In this section, I discuss survivorship bias and a series of tests I estimated to evaluate the effects of various economic and global factors: the effect of state-owned enterprises (SOEs), the effect of the long US bull market following the global financial crisis (GFC) of 2007–2009, the stability of the results in the period prior to the GFC, the role of small-cap stocks, and the effect of limiting the strategy to long-only (top-quartile) portfolios. 
Survivorship Bias.
The corporate governance tests, which relied on MSCI scores, were conducted over the 2009–17 period, but MSCI did not begin issuing ratings until 2013. For prior periods, MSCI took its 2013 sample and worked backward to 2009, estimating ratings for the same set of companies. This methodology raises the question of potential survivorship bias in the tests. The results in this study were unlikely to be driven by this bias, however, for two reasons. First, inferences were based on in-sample comparisons across quartiles of governance or ESG scores, not on comparisons between the MSCI sample and a survivorship-bias-free non-MSCI sample. Second, for survivorship bias to influence the differential performance across these quartiles, it would need to be correlated with the sorting variable, but this phenomenon seems unlikely to explain the results. If one argues that survivorship bias affects companies with poor governance more than those with good governance (i.e., poor governance companies are less likely to survive), the effect would be to inflate the returns of the poor governance quartile in the surviving sample and attenuate return differences between the top and bottom quartiles, thereby biasing against an ability to find the results reported here.
Effect of SOEs in the Sample.
Although it has been reported that SOEs have relatively poor performance due to governance issues, SOEs were unlikely to be influential in the results reported here for two reasons. First, SOEs were unlikely to explain the monotonic relationship between governance quartiles and returns, which held across the entire distribution of governance. Second, 88% of the ACWI IMI sample consists of companies from the developed markets, which attenuated any potential SOE influence. 
Effect of the US Bull Market.
The corporate governance tests spanned the post-GFC period, when US equity markets have performed more strongly than those in the rest of the world. To examine whether this US “bull run” was driving the results, I excluded US equities from the original test sample and reestimated for corporate governance and for ESG. In the ex-US sample, the new composite governance score quartile had a coefficient of 8 bps monthly (t-statistic = 4.10), implying a 24 bp monthly return spread between the top and bottom quartiles. Without the United States, the material ESG score quartile had a coefficient of 10 bps monthly (t-statistic = 3.68), implying a 30 bp monthly return spread between the top and bottom quartiles.
Pre-GFC Effect.
The corporate governance tests spanned a nine-year post-GFC period, but a useful test, if data availability permits, might be to assess the stability of the results in a pre-GFC window. Because data for two governance factors, ownership dispersion and shareholder orientation, are available beginning in 2000, I reestimated the first and second regression specifications in over the longer, 2000–17, window to examine the efficacy of these two factors in the pre-GFC period. Ownership dispersion and shareholder orientation remained statistically significant at the 1% level in a one-tailed test (coefficient = 0.0014, t-statistic = 10.65 for the ownership dispersion tercile; coefficient = 0.0136, t-statistic = 9.42 for shareholder orientation).
Effect of Small-Cap Stocks.
Could the results have been driven by small-cap stocks? This outcome is unlikely for two reasons. First, all the portfolio-level tests and results used cap-weighted returns, and all the company-level tests controlled for market cap. Second, the results remained robust in the MSCI ACWI universe, which consists of large-cap and mid-cap stocks only.
Long-Only Strategy.
Some investors might consider ESG investing in a long-only fashion and would be interested in the performance of a long-only strategy (rather than a long–short strategy) versus an overall index consisting of all firms regardless of ESG data availability. I examined the performance of a long-only governance strategy and an ESG strategy versus the MSCI ACWI IMI universe (all companies in the universe regardless of whether they had governance or ESG data available). 
Figure 10 shows the cumulative cap-weighted performance of the top-quartile portfolio of corporate governance companies (as measured by the composite governance score) versus the MSCI ACWI IMI universe. Each month, the cap-weighted forward returns of the top governance quartile and the entire universe were calculated; the return streams were cumulated to arrive at the growth curves in . As shown, companies with the best governance outperformed the universe over the test period by 40 pps cumulatively. shows the cumulative performance of the top-quartile portfolio of ESG companies (as measured by the material ESG score) versus the MSCI ACWI IMI universe. In companies with the best ESG scores outperformed the universe over the test period by 17 pps cumulatively.
Notes: The figure shows the growth in the value of US$10 invested in the top quartile of a new composite governance score versus the MSCI ACWI IMI universe. The top quartile of corporate governance had 84,124 company-month observations in the test period; the MSCI ACWI IMI universe had 887,781 company-month observations in the test period. 
Figure 11. ESG Long-Only Strategy vs. Universe, January 2013–November 2017
Notes: The figure shows the growth in the value of US$10 invested in the top quartile of the material ESG score versus the MSCI ACWI IMI universe. The MSCI ACWI IMI universe consisted of all index companies. The top-quartile total of the material ESG score had 45,511 company-month observations in the test period; the MSCI ACWI IMI universe had 496,640 company-month observations in the test period. 
