The cost of being good.

Introduction

This paper examines the performance of U.S. equities through the use of socially responsible investment screens. Socially responsible investing (SRI) is reflected in the attitudes of the investor to apply social goals to their investment portfolio. We extend the SRI literature by examining a broader study of social screens in a more restrictive context. The sample used in Sauer (1997), based on the Domini Social Index, is "selected to minimize the potential negative side effects." In contrast, our initial approach is to examine the costs and benefits from the most extreme approach of socially responsible investing--the exclusion of companies not meeting a social investment screen. The exclusionary approach results in portfolios that maximize the costs to socially responsible investing. We examine the returns and risk-adjusted returns to investing in socially responsible investment portfolios using 20 social screens from KLD Research & Analytics, Inc. (KLD).

We extend the literature by uniquely examining the persistence in performance of SRI screens, using both Jensen's alpha and conditional alphas. Our results support the null hypothesis of no cost to investing in a socially responsible manner, or reject the hypothesis that SRI comes at a significant cost. Through either single or multiple screens based on value or equally weighted portfolios, we find no statistical significant difference among SRI screen funds. While the differences in portfolio returns are statistically insignificant, the equal-weighted "AL2SNX portfolio" of no exclusionary screened firms and only firms with at least two strengths is the best performer over the period from 1991-2004.

Literature Review

Previous literature has examined screens and socially responsible investing in both equity and bonds. Angel and Rivoli (1997) find that the reluctance of investors to invest in certain firms can lead to increases in the firm's cost of equity; however, the percentage of investors unwilling to invest has to be relatively large for the effect to be significant. Feldman, Soyka, and Ameer (1997) analyze the impact of the firm's environmental management system on stock prices, and find that improvements result, primarily, from a decrease in risk.

Research in the area of investing has led to inconclusive results. Kurtz (1997) finds that the universe of SRI stocks does not appear to underperform the market, but there are costs to diversification and information effects. Guerard (1997) finds that returns for a socially screened universe do not differ from the unscreened universe, but using multiple screens improves results. Statman (2000) finds that the Domini Social Index outperforms the S & P 500, and that socially responsible mutual funds do better than conventional funds, but the results are not statistically significant. Finally, Derwall and Koedijk (2005) find positive, but insignificant, differences between the performance of SRI funds and conventional funds, and evidence of time-variation in performance over business cycles.

Methodology

There are a number of ways to approach social investment screens. It is evident from the mutual funds available that some combination or variation of three approaches is taken--divest from or exclude firms that are inconsistent with the investor's social goals; reduce or underweight exposure to such firms; or take on active shareholder initiatives to change the actions of a firm. In an asset pricing context, we examine the first or most restrictive approach, which is to test the impact of exclusions on investment portfolios. Exclusionary screens imply that an investor is either fully in favor of or against a particular social screen or issue. A firm must pass a particular screen for an investor to include it in his or her portfolio. One way of interpreting such an investment policy in an optimization framework is that there is an infinite cost attached to the exclusionary belief. If the cost were not infinite, then the SRI process would be more reflective of an underweighting of a firm in the portfolio relative to a benchmark.

Asset Pricing Model of SRI and Econometric Issues

In this initial analysis, a simple model of social investment is sufficient. An investor is assumed to have an objective function that maximizes a combination of wealth and social good. Investors may have homogeneous beliefs for all firm returns and infinite costs on any one social screen. Zero investment portfolios are equivalent in the model to taking a long (short) position in the positively screened equal-weighted portfolio, and taking a short (long) position in the negatively screened equally-weighted portfolio.

The model is in line with the common assumption that adding a non-wealth criterion to the investment decision comes at a cost to the investor. This cost to socially responsible investing is also associated with a constrained optimization of the investment set where the constrained efficient frontier lies on the interior, or tangent to the unconstrained frontier (see Exhibit 1). Testable hypotheses arise from differences in returns and risk. If SRI constrained portfolios are inefficient, then they should either have significantly lower returns for the same level of risk, or higher risk for the same return level.

Performance Measures

Our performance measures are made relative to the Russell 3000 as a broad market index. While the KLD data covers the Russell 3000, it only does so after 2000. To get the longest time frame for analysis, the portfolios created in this study are restricted to the S & P 500 for which data are available back to 1991. Equal-weighted and value-weighted portfolios are reconstituted annually based on inclusion in the S & P 500 and the KLD data on screens. Returns are calculated on a monthly basis and then employed in generating risk-adjusted performance measures for the portfolios. Both a Jensen's alpha and a conditional alpha are calculated for each portfolio. We also generate alphas based on the three Fama-French factors.

The dynamic or conditional alpha is based on Christopherson et al. (1998). Both the conditional and unconditional models are estimated using 36 months of past returns. The Capital Asset Pricing Model (CAPM) and the Jensen's alpha fall out of equation (1) if the conditioning information variables, Z, are zero:

[r.sub.pt+1] = [a.sub.0p] + [A'.sub.p][z.sub.t] + [b.sub.0 pb][r.sub.bt+1] + [B'.sub.pb][z.sub.t][r.sub.bt+1] + [u.sub.pt+1]

[^.[beta].sub.pb]([Z.sub.t]) = [^.b.sub.0 pb] + [^.B'.sub.pb][z.sub.t]

[z.sub.t] = [Z.sub.t] - E(Z)

[^.[alpha]] = [^.a.sub.0p] + [^.A'.sub.p][z.sub.t]

Equation 1

--where [r.sub.p] is the return of the account portfolio in excess of the risk-free rate. Z is a vector of the four demeaned 36-month rolling lagged information variables: dividend yield, detrended bill rate (subtracting the 12-month moving average), January dummy, and term spread for the two conditional pricing models. Our null hypothesis is that there is no difference in returns among socially responsible screened portfolios. Another interpretation of the null is that there is no cost to SRI. HO: SRI investors' utility functions include both return and socially responsible behavior, E(Rn) [greater than or equal to] E([R.sub.SRI]) This study examines the cost to a risk-return utility function from the imposition of social screens.

Performance Persistence Methodology.

The final step in the analysis is to examine persistence or predictability in performance. Persistence is measured by a cross-sectional regression technique similar to one used by Christopherson et al. (1998) for the past 36-month alphas on future returns.

[r.sub.p(t,t+[tau])] = [[gamma].sub.0,t,[tau]] + [[gamma].sub.1,t,[tau]][[alpha].sub.pt.sup.CAPM] + [u.sub.p(t,t+[tau])]

[r.sub.p(t,t+[tau])] = [[gamma].sub.0,t,[tau]] + [[gamma].sub.1,t,[tau]][[alpha].sub.pt.sup.CCAPM] + [u.sub.p(t,t+[tau])]

[r.sub.p(t,t+[tau])] = [[gamma].sub.0,t,[tau]] + [[gamma].sub.1,t,[tau]][[alpha].sub.pt.sup.FF] + [u.sub.p(t,t+[tau])]

Equation 2

For horizons, T = 1, 3, 6, 12, 18, 24, and 36 months. Alphas are generated by a traditional CAPM, Conditional CAPM, and Fama-French (1989 and 1992) regressions.

The cross-sectional regression coefficients for each month are averaged over time similar to Fama and MacBeth (1973). The cross-sectional regression is a weighted-least squares (WLS) approach, where the weights are the residuals from equation 1. The result is an appraisal ratio, alpha divided by its standard error, based on Brown et al. (1992) to compensate for survivorship bias due to differences in volatility. If spurious persistence is created by differences in volatility, Brown et al. find that the use of an appraisal ratio compensates for the bias.

Data

There are several socially responsible data bases available. For mutual funds and exchange-traded funds, the most common benchmark or index is the Domini Social Index, which is a product of KLD Research & Analytics, Inc. The acceptance of the KLD Indexes both within the literature and industry led us to choose the KLD data on screening for this study. The database that we employ from KLD classifies 112 subcategories or screening questions for the period 1991-2004. The 112 subcategories are summarized into 20 main categories of socially responsible screens. Most of the category screens are represented in positive and negative screens of strengths versus concerns.