Print
E-MailBernard Black is the Nicholas D. Chabraja Professor at Northwestern University School of Law and Kellogg School of Management. The following post is based on a paper co-authored by Professor Black, Professor Antonio Gledson de Carvalho of Fundacao Getulio Vargas School of Business at Sao Paulo, Professor Vikramaditya Khanna at the University of Michigan, Professor Woochan Kim at Korea University Business School and Professor Burcin Yurtoglu at WHU – Otto Beisheim School of Management. Work from the Program on Corporate Governance about the relationship between corporate governance and firm value includes Learning and the Disappearing Association between Governance and Returns by Lucian Bebchuk, Alma Cohen, and Charles C. Y. Wang (discussed on the Forum here).
There is a vast and growing literature using multi-country studies to examine the effects of corporate governance on firm value. In our paper, Methods for Multicountry Studies of Corporate Governance: Evidence from the BRIKT Countries, forthcoming in the Journal of Econometrics and recently made publicly available on SSRN, we explore the empirical challenges in multicountry studies of the effect of firm-level corporate governance on firm market value, focusing on emerging markets, and propose methods to respond to those challenges. Our study has implications for multicountry studies in other spheres as well.
These kinds of studies face at least three critical problems: construct validity, limited data and endogeneity. Construct validity is central in corporate governance research, yet rarely addressed. A governance index is a construct that imperfectly measures unobserved underlying governance. There is no direct way to quantify the gap between the construct and the underlying concept. Moreover, what matters in corporate governance often depends on local norms and institutions, which vary widely across countries. Simply put, it is difficult to understand governance in emerging markets by studying developed ones.
A second core problem is lack of data on governance, especially lack of consistent, time series data across countries. In many countries the number of firms with governance data is limited and financial databases are often imperfect. Lack of data on governance also reinforces concerns with construct validity. As we will show, it is impossible to use public data to build a broad governance index based on elements common to all countries, even across the five countries we study. It is very difficult to do so even relying on private surveys of firms.
A third problem is endogeneity. Natural experiments can sometimes be found in individual countries, but address limited aspects of governance. The next best approach, and the one we pursue here, is to build panel data and use firm fixed (or at least random) effects, plus extensive control variables, to limit one central endogeneity concern, omitted variable bias (OVB). However, because of limited data, it is difficult to obtain time-series data on governance and detailed data on control variables, which serve to worsen OVB and exacerbate construct validity concerns.
Yet, most of the prior research on emerging markets comes from studies with most or all of these problems. In this article we describe the perils in multicountry studies of corporate governance, and seek to make progress on all three core dimensions mentioned above.
We construct, largely by hand, unique time-series datasets on governance in each of five important emerging markets—Brazil, Russia, India, Korea, and Turkey (“BRIKT” countries). Together, these countries provide a representative sample of the results one might expect in moderately developed emerging markets. Our governance dataset covers many, but far from all, public firms in each country. It is, we believe, close to the best that one can realistically build across multiple emerging markets. We use this rich dataset, plus extensive control variables to offer a more robust approach to understanding how corporate governance affects firm value in emerging markets.
We address construct validity by building country-specific corporate governance indices (CGI) which reflect local norms and institutions. Each is comprised (data permitting) of subindices for board structure, board procedure, disclosure, ownership structure, minority shareholder rights, and control of related party transactions. Each subindex is comprised of one or more “elements” that seek to capture specific aspects of governance that we consider relevant in each country. The indices and subindices for each country are broadly similar, and seek to capture similar underlying governance concepts. But the individual elements reflect the norms, institutions, and data limitations in each country. The new methodology we employ—conducting a multicountry study using similar-but-not-identical country-level governance indices—can be seen as a “middle way” between single-country studies, from which it is hard to generalize; and “massively multicountry” studies, which suffer from the three core problems noted earlier. We develop methods appropriate for (cautiously) generalizing across a number of countries, which may provide a way to develop a more credible connection between governance and firm value or performance.
Using our country CGI, we assess whether governance predicts firm market value (proxied by Tobin’s q) in each country, in firm fixed effects (FE) and random effects (RE) specifications. We find positive coefficients on country CGI in all five countries, which are statistically significant in RE (in all five countries) and in FE (in all but Brazil). We then pool the indices across our countries (except Russia, which we cannot use when pooling), to create a Pooled CGI index. We find strong evidence with both RE and FE that Pooled CGI predicts higher Tobin’s q.
We also generate a “Common Index,” which consists of the 15 elements available in all four countries and useful in at least two of them (we require this because we seek to assess the relationship between governance and Tobin’s q across countries). The Common Index has weak predictive power with RE and none with FE. In regressions including both the Common Index and either Pooled CGI or a “non-common” index built from the remaining elements, the Common Index has no power to predict Tobin’s q. Instead, power comes entirely from the country-specific elements included in the non-common index.
Omitted variable bias is important. In both individual country and pooled regressions, coefficients on CGI are generally higher in weaker designs. This suggests that firm effects are important and that an FE specification is preferred, if feasible. Coefficients are also generally higher with fewer covariates. This provides evidence that to limit omitted variable bias, extensive covariates are important, in addition to firm effects. In multicountry studies that use regressions on pooled data across countries, it is important to interact the covariates with country dummies, thus allowing for country-specific “response surfaces.”
We then assess the sensitivity of our estimates to remaining omitted variable bias using two sets of bounds, adapted respectively from Hosman, Hansen, and Holland (2010) and Altonji, Elder and Taber (2005). These bounds use the sensitivity of coefficient estimates to included covariates to estimate lower bounds on those coefficients under assumptions about the extent of bias from omitted covariates. The lower bounds for country CGI are positive in all five countries and statistically significant in Korea and Russia, as well as for Pooled CGI.
We study here only firm-level governance in emerging markets. But the concerns we raise with common indices also apply to multicountry indices in developed markets such as the Institutional Shareholder Services index (e.g., Aggarwal et al., 2009), widely used indices of anti-director rights and creditor rights (La Porta et al., 1997, 1998), and measures of economic competitiveness (e.g., World Bank, 2013). In all these areas, we face a choice between a common index, whose elements may poorly capture the underlying concept in some countries, and richer, country-specific measures with uncertain generalizability.
The full paper is available for download here.
