Corporate Governance Indices and Construct Validity Bernard Black Northwestern, Pritzker Law School and Kellogg School of Management (coauthors: Antonio Gledson de Carvalho; Vikramaditya Khanna; Woochan Kim; Burcin Yurtoglu) (GCGC Conference, June 2016]
Construct Validity Very much an exploratory project Term borrowed from education, psychology Research questions: What is “good” Corporate Governance? What goes into a “good” corporate governance index How does it vary across countries? Different local rules, institutions Construct validity questions: How good are our proxies? How can we tell? Very much an exploratory project
Prior research: Black et al. (J Econometrics 2014) Study Brazil, Russia, India, Korea, Turkey: “BRIKT” Related governance indices in each country Built from governance “aspects”, proxied by “subindices” Board structure Disclosure Board procedure (Minority) Shareholder rights Ownership structure Control of RPTs Different subindex “elements” across countries
Very different elements Governance elements must be: Measurable Meaningful (in judgment of local coauthors) attend to local rules, institutions We think they might reflect “good” governance Lots of judgment here! Significant variation across firms Not useful if required by law; nearly universal; or rare Not too similar to another element similar across countries to extent feasible (often not) Turns out: elements are very different
Brazil Corp Gov Index (BCGI) Use Brazil to illustrate approach and complexities Subindices (each 0 ~ 100) for: Board Structure (7 elements) Ownership Structure (5 elements) Board Procedure (6 elements) Disclosure (11 elements) Related Party Transactions (5 elements) Minority Shareholder Rights (7 elements) BCGI = [∑(subindices)/6] Range: [19, 92] BCGInorm = normalized [∑(normalized subindices)]
Importance of local institutions Brazilian institution: fiscal board Can be permanent (in charter) or near-permanent (demanded every year or most years by minority shareholders) We use (4 years out of 5) as out measure of “near-permanent” Function similar to audit committee often a substitute Audit committees rare (mean = 0.14) Fiscal board common (mean = 0.68)
Compare Brazil to Korea for Board Structure Brazil Element (NP = not public) Korea Element Board includes ≥ 1 indep directors (NP) Required ≥ 30% independent directors (NP) Requires 25% independent directors ≥ 50% indep directors (NP; mean = 0.20) in KCGI Strict majority of indep. directors CEO is NOT board chairman Not available Audit committee exists (rare; mean = 0.14) Near-permanent fiscal board exists Not meaningful Audit committee or permanent fiscal board includes minority shareholder representative Not available; rare Rare (NP) Compensation committee exists Outside director nom. committee Only common elements are: 50% outside directors (uncommon in Brazil) (NP) audit committee (rare in Brazil; misleading alone) Only public common element: audit committee Rare in Brazil, misleading alone
Example 2: Board procedure subindex Start with Brazil & India: see if avail in Korea, Turkey Overall Procedure subindex Brazil India Korea Turkey ≥ 4 regular board meetings per year NP X available Average board meeting attendance rate ≥ 80% Outside directors attend minimum % of meetings Firm has system to evaluate CEO Firm has system to evaluate other executives Firm evaluates nonexecutive directors Firm has succession plan for CEO Firm has nonexecutive director retirement age Directors receive regular board training Outside directors only annual board meeting NP, rare Board receives materials in advance Nonexecutives can hire own counsel & advisors Directors’ positions recorded in board minutes Firm has code of ethics Specific bylaw/policy to govern board Firm has ≥ 1 foreign outside directors Shareholders approve outside directors’ aggregate pay NP = not public (use surveys). With public data, can’t build Brazil subindex at all Even with surveys, can’t build consistent subindex across countries
Lesson: CG index must be country-specific If require same elements in each country: Can measure little What we measure may not be very relevant Problem gets worse if add more countries
Severe construct validity questions We’re not sure how to measure “governance” Not sure what counts as “good” CG, for which firms, in which countries We have. . . Different overall indices in each country Different subindices in each country Very different subindex elements in each country We hope: CG indices & subindices proxy for similar concepts
Gov Value: Neither Necessary nor Sufficient One can ask (as all of this literature does): Does gov predict firm value (proxied by Tobin’s q) CGI as a whole, or particular subindices Suppose not, why? Maybe no effect of gov on value Maybe bad construct Suppose yes: subindex predicts the outcome, but is poor construct; measures something else Plus usual “endogeneity” concerns
Construct validity is a hard problem Widely ignored No easy solutions We offer an exploratory study We hope to make (some) progress . . .
If “valid construct”, what might we hope for Build the elements and subindices first “Design stage” (Rubin, 2008) Cannot aim directly for construct validity If you do, tests for validity are meaningless
Once subindices are built Some correlation across subindices Reasonable Cronbach α Standard (but crude) measure of construct validity But too high a score not desirable Reasonable inter-element correlation (within subindex) But too high bad element choice Principal components are coherent May onto subindices
Correlations Between Subindices Brazil DS BS OWN BP SR RPT Subindex complement 0.57 0.24 0.18 0.29 0.47 0.08 Disclosure Index (DS) 0.19 0.40 0.61 0.10 Board Structure Index (BS) 0.28 0.23 0.05 Ownership Structure Index (OWN) 0.04 Board Procedure Index (BP) 0.15 -0.01 Minority Shareholder Rights Index (SR) 0.07 India 0.17 0.09 0.13 0.14 -0.04 Korea 0.43 0.51 -0.09 0.44 0.46 0.42 -0.06 0.36 0.38 0.39 -0.12 Turkey 0.58 0.62 0.52 0.20 0.01 0.27
Correlations between Elements
Cronbach α measure of correlation between elements of a multipart measure (0 ~ 1): 𝛼= 𝑛𝑟 1+ 𝑛−1 𝑟 n = number of governance elements r = mean correlation among elements High α valid subindex? We argue Low r worry sign High r also a worry sign!
Cronbach α results Brazil India Korea Turkey All governance elements Brazil India Korea Turkey All governance elements Cronbach α 0.80 0.70 0.76 0.88 Mean r 0.09 0.05 0.10 0.14 All subindices 0.56 0.31 0.50 0.58 0.18 0.08 0.20 0.22 Board Structure Subindex 0.38 0.74 0.63 0.13 0.29 Board Procedure Subindex 0.49 0.61 0.19 0.07 0.24 Disclosure Subindex 0.84 0.69 0.43 0.86 0.32 0.15 0.21 Ownership Structure Subindex 0.64 ─ 0.40 0.26 Shareholder Rights Subindex 0.68 0.11 0.33 0.42 0.23 0.03 RPTs Subindex 0.77 0.36
Why worry if high r? Suggests elements are not measuring different things Narrow subindex, many similar elements vs. broader subindex, disparate elements Lower α But better capture of governance complexity
What counts as “high enough” α? No theory Some rules of thumb from psychology, education But can increase α by having more elements Higher n Can lead to search for suspect elements Or having more-similar elements Higher r But can be spurious guide to validity
Principal Component Analysis (PCA) Principal Component Analysis (varimax rotation) Example: Turkey Element level: loadings > 0.4 Subindex Level Element Level Elements (mean loading) Subindex Factor1 Factor2 Factor 3 Board Structure 0.7039 -0.0327 5 (0.84) Disclosure 0.8036 0.1283 7 (0.81) 4 (0.79) Shareholder Rights 0.0964 0.8799 Ownership Structure -0.0066 0.8908 Board Procedure 0.7803 0.0273