You are on page 1of 16

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

AN APPLIED ECONOMETRICIANS VIEW OF EMPIRICAL CORPORATE GOVERNANCE STUDIES BY AXEL BORSCH SUPAN AND JENS KOKEA CRITICAL REVIEW PART A INTRODUCTION(I) AUTHORSThe article under review is authored by veteran researchers, Axel Boorsh Supan and Jens Koke. Axel Supan is Professor of Macroeconomics and Public Policy at the University of Mannheim. He is also the founding Executive Director of the Mannheim Research Institute for the Economics of Aging (MEA) and Research Associate of National Bureau of Economic Research (NBER) and Member of the Board and Research Director at the Centre for European Economic Research (ZEW). Co author, Jens Koke worked as a staff member of ZEW at the time of writing the article. (II) TIMELINESS OF ARTICLEThe paper was published in the third volume of the German Economic Review in 2002. To borrow the authors words, this was a time when the economic analysis of corporate governance was in vogue. One may safely deduce that the trend of analyzing corporate governance issues became the prevalent rage of the times on the back of corporate casualties of titans such as Enron in 2001 and World Com and Global Crossing in 2002.It was this string of corporate casualties which made investors discern the need of analyzing and monitoring firms from the perspective of corporate governance and to cognize that good corporate governance may increase a company's valuation and boost its bottom line in addition to averting disasters. A nearly concurrent publication of a report in 2000 by the reputed global consultancy firm McKinsey which found that 75% of the 200 institutional investors it surveyed regarded board practices as important as financial measures for assessing companies, also established the significance of the study of corporate governance dimensions.

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

But more pertinently, this paper came in the wake of the first broad survey of corporate governance by Shleifer and Vishny (1997) and Kaplan (2001). Preceding studies by Verbeek and Neijman (1996), Zingales (1998), La Porta et al (1997, 1998), Holmstrom (2001), Poland, Grosfeld and Tressel (2001), Grosfeld and Tressel (2001) then became the basis for this seminal paper by Supan et al. Corporate governance entails the keeping of the management in check and ensuring financial disclosure, board independence, and shareholder rights. As corporate governance is a continually evolving process and not a permanent state, there is no universally accepted definition of the term. The Cadbury Committee in 1992 defined it as "the system by which companies are directed and controlled., while the IFC viewed it as "the relationships among the management, Board of Directors, controlling shareholders, minority shareholders and other stakeholders". (III) PURPOSE OF ARTICLESupan et al reviewed the recent empirical literature and research on corporate governance with the dual objectives of specifying the systematic methodological flaws which beleaguered empirical studies and recommending techniques to overcome such problems. The authors assert that efficient corporate governance is imperative as it may lead to a reduction in the agency costs which result from the conflict of interest between the managers and owners of a firm. The cost reduction may in turn boost an individual firms productivity at the micro level and consequently improve the productivity of the entire country at the macro level. Theoretical research in this regard is insufficient to provide a complete understanding of governance issues on its own .It has to be supplemented by empirical evidence so that a comprehensive understanding of governance can be made. In this context it is essential that governance and firm performance variables be measured correctly so as to derive meaningful results. (IV) GENERAL IMPACT OF PAPERThe general impression of the article is that of a thoroughly researched and systematically planned work. The relevance of the article is enhanced by the fact that As Supan et al draw only on recent studies in corporate governance while excluding research of the past. This purposely
2

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

imposed screening criterion helps the paper in keeping a current outlook and assessing only the latest and most prevalent techniques and standards in the empirical field. The paper is logically ordered with the authors diagnosing four major flaws of contemporary empirical research i.e. reverse causality, sample selectivity, missing variables and measurement error in variables. Consequently, the paper is divided into four categories with each category devoted to the analysis and brief of a particular econometric problem and the recommendation of a remedial technique. In conclusion, the authors synthesize their aforementioned solutions to econometric problems to build a framework for future empirical studies which would be devoid of the most common econometric and technical problems and thus yield more reliable and accurate results which could provide a better insight into the nature, magnitude and direction of the relationships between metrics of corporate governance and firm performance. PART B (I) METHODOLOGY OF PAPERThe survey is very well planned with each of the four problems being analyzed in the following framework, i. The first part of every section expounds the econometric problem theoretically and cites examples of papers which were plagued by the same problem. ii. The second part of every section recommends tests for detecting the presence of the problems. iii. The third part provides suggestions as to circumventing these problems in future empirical studies. In view of the systematic framework used by Supan et al one may aver that the study would fulfill the authors stated purpose of detecting and avoiding the said problems in future research as a theoretical analysis of the specified econometric problems is coupled with an applied solution of dealing with them.

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

(II) SUMMARYMost empirical studies try to establish a relationship between measures of corporate governance and firm performance by employing a multiple regression model and regressing proxy variables of the former on the proxy variables of the latter. Most often proxy variables used as metrics of corporate governance / regressands are, a. Ownership structure b. Capital structure c. Board Structure d. Market for corporate control Proxy variables used commonly as metrics of firm performance/ regressors are, a. Equity Value b. Total factor productivity Supan et al establish that empirical studies conducted in the domain of corporate governance have been replete with econometric problems. The authors identify the following four econometric issues as recurrent problems and as those which occur most frequently in empirical research. 1. Firm variables are assumed to be exogenous when they are in fact endogenous. 2. The sample is not truly random. 3. Core variables have been omitted. 4. Variables are measured inaccurately and thus exhibit large errors. The econometric issues are then analytically and technically discussed in four independent sections. (i-a) REVERSE CAUSALITY AND SPURIOUS CORRELATION(i) The primary objective of empirical studies in the field of governance is to establish the relationship between firm performance and a measure of governance which is often proxied by the concentration of ownership. But studies by Jensen and
4

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

Meckling (1976) and Leland and Pyle (1977) have come to opposite conclusions regarding the direction of causality. The former concludes that high management stakes in equity precede better firm performance while the latter deduce that well performing firms lure clusters of investors and managers who invest en bloc in companies which are already well performing; concentrated ownership proceeds better performance. On the basis of the findings of the above cited studies Supan et al infer that the direction of the relationship (ex ante or ex post) between the dependent and independent variables is ambiguous and leads to the presence of the condition of reverse causality. (ii) The authors build upon Himmelberg (1999) to state that both ownership structure (X variable) and firm performance as measured by profitability (Y variable) may be simultaneously influenced by a common intervening variable e.g. market power. The absence of Market Power as determinant of both the dependent and independent variable will lead to spurious correlation, which is a variant of reverse causality. (iii) The presence of structural reverse causality or spurious correlation would render the estimated coefficient of the independent variable biased and inconsistent. However a structural assumption about the nature of the X variable which takes into account the endogeneity of the X variable may resolve this problem. Supan et al expound the correct model as, Z= X + _____ (i)

X= Z + ---(ii) where E()= E () = 0 and Cov (,) = (), = Cov(X,Z) = Cov ( X, X + Var ( X) Var (X) = + Cov (X, ) _______ Var (X) (ii)

Z=Firm Performance, X=Ownership concentration. Cov( x,) = Cov (Z + ,) = Cov[(X+) + , ] = + Cov( X, ) + ---(iiii) If 0 and 0 the estimated will be biased and inconsistent.

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

Ordinary regression models are built on the Gauss Markov assumption that (i) = E( ) ie the population or true is equivalent to the expected , E( .
(ii) = 0 or Cov ( ) = 0 ie. There is no autocorrelation between disturbance terms .

But if this assumption is violated it becomes problematic if a regressor is correlated with the error term or equivalently, the omitted variable both affects the independent variable and separately affects the dependent variable. The problem of endogeneity arises when the factors that are supposed to affect a particular outcome, depend themselves on that outcome. In the above case of firm performance and profitability, a loop of causality between the independent and dependent variables of a model leads to endogeneity. Supan and et al concur with the findings of the preceding critical survey of Himmelberg (1999) that cross sectional data even if it incorporates an intertemporal dimension does not resolve the endoneity problem. This assertion counters Morck (1988) which used a model in the spirit of Demsetz and Lehn (1985) concluded using a large cross sectional dataset of a371 companies that there is no significant relationship between ownership and firm performance. (i-b) IDENTIFYING AND REMEDYING REVERSE CAUSALITYCausality is simply understood as if event A happens before event .B then it is possible that A is causing B but it is impossible that B is causing A.( Koop). Francis Diebold used the term predictive causality to state that if Yi contains useful information for predicting Yj above the histories of other variables in the system ,we can concisely say that Yi causes Yj. Supan et al recommend the classical Granger Causality test (1969) to test for causality. Applying this test would entail running a restricted and unrestricted regression and then testing two hypotheses, Zit= Zit = jZit-J + jZit-j +it j Xit-j + it ---- (Unrestricted Regression) ----- (Restricted Regression)

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

Two null hypotheses would be tested using the residuals from both equations using the F test. (i) (ii) Ho: i = 0 and i=1,2,..n Ho: , = 0

F = (RSS r RSS ur)/m RSS ur / (n-k) Under the accept/reject criterion If F calc F table then accept Ho that variable under

consideration (Z ) does Granger cause the other variable (X) otherwise reject Ho. Because it is being tested whether changes in X precede (cause) changes in Z therefore a regression of Z on X should include its past values. The inclusion of lagged values is I backdrop of using a stationary series of data. Supan et al opine that the use of panel data and models using models which include all relevant firm variable collected over time can help resolve the problem of reverse causality. (ii-a)-SAMPLE SELECTION BIASSupan et al aver that most empirical researchers do not use a truly random sample. Most studies are conducted using data sets composed of the largest and often listed (Atkiengeselleschaft) companies. Incidentally the largest and listed companies are those which are also the most profitable (well performing) firms in the population. In such cases, firms which constitute the sample are those which have already been influenced by the endogenous variable. The consequences of building a sample which consists of only of the largest and publicly traded firms causes results to be biased. Supan et al term this bias as Sample selection bias. It is imperative for results to be reliable that the sample selected should be representative of the entire population and not reflective of a single stratum of the population. According to the properties of the Classical Linear Regression Model (CLRM) or the Gausssian Standard, an estimator of the exogenous or dependent variable will only be unbiased when E(/Xi) ) =0. This is to say that given the value of Xi, when Xi is stochastic, the mean or expected value of the random disturbance term
7

i is zero. When this property holds the true

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

population would be equal to one.But when the sample is already correlated with the independent (Y) variable this property is violated and E ( be 1. 0 and hence the estimated would

When only the largest firms are selected as part of the sample, it amounts to purposely including firms which exceed some hypothetical value Y*.When this happens negative residuals which emanated from low Xi values are omitted causing t he estimated to be exhibit a downward bias or to be underestimated. Whether the bias is under or over estimated depends upon the covariance between and Xi. If this covariance can somehow be measured and then included in the regression, it would solve the problem of getting biased estimators. The following table shows the volume of studies which were beset by a sample bias.

In addition Supan et al observed that the market value of a firm is often taken as a proxy variable for firm performance. (Morck et al 1988, McConell and Servaes 1990).But market value is only
8

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

available for listed companies and usually most listed companies are those which were already the best performing. Once more, the estimators will be biased and inconsistent. Supan et al also noted that most researchers are inclined towards using balanced panel data, but the use of a balanced panel raise the probability of a selection bias, as in a representative sample there should be a sizable number of companies which should exit the market due to insolvency or other reasons and thus lead to a gap in reported data points of market value or other measures of performance. Consequently Supan et categorize the work of Demsetz et prone to be biased. (ii-b) -IDENTIFYING AND REMEDYING SAMPLE SELECTIVITY1a. If a sample does not contain a particular type of firm 9 (non listed or small private firms) then there is no test which can indicate that the sample is biased as it has entirely omitted a particular type of firm. 1b. The only way to deal with such a bias is to include all relevant firms in the data set irrespective of whether their inclusion results in an unbalanced panel or entails the difficulty of keeping track of non listed or small companies which may have exited the market due to insolvency or changed their name during the period under study. 2a.The bias which emanates from using a data set in which the data points of some firms have gradually undergone diminution due to the dropping out of firms owing to bankruptcy or acquisition, may be dealt with econometrically. 2b.The researcher must bear out the reasons as to why data isnt available for particular years. This information would then be used to apply the Heckman selection test (1976).The Heckman Test is applied keeping in view that the sample is truncated, i.e. Y is known only if some criterion defined in terms of Y is met and X variables are observed only if Y is observed. The remedy proposed by Supan et al builds on the proposition that firm performance has a truncated distribution and only firms which exhibited a performance level of Y* are included

in the sample. Performance of firms in the sample is conditional on the probability that the firm is part of the sample, E(|x). If this conditional probability is ignored it would result in the estimators being biased and inconsistent.

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

The recommended sample selection model would be as indicated below, Yit = Xit + it ------ (a) Xit =exogenous characteristic, Yit= Firm Is performance

Exit is described by, Rit = Zit + it Rit is a dummy variable and { or {

The following observation rule would be, Yit = Yit* , Rit =1 if Rit*> 0 and if Yit not observed , Rit =0 if Rit* 0.

Equation (a) would become, Yit = Xit + [{ ( Zit)}/{( Zit )}] + it (iii-a)-MISSING VARIABLESSupan et al categorize the third most commonly made econometric mistake as 1. Model misspecification or using an incorrect functional form for the model or 2. Omitting an interaction tem. When this happens the influence of the missing variable is captured by the stochastic disturbance term . Many studies such as Demsetz and Lehn 1985 use a linear regression model while studies by Jensen and Meckling (1976) and Fama and Jensen (1983) point out that managerial ownership (X) is quadratic (U shaped) in variable. If a polynomial of the 2nd degree is omitted it will create a left out variable bias causing the estimated to be biased and underestimated. This was supported by Mork et al (1988). When an interaction term or any other relevant variable is missing, it means that estimators will be unbiased only when, Y = + X +Z +, = + Cov (X, Z) Var (X) But since Cov (X, Z) is in case of Z being market power, will be underestimated.
10

and,

=0 and Cov (X,) =0

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

(iii-a)-IDENTIFYING AND REMEDYING MISSING VARIABLE BIASThe authors opine that there is no method of econometrically testing for missing variables. But they make two recommendations, 1. Variables must be taking the form (linear or quadratic) which is supported by theory. Otherwise it is advisable to run two regressions, one with a linear and other with a quadratic form. 2. The interaction of proxies for corporate governance mechanisms should be taken into account as some interactions either substitute or complement one another.(McWilliams and Sen 1977) 3. Data which was previously unavailable can now be used to serve as proxies for firm specific variables which were omitted in previous studies, besetting them with a missing variable bias and causing estimators to be inconsistent and biased. (iv-a)- MEASUREMENT ERROR IN VARIABLESEmpirical studies in governance usually explore the linkages between a measure of firm performance and a measure of corporate governance. In this regard it is quite worrying to note that there is no consensus on a variable which may be accepted as the best and most appropriate proxy of firm performance. Consequently, studies used a variety of variables i.e. Tobins q, ROA and ROE as metrics of firm performance. The use of such an assortment of measures of performance has made comparison of the results of such studies difficult to compare and thus their findings cannot be generalized across the board. It is even more disturbing to note that these proxies have a low correlation indicating that they are not perfect substitutes. As there is there disagreement on what is the best measure of firm performance there is a high probability that the proxy chosen for the endogenous/dependent variable will not be an accurate measure of performance and will contain an error in measurement.

11

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

Supan et al conclude that , 1. If the dependent variable (Y) contains an error in measurement, the estimators will not exhibit a bias as will capture the additional variation, yet this will cause estimation

results to be weakened due to a low signal to noise ratio. This is to say that there will be a higher probability of results being insignificant.

Statistical significance can be considered to be the confidence one has in a given result. In a comparison study, it is dependent on the relative difference between the groups compared, the amount of measurement and the noise associated with the measurement. In other words, the confidence one has in a given result being non-random (i.e. it is not a consequence of chance) depends on the signal-to-noise ratio (SNR) and the sample size.

In other words, the dependence of confidence is high if the noise is low and/or the sample size is large and/or the effect size (signal) is large. The confidence of a result (and its associated confidence interval) is not dependent on effect size alone. If the sample size is large and the noise is low a small effect size can be measured with great confidence. If results are insignificant they do not contribute much to existing literature and the purpose of such studies is nullified. 2. If the independent variable has an error of measurement it will lead to biased and inconsistent estimated values. Econometrically this can be expressed as, = Cov (X, Y) Var (X) = Cov (X* + v, X* + Var (X) /x} ,

= Cov (X *+ v , X*- V + ) = Cov ( X* + v ,(X*+v)- v + = { Var (X) Var(X)

If x > v then the X variable has substantial noise and beta will be biased towards zero.

12

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

The authors recommend the improvement of data and application of a structural analysis for circumventing such complications stemming from errors in measurement. (III) STRENGTHS OF THE ARTICLEThe article is useful as, i. It is well written, not difficult to comprehend and on the whole a readable article. This is a great strength for a survey of existing literature. The diction of the article has been chosen with exquisite care. ii. It has been meticulously researched with the authors drawing upon 25 studies as borne out by the numerous citations. Thus the study is very comprehensive and thorough. iii. iv. Almost all studies cited by the authors had used extraordinary large data sets. The results of various studies have been interpreted in consonance e.g. Jensen and Meckling (1976) and Fama and Jensen (1986) to prove that is some variables have a quadratic functional form. This amalgamation of various studies gives the paper an added dimension. v. It has been systematically planned and subdivided into categories according to the econometric problems. vi. It delves into the details of each of the four econometric problems, with the authors citing not only on the major linkages the variables but also interactions which exist between variables eg the interaction between the confounding variable of market competition, ownership structure and profitability. (Bally and Gerbasch 1995). vii. It builds primarily on the properties of the CLRM and exposes truly profound complications which are entailed by a violation of the properties and the consequences of arriving at biased and inconsistent estimators. What is even more surprising is the fact that these properties have been widely violated in contemporary studies. viii. It highlights an immensely important feature that if studies disclosed the assumptions on which proxy variables were selected or if they used a single measure of firm performance, results could be rendered more comparable. ix. The authors recommend the use of simple tests eg the Granger causality or the Hausman test for resolving econometric problems rather than proposing some complex applications.
13

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

x.

The study has an added importance in that such econometric problems are not specific to regression analysis alone but also crop up in simple correlation analyses as well.

(IV) WEAKNESSES OF THE ARTICLEi. Although the authors explicitly state that this study pays particular attention to the institutional setting in Germany, it uses a significant number of studies from USA and UK .It is questionable as to how conclusions can be drawn for Germany by using studies which and have been conducted in a different macroeconomic environment, with varying regulations, tax regimes and incentives for listing on stock exchanges and accounting standards.(Jensen and Meckling 1976, Fama and Jensen 1983) ii. Even though the study is aimed at analyzing contemporary empirical studies there are instances when the authors draw upon studies which are quite old. (Cable 1985, Grossman and Hart 1980) iii. Supan at al repeatedly use the findings of Demsetz and Lehn 1985 for making deduction purposes when they expressly qualify the former for using a cross sectional data set and purposely imposing the restriction that ownership concentration is a linear functional form when it is actually a quadratic form. This amounts to a contradiction in terms. iv. Although Supan et al enunciate that there should be only one proxy for measuring firm performance, they do not recommend a metric themselves. v. Although the study recommends the use of Heckmans test (1976) for overcoming sample selection bias, it qualifies the assertion by stating that the use of Heckmans test may cause problems of its own. Supan et al offer no opinion of dealing with sample biases in such cases. (V) ORIGINALITY OF PAPERThe article is not the first survey of the econometric problems which afflict empirical studies in the domain of corporate governance. It was preceded by another such study by Himmelberg et al (1999) and was concurrent with Himmelberg (2002).However the paper makes an original contribution in that, i. It is a broad analysis made with applicable recommendations for econometric problems. Unlike Himmelberg et al (1999) which discussed only reverse causality
14

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

and sample selectivity , Supan et al make an original contribution by adding the facets of errors in measurement of variables and missing variables to the existing list of econometric problems. ii. The authors forward a logical and a mathematical proof of how errors in measurement of variables and missing variables create econometric problems in empirical studies. Thus the paper under review had a higher value quotient as compared to other studies at the time it was written. (VI) VALIDITY OF PAPERi. The review conducted by Supan et al is valid as it builds upon a plethora of studies with a particular emphasis on the seminal research by Jensen and Meckling (1976) and Fama and Jensen (1986) alongside the most recent empirical studies in corporate governance. ii. Each of the four cited problems is supported by numerous studies which were beset by the specified problems thus providing compelling evidence of the extent to which such problems are encountered in empirical studies and how they distort results. PART C CONCLUSIONSupan et als assertion of contemporary research being beleaguered by the problems of reverse causality, sample selectivity, missing variables and errors in variable measurement has been proved conclusively through the large number of studi.es cited by the authors which were conducted in the presence of the said problems. Thus the recurrent and distortive nature of the cited econometric problems is established very soundly. The paper took into account various dimensions of the proxies for the endogenous and exogenous variables such as their interactions and the impact which confounding variables exerted on them and consequently the consequences for the results which were attained in the presence of such biases.

15

A Critical Review of an Applied Econometricians View of empirical Corporate Governance Studies (2002)

A final evaluation of the paper taking into account its numerous strengths and the few weaknesses, would tend a reader to conclude that this was a meticulously planned and researched survey which would have served as an invaluable addition to then published surveys on empirical econometric problems. In view of the suggestions forwarded and the contribution made by Supan et al, one would presume that future empirical studies would not be as riddled with econometric problems of such profound nature and such serious consequences as the preceding studies were found to be.

16

You might also like