The last few decades have seen an explosion of corporate board reforms. Fauver, Hung, Li, and Taboada (2017, “FHLT”) examine the impact of board reforms on the value of firms in 41 countries from 1990 to 2012. Using staggered difference-in-differences (“DiD”) estimates, FHLT find that firm value increases after board reforms. However, the robustness of these results is contested by Baker, Larcker, and Wang (2021, “BLW”) in a study that reviews the latest econometric theory on staggered DiD estimation. BLW report that after applying alternative estimators to address biases identified in recent research, most of the results in FHLT become insignificant.
In a study conducted in response to BLW, we show that the insignificant findings obtained by BLW are largely due to selective modifications made by BLW to FHLT’s data and model specifications. Importantly, these changes result in insignificant results even before the new econometric approaches are implemented. We perform a battery of tests that apply alternative approaches to address the biases suggested in the recent literature. In contrast with BLW, we find that FHLT’s results remain significant.
Traditionally, staggered DiD is implemented using a two-way fixed effects regression. The regression includes dummies for the units, dummies for the time periods, and a binary treatment indicator to capture the differential estimate of the treatment. However, recent studies point out that the two-way fixed effects estimator is not easily interpretable, because it is a weighted average of all possible two-group/two-period DiD estimators. In addition, it may be biased downward or have a different sign due to the use of already-treated units as controls. To mitigate these potential biases, several studies recommend using event-based staggered DiD with alternative groups as controls. To date, there is no consensus on the choice of controls; however, these studies generally use control groups that are never treated, are not yet treated, or remain untreated during the event window.
FHLT use a two-way fixed effects estimator to assess the impact of country-level board reforms—both major reforms and first reforms—on the value of firms in 40 countries from 1990 to 2012. They focus on [-5, +5] year windows and use Tobin’s q to capture firm value. FHLT find that Tobin’s q increases following a board reform, and that this increase materializes when or after the reform becomes effective in the firm’s country. Revisiting FHLT’s analysis, BLW combine the following modifications in their analyses: 1) using the full data set but truncating observations after the final treatment (2007 for major board reforms and 2006 for first board reforms); 2) excluding already-treated observations from the control group when implementing event-based stacked regressions; and 3) coding the effective year of the board reform as year 0 and using year -1 and the most negative timing year as benchmark years.
This approach has several limitations and results in low power. First, by using the full data set but truncating post-treatment observations, BLW exacerbate the influence of confounding events and limit the number of post-treatment observations that can be used to test for treatment effects. This truncation casts doubt on whether the data are presented fairly to detect treatment effects.
Second, by failing to include never-treated countries or countries untreated during the event window as controls, BLW’s approach results in relatively few control observations for later reforms and noisy estimates. For example, Indonesia (Italy) is the only control country for countries with major (first) reforms in 2006 (2005). Furthermore, BLW’s approach is highly sensitive to the choice of sample period and fails to define the test sample ex ante. For example, if FHLT were to begin their sample period in 2000 (instead of 1990, covering the 1992 and 1998 reforms in the U.K.), then U.K. firms would be included in the control group based on BLW’s approach.
Third, BLW do not consistently apply the coding of the timing variables suggested in the literature. For example, while BLW cite Borusyak and Jaravel (2017) to justify using year t-1 and the most negative timing year as benchmark years, they omit Borusyak and Jaravel’s (2017) suggestion that this approach should only be used to assess whether the parallel trends assumption holds, as it does not estimate the treatment effects efficiently. To obtain effective dynamic ex-post treatment effects, Borusyak and Jaravel (2017) recommend setting all lead indicators to zero.
While BLW discuss their empirical choices, they do not disclose the specific modifications used to generate their various plots. To clarify their empirical choices, we present their results in regression format and report the associated data and model specifications. Our additional analyses suggest that BLW’s empirical choices largely explain their insignificant results—even before applying the advanced staggered DiD methodology.
To assess the robustness of our results to alternative estimation approaches, we perform a variety of tests. We add three countries (Ivory Coast, Venezuela, and Vietnam) with no board reforms from 1990 to 2012. Clearly, these countries are not comparable to the sample countries in FHLT (which consist mostly of developed economies). Our intention is not to suggest that future research on board reforms should rely on such countries as controls; rather, we aim to illustrate the robustness of FHLT’s results to using alternative controls and the subjective nature of truncating the sample period based on the last reform (i.e., BLW’s approach). We then run event-based stacked regressions using observations that are never treated or not yet treated (alternatively, untreated) during the event window as controls. Further, we adjust the sample period to start in 1997, five years after the first board reform in the U.K. Again, the results of our analyses using only U.K. firms or both U.K. and never-treated firms as the control group are similar to those reported by FHLT.
Overall, our new analyses emphasize the importance of fair data representation when applying the advanced staggered DiD methodology suggested in recent studies. The pervasiveness of board reforms likely reflects their economic impact worldwide, and our additional analyses strengthen the conclusion that these reforms, on average, increase firm value.
Larry Fauver is a James F. Smith, Jr. Professor in Financial Institutions at the University of Tennessee
Mingyi Hung is a Fung Term Professor of Accounting and Chair Professor at the Hong Kong University of Science and Technology
Xi Li is an Associate Professor of Finance at the University of Arkansas
Alvaro Taboada is a BancorpSouth Associate Professor of Finance at Mississippi State University
