Corporate Social and Financial Performance Re-Examined: Industry Effects in a Linear Mixed Model Analysis

Abstract

In this research, we shed new light on the empirical link between corporate social performance (CSP) and corporate financial performance (CFP) via the application of empirical models and methods new to the CSP–CFP literature. Applying advanced financial models to a uniquely constructed panel dataset, we demonstrate that a significant overall CSP–CFP relationship exists and that this relationship is, in part, conditioned on firms’ industry-specific context. To accommodate the estimation of time-invariant industry and industry-interaction effects, we estimate linear mixed models in our test of the CSP–CFP relationship. Our results show both a significant overall CSP effect as well as significant industry effects between CSP and CFP. In conflict with expectations, the unweighted average effect of CSP on CFP is negative. Our industry analysis, however, shows that in over 17% of the industries in our sample, the effect of CSP on CFP for socially responsible firms is positive. We also examine the multidimensional nature of the CSP construct in an industry context by exploring the CSP dimension–industry nexus and identify dimensions of social performance that are associated with either better or worse financial performance. Our results confirm the existence of disparate CSP dimension–industry effects on CFP, thus our results provide important and actionable information to decision makers considering whether and how to commit corporate resources to social performance.

Appendix

Intrinsic Value Derivation

We compute the intrinsic value of a share of common stock using the residual income valuation model:

$$ V_{t} { = }B_{t} \rm{ + }\sum\limits_{{{\tau = 1}}}^{T} {R^{{{ - \tau }}} E_{t} \left[ {x_{{t{ + \tau }}}^{r} } \right]{ + }R^{{{ - }T}} E_{t} \left[ {V_{t + T} { - }B_{t + T} } \right]} . $$

In the residual income model, the value of a share of stock at time t, V _t , equals the current per share book value of common equity, B _t , plus the present value of future residual income. In the equation above, R denotes one plus the required return on equity (r), and x _t denotes net income year t. E _t is an expectation operator conditioned on information available at time t. Residual income in year t equals net income in excess of the opportunity cost of beginning-of-period book equity: $x_{t}^{r} { = }x_{t}{ - }rB_{{t{ - 1}}}$. Penman and Sougiannis (1998) provide a complete derivation and discussion of the residual income model’s properties.

We compute the intrinsic value of each firm following Courteau et al. (2001), who showed that intrinsic values calculated by residual income possess very high explanatory power (i.e., R-square) for stock price. They also demonstrated the statistical equivalence of residual income and discounted cash flow models. We use forecasts of earnings, dividends and stock prices from the Value Line Earnings and Projections file. This dataset contains forecasts for the current year, the following year, and the 3- to 5-year ahead period. Each week, Value Line publishes updated reports for approximately 125 firms, so that projections are updated once per quarter for all firms in its coverage universe. To ensure that forecasts incorporate information from each firm’s most recent annual report, we take Value Line data from the last published report in May of each year from 2001 to 2008. In addition to each firm’s current stock price, these reports include all updated forecasts from the prior 13 weeks inclusive. For example, for the 2005 sample year the Value Line report date is May 18, 2005. This report contains earnings and dividend forecasts for fiscal years 2005 (τ = 1), 2006 (τ = 2) and for the 2007–2009 period. We take the mid-point of the 2007–2009 period as our 2008 (τ = 4) forecast and compute the 2007 (τ = 3) forecast by simple interpolation between 2006 and 2008. Value Line also projects the high and low points of the expected range of future stock price 3- to 5-years ahead. We use the average of these high and low points as our 2008 (τ = 4) stock price and assume that at that point intrinsic value equals stock price, i.e., V ₄ = P ₄. Beginning book value per share (B ₀) is calculated from net worth and shares outstanding taken from the Compustat data for the most recently reported fiscal year (i.e., 2004). Thereafter, book value is calculated according to the clean surplus relation: B _t+τ = B _t+τ−1 + x _t+τ − d _t+τ.

As described in §4, the required return on equity (r _i) is estimated from the Fama–French (1992) 3-factor model and calculated according to the following:

$$ r_{\rm{i}} { = }r_{\rm{f}} \rm{ + }\left( {r_{\rm{M}} { - }r_{\rm{f}} } \right){\beta }_{{\rm{i,M}}} \rm{ + SMB \times \beta }_{{\rm{i,SMB}}} \rm{ + HML \times \beta }_{{\rm{i,HML}}} $$

One-year T-bill yields, r _f, are obtained from the US Federal Reserve Weekly Statistical Release H15. Factor risk premia are the monthly time series averages of returns from July 1931 through December 2010.¹² For each company and year, the factor loadings (i.e., the betas) are estimated from returns during the prior 60 months. Because estimates for individual stocks are highly imprecise (see e.g., Fama and MacBeth 1973), we estimate factor loadings for each of the 30 industry portfolios defined in Fama and French (1997) and assign the industry loadings to individual companies by matching on SIC codes. For each company and year, the loadings are estimated during the prior 60 months for the corresponding SIC-matched industry portfolio.

Our results are robust to alternative calculations of required return. In addition to the Fama–French 3-factor calculations, we computed intrinsic value using two alternative methods. First, we used the standard Capital Asset Pricing Model (CAPM) with Value Line-supplied betas. Next, we used the standard CAPM with betas estimated from market model regressions of industry returns on the value-weighted NYSE/AMEX/NASDAQ index. We assigned these industry betas to individual stocks matching on SIC codes. All three approaches produced qualitatively similar results. We report our results based on the Fama–French 3-factor model as it has been demonstrated to possess superior explanatory power for the cross-section of average stock returns (Fama and French 1992).

See Tables 5, 6, and 7

Table 5 Complete estimation results for Eq. 1

Table 6 Complete Estimation Results for Eq. 2

Table 7 S&P GIC