The Effect of Board Structure on Firm Disclosure and Behavior: A Case Study of Korea and a Comparison of Research Designs

• Published date: 01 June 2021
• Author: Bernard Black,Woochan Kim,Julia Nasev
• Date: 01 June 2021
• DOI: http://doi.org/10.1111/jels.12280

Journal of Empirical Legal Studies

Volume 18, Issue 2, 328–376, June 2021

The Effect of Board Structure on

Firm Disclosure and Behavior: A Case

Study of Korea and a Comparison of

Research Designs

Bernard Black,*Woochan Kim, and Julia Nasev

We exploit a large legal shock to the board structure of Korean firms, using a strong

research design—combined difference-in-differences and regression discontinuity—to

study whether this board structure change affects firm financial reporting (disclosure,

MD&A length, and abnormal accruals), investment and growth (sales growth and capital

expenditures), and firm value (proxied by Tobin’s q). We also compare results from the

annual DiD/RD design to those from simpler panel and “causal”methods, and assess how

results vary across methods. We find robust evidence across methods that the shock pre-

dicts improved scores on a Disclosure Subindex, confirm prior findings of an increase in

Tobin’s q, and find some evidence for a drop in sales growth, but no convincing evidence

of significant change for other outcomes. By comparing results across methods, we illus-

trate how using multiple causal designs can provide insight into and evidence of robust-

ness not available from a single design, as well as case study evidence that panel methods,

simple DiD, and its close cousin, shock-based IV, can produce apparent false positives.

I. Introduction

We study two research questions. The first involves the effects of a shock to the board

structure of Korean firms on firm behavior. We study whether this board structure shock

affects other aspects of governance (disclosure, board procedure, shareholder rights),

financial reporting (absolute and signed abnormal accruals, MD&A length), firm

*Address correspondence to Bernard S. Black, Nicholas D. Chabraja Professor at Northwestern University School

of Law and Kellogg School of Management, 375 E. Chicago Ave., Chicago, IL 60611; email: bblack@northwestern.

edu. Kim is Professor in Finance at Korea University Business School; Nasev is Associate Professor in Accounting at

LMU Munich School of Management.

We thank seminar participants at LMU Munich School of Management (2020), McCombs School of Business

(2012), Chicago Booth School of Business (2012), participants of the Seventh International Conference on Asia-

Pacific Financial Markets (2012), American Accounting Association annual meeting (2013), Financial Management

Association annual meeting (2013), Shuping Chen, Martin Dierker, Dain Donelson, Christian Hofmann, Carsten

Homburg, Zi Jia, Christian Leuz, Jim Naughton, Joshua Ronen, Thorsten Sellhorn, and Ira Yeung for comments.

We also thank financial support by Korea University Business School.

328

financial outcomes (profitability, leverage, growth, and investment), and firm value

(proxied by Tobin’s q). We do so using a strong, “causal”research design—a design that

combines annual difference-in-differences (annual DiD) regressions with a regression dis-

continuity (RD) design, that draws strength from both design aspects. With this annual

DiD/RD benchmark design, we find evidence that the shock predicts improved scores on

a Disclosure Subindex and higher Tobin’s q, some evidence for lower sales growth, but

no significant change in the other outcomes.

We also provide case study evidence on the sensitivity of research results to choice

of research design. Finance and accounting research has moved strongly toward relying

on natural experiments, using various research designs, including (DiD) and RD sepa-

rately (not often combined), and instrumental variables (IV) with the shock as an instru-

mental variable. But natural experiments are often hard to find or answer limited

questions. Classic panel data approaches—with firm fixed effects (FE) or random effects

(RE)—are often the best available designs. We assess how our results would change if we

used either panel methods or simpler “causal”methods. The potential limitations of

panel methods, as a guide to causal effects, are known (e.g., Leamer, 1983;

LaLonde, 1986), but we know little about how important these limitations are in applied

research. We also know little about how reliable simpler causal designs are as guides to

causation, or about how results differ across methods, including different causal methods

that rely on the same shock.

We exploit a large, exogenous shock to the board structure of large, public Korean

firms (assets over 2 trillion won, about US$2 billion (2T)), with no similar shock to

smaller firms. Korean rules, adopted in 1999 following the East Asian financial crisis of

1997–1998, require large firms to have at least 50 percent outside directors, an audit com-

mittee (with at least two-thirds outside directors and an outside chair), and an outside

director nominating committee (with at least 50 percent outside directors). Before the

rules were adopted, essentially no Korean firm had any of these governance elements.

Thus, this legal shock strongly affects two core governance institutions—outside directors

and audit committees. Prior research has found that the shock strongly affects the market

value of Korean firms, reflected in both share prices and Tobin’s q(Black et al., 2006;

Black & Kim, 2012).

For the first research goal, we find evidence that this shock leads to higher scores

for a “Disclosure Subindex,”but no evidence for a significant change in other aspects of

firm governance (a Board Procedure Subindex or a Shareholder Rights Subindex) or

other aspects of disclosure (signed and absolute abnormal accruals, and word length of

the management’s discussion and analysis (MD&A) section of annual reports). Turning

to financial outcomes, we confirm prior findings that the board structure shock predicts

higher Tobin’s q(Black & Kim 2012), and find some (less than definitive) evidence of a

drop in sales growth, but no overall evidence of significant change in other outcomes,

including profitability measured by return on assets (ROA), or investment (measured by

capital expenditures and research and development spending (R&D)). This evidence,

taken as a whole, is consistent with independent directors and audit committees having a

real, although modest, effect on selected aspects of firm behavior.

The Effect of Board Structure on Firm Disclosure and Behavior 329

For the methods goal, true causal effects are unknown and, as with any case study,

our results may not generalize to other settings. However, the Korean shock provides

an unusually “clean”shock. It is plausibly exogenous (the 2T threshold is apparently

arbitrary), it strongly alters board structure, and it occurs at a specific point in time.

1

We compare assumed “truth”from our benchmark an nual DiD/RD design t o results

from simpler “causal”designs—including “simple Di D”(comparing after to before the

shock), annual DiD, RD, and instrumental variables (IV) using the shock as an

instrument.

With regard to whether our benchmark design can recover approximate truth,

studies across different fields provide evidence that RD can often recover results close to

those from a randomized experiment.

2

From a theoretical perspective, within a band-

width around the discontinuity, one can treat the RD sample as if it came from a random-

ized experiment, with the running variable providing the assignment mechanism that

determines which units are treated (Mattei & Mealli 2016). Combining RD with DiD

addresses a principal limitation of RD, that results are valid only near the discontinuity

(Mealli & Rampichini 2012), and permits use of a broader bandwidth, which increases

sample size and power. Using a combined DiD/RD design also allows us to relax the

usual RD assumption of similar levels between treated and control and rely instead on

the weaker DiD assumption of parallel changes (parallel trends). Using annual DiD allows

one to assess whether the treatment effect appears when it should relative to the shock,

and whether there are nonparallel pretreatment trends that could explain an apparent

treatment effect. Thus, our benchmark annual DiD/RD design provides a reasonable

basis for comparison with other methods. Moreover, given the rarity of true randomized

experiments in finance and accounting, we know of no other way to address the reliabil-

ity of results across different research design.

We compare results from the benchmark design to simpler causal designs and to

panel data designs using either firm fixed effects (FE) or random effects (RE). Overall,

as we move from our benchmark design to annual DiD alone (which we view as the next

strongest design if RD is not available) to other causal designs, firm FE, and finally to

firm RE, there is a general tendency for more outcomes to become statistically significant.

We term these results “apparent”false positives. We do not know the true causal effect,

and it is possible that a significant result with firm FE measures the true causal effect, but

a firm FE result is not convincing given insignificance from stronger designs. Similarly, a

simple-DiD result is not convincing if not supported by the annual DiD analysis (if,

e.g., the annual analysis shows nonparallel pretreatment trends).

For our methods goal—assessing whether results are consistent across methods—

there is no clear winner in terms of not generating apparent false positives. However, sim-

ple DiD is a clear loser—it generates more apparent false positives than other causal

1

We discuss below other limitations of this setting. A principal concern is limits on statistical power due to the

modest number of Korean firms above the 2T size threshold.

2

Buddelmeyer and Skoufias (2004); Black et al. (2007); Cook and Wong (2008); Cook et al. (2008); Green

et al. (2009); Berk et al. (2010); Shadish et al. (2011); Gleason et al. (2012); Moss et al. (2014).

330 Black et al.