Bernard 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 and Vladimir Atanasov at the Mason School of Business, College of William and Mary.
Much corporate finance research is concerned with causation—does a change in some input cause a change in some output? Does corporate governance affect firm performance? Does capital structure affect firm investments? How do corporate acquisitions affect the value of the acquirer, or the acquirer and target together? Without a causal link, we lack a strong basis for recommending that firms change their behavior or that governments adopt specific reforms. Consider, for example, corporate governance research. Decisionmakers—corporate boards, investors, and regulators—need to know whether governance causes value, before they decide to change the governance of a firm (or all firms in a country) with the goal of increasing firm value or improving other firm or market outcomes. If researchers provide evidence only on association between governance and outcomes, decisionmakers may adopt changes based on flawed data that may lead to adverse consequences for particular firms.
In our recent working paper, Shock-Based Causal Inference in Corporate Finance Research, we study what researchers do in major journals to address causal inference, and then build on this survey to provide an overview of credible causal inference designs. We survey 13,461 articles in 22 major journals in accounting, economics, finance, law, and management over 2001-2011, and identify 863 empirical corporate governance papers which study whether corporate governance predicts firm value or another dependent variable. We classify the strategies these papers use to examine potentially causal relations between governance and firm outcomes and assess how successful these strategies are.
A central theme of the paper is that credible causal inference strategies usually rely on “shocks” to governance. These shocks generate “natural” or “quasi” experiments, which can provide reason to believe that a change in governance causes a change in the firm’s value or behavior. Here, a “shock” is a discrete, external event that causes some firms to be treated; the others become “controls.” The assignment of firms to treatment versus control should be plausibly exogenous—uncorrelated with firm characteristics (observed or unobserved). Usually the shock will occur at a known time and we can measure outcomes both before and after the shock (discontinuity designs can be an exception). Most convincing shocks, in turn, come from legal rules, rule changes, and law-based discontinuities (together, “legal shocks”). We identify 77 papers with shock-based research designs (involving 42 distinct shocks), and use these papers to provide a guide to shock-based design.
A second central theme is a focus on shocks and general features of shock-based design, which cut across the particular econometric methods that are used to exploit shocks. We show that difference-in-differences (DiD), regression discontinuity (RD), event study (ES), and instrumental variable (IV) designs must all satisfy similar exogeneity, relevance, and “only through” conditions. These common elements of shock-based design have been obscured because the causal inference literature has largely treated each design separately.
Finding a “credible shock” (one which provides a credible basis for causal inference) is central. Given such a shock, several designs can often be used to exploit the shock, including difference-in-differences (DiD), regression discontinuity (RD), event study (ES), and IV. Often, these approaches can be combined. For example, if a shock involves a discontinuity (e.g. some portions of SOX apply only to firms with public float above $75 Million), a combined DiD/RD design will often be attractive. Inference in event studies relies on a combination of DiD and IV assumptions. One design can often be substituted for another; for example, most shock-based IV designs can be recast as DiD.
We seek to provide guidance on how to improve causal inference, even if inference remains imperfect. In almost all of the shock-based papers in our sample, there is substantial room for better design. We share neither the perspective of some researchers, whose view can be caricatured as “endogeneity is everywhere, one can never solve it, so let’s stop worrying about it”; nor the “endogeneity police”, whose attitude is that “if causal inference isn’t (nearly) perfect, a research design is (nearly) worthless”; nor that of authors who know they have an endogeneity problem, but say little or nothing about it in their paper, hoping the referee won’t notice, or offer a weak instrument to address endogeneity and hope the referee won’t object.
We do believe that useful shocks can often be found. Even true randomized experiments can sometimes be found or created. We collect in a public database (Atanasov and Black, 2014, Database of Shocks Used in Corporate Finance and Accounting Research), the shocks used in our sample. Many of these shocks can be put to additional uses. Many additional useful shocks surely exist, but have not yet been exploited. We plan to update our shock database to include additional shocks based on examination of papers outside our sample or inputs from the broad corporate finance research and practitioner communities.
The full paper is available for download here.