Anderson, Anne Marie, and David H. Myers. "Performance and Predictability of Social Screens." Working Paper, Lehigh University, June 26, 2007.

Compares the performance of four groups of portfolios based on KLD social screens with the Russell 3000 stock index (a broad benchmark for U.S. equities) for the 1991-2004 time period. The four portfolio groups are Positive Portfolios (positive results for a given screen), Negative Portfolios (negative results for a given screen), Neutral Portfolios/Concerns (neutral results for a given screen, but negative ratings in other categories), and Neutral Portfolios/Strengths (neutral results for a given screen, but positive ratings in other categories). Each individual portfolio is formed on one of the 20 KLD social ratings categories, and is composed of companies in the S&P 500. Results are computed using both equal weights.

The study tests the hypothesis that "there is no difference among screened positive and negative screened portfolios and the unscreened S&P 500 firms' returns." This is tested using a variety models, including Jensen's alpha and a persistence analysis modeled on Ferson and Schadt (1996). This work finds that "no clear pattern is seen in the differences among the alphas and the betas," and that alphas do not tend to persist (i.e., positive alpha in one period tends to be negative in the next time period, and vice-versa).

The authors then do a more comprehensive analysis by creating four portfolios:
WHITE: "all firms with at least 2 strengths and none of the six exclusionary concerns (41-96 firms)"
BEIGE: "all with strengths (73-240 firms)"
BROWN: "all with concerns (279-470 firms)"
BLACK: "all with any of the six exclusionary concerns"

Returns to these portfolios are computed using both equal and market weights. Again, no apparent pattern is observed, although portfolio performance differences appears economically meaningful. For example, the highest conditional monthly alpha is recorded by the WHITE equal-weighted portfolio, while the lowest is achieved by the WHITE value-weighted portolio (!).

The authors conclude that "there are no significant differences in the returns of the screens over the periods of 1991-2004. This includes positive, negative, and neutral screening criteria as well as combinations of screens in the four portfolios." The authors comment in their abstract that "investors are no worse off investing in accordance with their social beliefs."