Most of the literature on the effect of corporate governance on firms’ behavior, including financial reporting, investment and growth, provides evidence on association, not causation. How likely are these results to survive, if one could apply causal methods? We provide evidence on that question, by comparing “classic” panel data research designs (pooled OLS, firm random effects, firm fixed effects, and first differences), simpler causal designs (simple difference-in-differences (DiD), regression discontinuity (RD), instrumental variables (IV), and year-by-year DiD), and combined causal designs (combining simple DiD, IV, and year-by-year DiD with RD), estimates. We use a dataset that permits us to use all of these methods, from a case study of Korea. Under 1999 Korean legal reforms, large firms (assets over 2 trillion won, about US$2 billion) faced a major, exogenous shock to board structure, with no similar shock to smaller firms. This shock to board structure predicts higher firm value (proxied by Tobin’s q) and improved scores on an overall disclosure index with both panel data and causal methods. The shock predicts lower investment, slower growth, lower absolute abnormal accruals, and more extensive MD&A disclosure with panel data methods, but these results fall away with causal methods. Some results survive with simpler causal methods, but weaken with combined designs. The shock has limited impact on other outcomes with both classic and causal methods. Our case study provides evidence that classic panel methods can provide a weak guide to causation, and simpler causal methods can also be unreliable.
|Original language||English (US)|
|Publisher||Social Science Research Network (SSRN)|
|Number of pages||55|
|State||Published - Apr 13 2016|