Trebucq and Russ (2005)
Trebucq, Stephane, and Robert Russ. "The case of earnings and stakeholder management: towards an integrated theory of managerial behavior." Paper presented at European Accounting Association 28th Annual Congress, 2005.
This unique and interesting study examines the relationship between corporate social records (based on data from the KLD database) and accounting accruals, a measure of earnings quality. According to the authors, "our main argument states that managers behave with accounting like all other social fields. If they usually cheat with stakeholders, they will also cheat with accounting numbers."
Using a sample of 587 U.S. firms, the authors conduct a regression analysis covering the 1990-2001 time periods. The independent variable is a series of summary social ratings computed from KLD data. The dependent variable is an estimate of discretionary accruals developed from a Modified Jones Model as described by Hribar and Collins (2002) - this includes control variables for cash flow, size, distress (Altman Z-Score), leverage, and growth. The authors control for industry effects by estimating separate models for each SIC code.
The regressions don't fit particularly well (highest r2 is 12%), but many of the KLD variables show statistically significant impact. For example, testing in individual issue areas showed that positive social characteristics tended to be associated with lower accruals (higher quality-of-earnings), while social concerns tended to be associated with higher accruals (lower quality-of-earnings). A complex sub-analysis raises many questions deserving study. Some are confusing (variables with the opposite sign from that predicted), others are intriguing, including the possibility that some companies manage earnings down to improve their bargaining position with employees and regulators.
Concludes that "significant correlations between the KLD ratings and discretionary accruals suggest that ethical managers who take care of their stakeholders...do not manipulate earnings." The authors acknowledge that "while the results presented are mixed," there clearly is some connection between social ratings and accruals. The authors speculate that the addition of additional control variables might clarify the results.
LK comment: see Kim, Park, and Wier (2012)