996, nparametric regression using Bayesian variable selection, Journal of Econometrics, vol. 75, pp