How Useful are Commercial Corporate Governance Ratings in Emerging Markets?

A central issue in evaluating the effects of corporate governance (CG) is how to measure it. Some researchers measure firm-level CG using country-specific indices (CSIs), tailored to each country’s laws and institutions; several studies report that these indices can predict Tobin’s q in emerging markets, in a panel data framework with firm fixed effects. In contrast, commercial CG ratings (CCGRs) apply the same or similar elements across many countries. However, their power to predict relevant outcomes is not known. In our paper, How Useful are Commercial Corporate Governance Ratings in Emerging Markets?, we assess the three best available CCGRs that cover emerging markets over a reasonable time period, Asset4, Thomson Reuters, and MSCI. We find that these ratings have no power to predict Tobin’s q or profitability. We also provide suggestive evidence that the likely root cause is poor construction of the ratings, rather than CG’s inability to predict Tobin’s q.

A substantial body of CG research studies the extent to which firm-level CG choices, often captured in CG indices, predict firm value, profitability and other outcomes. This research involves whether “better” CG has a payoff in firm performance and which aspects of governance are indeed better, as well as what firm attributes predict governance. This research is important because it can guide firm choices of which governance measures to adopt, and investor decisions on which governance measures to support, and which firms to invest in. This research is necessarily conducted at the firm level. It can be conducted in individual countries, or across multiple countries, in both developed and emerging markets (EMs).

A related body of research, deriving from La Porta, Lopez-de-Silanes, Shleifer and Vishny (1998) studies the effects of country-level CG rules. Studies of the effects of country-level rules often rely on natural experiments involving changes in rules. But for firm-level choices natural experiments are, almost by definition, not available. One must instead rely on panel data, ideally with firm fixed effects, to study either individual governance elements (e.g., does a firm have an audit committee or a majority of independent directors) or broader indices which seek to capture the combined effect of multiple individual governance measures.

We contribute here to the literature on firm-level CG indices, by studying the relative value of CSIs, constructed by researchers, versus CCGRs. Both approaches have been used in prior work. Each approach has potential advantages. CSIs can be tailored to address country-specific rules and customs, they can rely only on objective elements which the researchers believe are related to the imperfectly defined underlying concept of governance, and data can be collected for a broad range of firms in a given country. However, CSIs are expensive and time consuming to construct and collect the data for. Moreover, because the indices are country-specific, generalizability of results to other countries is unclear.

Multicountry CCGRs are a potentially attractive alternative. They can cover a large number of countries over a substantial time period, although usually only the largest firms in each country. If CCGRs perform well, there might not be sufficient extra value from CSIs to justify building them. However, the available CCGRs have potential downsides. Often, they include subjective elements. How they are built is often not transparent—indeed, their commercial value depends on the CCGR not being easily replicable. Bond ratings face similar concerns but the value of the ratings can be assessed ex post by assessing how well they predict future default. CCGRs lack a natural ex post outcome, similar to default, that can be studied. The best one can do is assess the evidence for a basic property that a good CG measure should have—it should be correlated with firm value. One can potentially study other outcomes, such as profitability or share liquidity.

There is, however, a joint hypothesis problem in testing the value of CSIs and CCGRs. One is jointly testing whether CG predicts firm value (or another outcome) and the construct validity question of whether the index does a good job of capturing important aspects of unobserved underlying governance (Black et al., 2017).

There is also a power issue. To observe a relationship between CG and firm value, one needs sufficient cross-sectional variation in CG in the sample. To use a classic panel data design with two way (firm and time) fixed effects (TWFE design), one further needs within-firm variation over time. Yet without firm FE, an observed correlation between CG and firm value could reflect unobserved firm characteristics which are correlated with both CG and firm value, rather than a causal relationship. The power concern is greater in developed countries, which generally have strong CG laws and limited variation both across firms and across time within firm. Limited power may explain why it has been challenging for researchers to find a robust correlation between CG and value in developed countries, the U.S. in particular. Compare the null results from, e.g., Daines, Gow, and Larcker (2010); Koehn and Ueng (2005); Bhagat, Bolton, and Romano (2008); and Ertugrul and Hedge (2009) with Ammann et al. (2011) (positive association in multicountry study between Governance Metrics International CCGR and firm value); and Guest and Nerino (2020) (event study of downgrades of U.S. firms using the ISS CCGR).

EMs may offer more fruitful ground for studying the impact of firm-level CG choices, for a number of reasons. Minimum standards, set by local rules, are often low, allowing for more firm-level variation. Also, as firms and countries develop, and firms face investor pressure to improve their governance, firm choices can vary substantially over time. Moreover, unlike the situation in developed markets, a well-constructed CSI can predict firm value with a TWFE design (e.g., Black et al., 2014, Cheung et al., 2011). Thus, if a CCGR fails to predict firm value, the likely cause is lack of construct validity for the CCGR, not with whether CG, if measured in a manner that reflects country-specific rules and norms, can predict value. However, the predictive power of CCGRs for EMs has previously been studied only in cross-section, and involves older indices which are no longer available (e.g., Klapper and Love, 2004; Durnev and Kim, 2005).

How well CCGRs perform in EMs is an empirical question, that we study here. But we note initially that there are reasons for concern about CCGR quality. First, their elements of the available CCGRs are often US-centric and may poorly capture variation in governance in EMs. For example, the available CCGRs do not measure firm-level disclosure, perhaps because U.S. disclosure standards are relatively high, leaving less room for cross-sectional variation. Second, many U.S. firms have dispersed ownership, leading to different agency problems than in EMs, where most firms have a controlling family or group, which may need different CG strategies (Bebchuk and Hamdani, 2009).

This paper investigates two basic questions with regard to CCGRs in EMs. First, do they predict firm value (measured by Tobin’s q) or profitability (measured by EBIT/assets). We assess the value of the three available CCGRs—Asset4, Thomson-Reuters (TR), and Morgan Stanley Capital International (MSCI)—which cover a substantial number of EMs for significant time periods. We find no evidence that any of these ratings predict firm value or profitability in a TWFE framework, either across countries or in individual countries. We also find no predictive value using year FE and firm random effects (firm RE).

Second, we investigate the joint hypothesis issue, which has not been addressed in the prior CG literature. We provide evidence that the likely root cause of the failure of the CCGRs is poor construction of the ratings, rather than the inability of a well-specified measure to predict Tobin’s q. We compare the power of the CCGRs to predict Tobin’s q to the predictive power of the CSIs from Black et al. (2014, below BCKKY), for firm-year observations common to both samples, from Brazil, India, Korea, and Turkey. BCKKY show that their CSIs are known to have power to predict Tobin’s q with TWFE in India, Korea, and Turkey, and with firm RE in Brazil. Despite the small size of the overlap samples, the BCKKY CSIs retain reasonable predictive power. Yet, for identical samples, the CCGRs have no predictive power.

We also investigate the correlation between different CCGRs, and, for overlapping samples, the correlation between the CCGRs and the BCKKY CSIs. The Asset4 and TR ratings correlate moderately with each other but this correlation is mechanical, since the TR index was built on the Asset4 index after Thomson-Reuters purchased Asset4. However, the Asset4 and TR ratings correlate poorly with the MSCI rating, and all three ratings correlate poorly with the BCKKY CSIs for the overlapping firm-year observations. This low correlation further suggests that the CCGRs’ lack of predictive power likely derives from their poor index construction.

Whatever the explanations, the bottom line is clear: across all three CCGRs, and many robustness checks, the coefficients on the CCGRs remain insignificant when pooled across countries, with occasional significance in individual countries but no apparent pattern in the country-specific results. Thus, the available CCGRs for EMs are not currently useful for analysis of the value of governance. Nor are they useful as benchmarks against which firms can assess their own governance or investors can push firms to change their governance.

The full paper is available for download here.