Manescu, Cristiana, and Catalin Starica. "Do Corporate Social Responsibility scores predict firm profitability? A case study on the publishers of the Dow Jones Sustainability Indexes." Working Paper, Center for Finance, Schoool of Economics, Business Administration and Law, Gothenburg, Sweden, June 2007.

This study analyzes Return on Assets for a sample of 386 non-financial firms (45% European, 30% North American, 15% Japanese) for the year 2004. The sustainability scores used are taken from Sustainability Asset Management of Zurich, while financial data was sourced from Worldscope Datastream.

Return on assets is evaluated using both traditional linear regressions and non-linear methods. The authors argue that traditional linear approaches are likely to be inadequate, as "the assumption of linearity has no real economic base." A comparison of traditional linear techniques such as OLS, stepwise regression, and Breiman's method with non-linear techniques such as boosting bagged regression trees show that the non-linear techniques substantially reduce prediction error.

The study shows that including CSR scores generally improves the explanatory power of the model - the reduction in prediction error ranges from -2% (OLS) to 3.3% (Boosting / Spline non-linear approach). The authors comment that "overall, adding the CSR scores to the set of explanatory variables improves the prediction."

"We find that a number of CSR scores like Corporate Governance and Codes of Conduct seem to have important explanatory power. In particular, the Corporate Governance positions ahead of common profitability explanatory variables such as leverage, capital intensity, change in sales or size. Above-average scores in corporate governance can add as much as 3 to 4% to Return on Assets (depending on the method used)."

See Belu and Manescu (2009) for further work in this area.