The performance of the Domini 400 Social Index as well as that of other socially responsible investment funds has led to two assertions about the performance impact of social-guideline investing. The milder statement is that there is no significant cost to social investment. The stronger claim is that there actually is positive performance value.
This article summarizes research that extends the performance impact of social screens to active value-style management. It also tests a longer time period than previous work, namely the third quarter of 1984 through the end of 1997, which includes the down market of 1990-'91. The effects of beta, growth, size and dividend yield are isolated from value-focused performance variables.
Our analysis used the social screens developed by Kinder, Lydenberg and Domini. We found that when we control for non-model performance factors, there is no significant cost for any KLD social exclusions for the study period for a wide range of value-focused active investment styles.
This analysis addresses only social screening - that is, prohibiting investment in the securities of companies or industries that the investor perceives to be engaged in socially negative behavior. It does not address positive social tilting, which is proactive investment in the securities of companies the investor perceives to be engaged in socially positive business activities.
In our study, the securities universe is the intersection of stocks common to three databases: CRSP, Compustat and I/B/E/S. Since one of our control factors, growth, is the five-year average sales growth rate for the prior five years, stocks must have been in the Compustat data base for five years prior to the time of portfolio formation. The crux of our methodology is to create two cross-sections of forecast-ranked, factor-controlled portfolios in each of 54 quarters from the third quarter of 1984 through the end of 1997. One cross-section is created from the overall security universe, and the second is created from a socially screened subset of the overall universe, but using the same security return forecast and the same portfolio formation algorithm. Thus, the only difference between the two cross-sections is the exclusion of stocks on the basis of one or more social screens.
What is particularly important in terms of performance assessment methodology is how we control for the impact of performance-affecting factors not in the return-forecasting model. Hereafter, we refer to non-model performance factors as control factors. In this study, we controlled for four non-model performance factors: beta, growth, size and dividend yield.
It is crucial to understand that, when we say "control for factor exposure," we are not referring to the conventional practice of after-the-fact adjustments for average differences in portfolio factor exposures using a measured factor price for the time period. We control for factor exposure at the time of portfolio formation by constraining the portfolio selection algorithm to make every portfolio in the cross-section have the same portfolio-weighted average value of each control factor.
A value-focused model
Most practicing analysts, both growth-focused and value-focused, center their analysis on earnings and cash flow forecasting. With many companies having negative earnings and with many growth companies having negative free cash flow, recent attention also has focused on the sales growth rate and the sales-price ratio. In addition to the income statement indicators of value (sales, cash flow, earnings and dividends), many value-focused analysts also consider balance-sheet variables, especially the book-to-market ratio. This very quick and admittedly superficial overview of value measures is intended to justify a set of value variables relative to stock price. The income statement measures are dividends, earnings, cash flow and sales. The key balance sheet measure is net common equity (net book value).
In the security return forecasting model used to represent a formal value-focused investment style, we selected four of these five value variables, each expressed on a per-share basis and scaled by stock price per share. They are: the earnings-price ratio; the cash-price ratio; the sales-price ratio; and the book-price ratio (hereafter called the book-market ratio in accord with conventional terminology).
We used two measures of each of the four value variables, a current measure and a smoothed relative measure. For instance, for the earnings-price ratio, the current measure is the most recent annual earnings per share divided by the current share price.
The smoothed relative value is the most recent earnings-price ratio relative to the five-year average value. We estimate a regression model in which the coming quarterly total returns variable is the dependent variable and the independent variables are the eight value variables and an earnings forecast variable to reflect the most recent consensus earnings forecast from the I/B/E/S databases.
The regression-estimated coefficients indicate the relative importance of each value measure and the consensus I/B/E/S forecast in determining recent returns. The regression weights plus the most recent variable value produces a return score.
Performance assessments
Constructing cross-sections of factor-neutral portfolios provides a robust alternative to linear factor corrections for comparing performance. We shall refer to this portfolio-level forecast performance assessment framework as a factor- neutral portfolio assessment. The crux of the factor-neutral portfolio assessment is constructing a set of portfolios that vary systematically in one attribute (forecasted return) but are matched for key control factors, in this case beta, growth, size and dividend yield. Matched means that each portfolio in the cross-section has the same portfolio average value of each control variable. For the cross-section of factor-neutral portfolios, there is no cross-sectional variation in the portfolio-average value of any control variable.
Impact of social screens
The logic for comparing performance possibilities for the screened subset of the security universe with the overall universe is straightforward. First we created a factor-controlled cross-section for the overall universe. Next, we chose a KLD screen or combination of screens and excluded the screened securities to create the screened universe. We then repeated the logic to create a factor-controlled cross-section for the screened universe. Figure 1 plots the 54-quarter time-average cross-section for the overall universe and all KLD screens in combination. Each realized return point in the cross-section is the time average quarterly return realized for a factor-neutral portfolio of a given forecast rank. The right-most point on each cross-section is the time average value for the portfolio with the lowest predicted return. The next point over from the right is the time-average return for the portfolio with the next lowest predicted return. As one moves from right to left along each cross-section, one moves to progressively higher predicted returns and, generally, to progressively higher realized returns. The left-most point is the time-average return for the portfolio having the highest predicted return score in each quarter.
Figures 2, 3, and 4 show average realized return vs. predicted return score for the factor-neutral return cross-sections for three subperiods using all KLD screens in combination.
We examined the impact of social screening using four KLD screens: alcohol, tobacco and gaming; environmental; defense; and nuclear power. For a long-only investor, the most pertinent points are the left-most points representing the factor-neutral portfolios with the highest predicted return. The plot indicates very little difference in realized return.
Formal statistical tests confirm the conclusions: There were no significant differences in realized return for any KLD screens, individually or in combination. Additionally, there are no significant costs to social screening in subperiods of the 1984-'97 period.
John B. Guerard Jr. is adjunct professor, University of Pennsylvania and a member of the Virtual Research Team, GlobeFlex Capital; and Bernell K. Stone is Harold F. Silver Professor of Finance Marriott School, Brigham Young University. Copies of their working paper can be requested at [email protected]
