The Effect of Ethical Fund Portfolio Inclusion on Executive Compensation

James Brander is the Asia-Pacific Professor of International Business in the Sauder School of Business at the University of British Columbia. An economist, his research areas include international trade, industrial organization and finance. He is a former editor of the Canadian Journal of Economics and has published widely, including the textbook “Government Policy Toward Business”.

Abstract

This paper divides firms in the Standard and Poor’s 500 (S&P 500) into two groups based on inclusion in or exclusion from the Domini Social Index (DSI). Inclusion in the DSI is interpreted as a positive indicator of ethical status. Using data for the 1992–2003 period, I provide evidence that chief executive officer (CEO) compensation, other executive compensation, and director compensation tend to be lower in DSI firms than in other firms in the S&P 500. This applies to the unconditional group averages (and medians) and is particularly striking given that DSI firms as a group had better financial performance than the other firms. This finding is also true in a regression framework that controls for other influences on compensation, including firm size and financial performance. In a regression context, the estimated discount for CEOs of DSI firms is approximately 12% for both current compensation (salary and bonuses) and total compensation (including the value of options). These results are consistent with the expectation that some senior executives require a “compensating differential” to accept positions in firms with less attractive ethical status. It is also consistent with the expectation that some firms with positive ethical status might use more restraint in setting executive compensation.

Appendices

Appendix 1. Data used

Most of the data used in this study is from the Compustat Executive Compensation (EC) database. Leverage is constructed from the corresponding Compustat Industrial Annual (IA) database. The data was downloaded in April 2005 using the Wharton Research Data Services interface. The databases were last updated on December 31, 2004. Data was available from 1992 to 2003. The Compustat Executive Compensation Database uses abbreviations to identify data series. The Compustat abbreviations of the data series used are in Table A1.

Table A1  

All financial variables were converted to real 2003 dollars using the consumer price index (CPI).

Appendix 2. Specification and estimation issues

(a) Linear, log-linear, and semi-log functional forms

One important specification issue concerns whether to estimate equation (1) in linear or log-linear (or some other) form. I use the log of compensation as the dependent variable and the log of assets, employment and experience as control variables. One reason for using logarithms is that compensation, size measures, and experience are all skewed in the sense of having a few very large values. Taking logarithms reduces this skewness and therefore reduces the influence of outliers.

A second reason relates to the expected relationship between the dependent variable and explanatory variables. If the linear form with unlogged variables is used, a given absolute change in employment is associated with a constant absolute change in compensation. For example, we would estimate that an additional 1000 employees is always associated with an additional $17,000 per year for CEO total compensation. The log-linear form implies that this premium declines on a per-employee basis as the number of employees rises such that the “elasticity” is constant, which means that a given percentage change in assets is associated with a constant percentage change in compensation. Thus, for example, we are able to say that a 1% increase in employment is associated with a 0.15% increase in CEO current compensation.

If employment were logged and compensation were left unlogged (referred to as a “semi-log” specification), a given percentage change in employment would be associated with a constant absolute change in the level of compensation. While not suitable for employment or assets, the semi-log form is natural for variables already in percentage form. In this case, for example, we can say that a 1(percentage) point increase in the total return to shareholders is associated with a 1.4% increase in total compensation. See Wooldridge (2003), p. 46 for a discussion of interpreting coefficients in log-linear and semi-log specifications.

In this case, it seems most reasonable and most consistent with common practice to use logarithms of compensation, assets, employment, and experience, and unlogged levels of ROA, TRS, volatility, and time, along with the DSI indicator variable. It is of course also important to consider which form is the better approximation to the actual data under consideration. In this case, the proposed (logarithmic) form does well in specification testing, although the pure linear form would also be reasonable.

(b) Industry fixed effects

When estimating wage or income equations, it is not uncommon to use industry fixed effects to capture unexplained industry-level differences in compensation. That would be inappropriate in this case as a significant part of the selection for inclusion in the DSI is done on an industry basis (as, for example, with tobacco companies). Accordingly, “correcting” for industry differences would be expected to leave out an important part of the effect of ethical status on compensation. In short, ethical status varies by industry and we would not want to abstract from that effect.

(c) Panel data issues

A second point to consider is that the data used in this study is panel data in the sense that there is a set of cross-sectional units (firms), each of which is tracked for a period of time (12 years). Our estimates make use of both cross-sectional and time-series variation (and there is a substantial amount of time-series variation in total and current compensation for virtually every firm). However, it is not appropriate to use standard fixed effects or random effects panel estimation because, by construction, there is no time-series variation in the DSI indicator for any given firm. Time-series variation does contribute in an important way to estimates of the sensitivity of compensation to the various control variables, however. Accordingly, the most suitable approach is simply to use all the observations in a standard regression estimated using ordinary least squares.