Improving Diversification Opportunities for Socially Responsible Investors

This specification has been the most popular in the literature. See Ortas et al. (2014) for an alternative use of this model in the SRI literature.

Schüler and Schröder (2003) pointed out that by using this methodology, all of the estimations are conducted using Maximum Likelihood under the assumption that the residuals follow a normal distribution. However, as these authors also pointed out, this assumption is, in fact, not true because in most cases, the standardised residuals exhibit leptokurtosis. Thus, the application of the normal distribution leads to a so-called Quasi- or Pseudo-ML estimation. According to Weiss (1986), this application leads to a consistent estimation of the parameters if the equations for the (conditional) means and variances are specified correctly. Some authors tried to solve the problem by using different distributions. However, it was proven that when a distribution different from the normal one is used and this distribution is not the true one, then the estimates are, in most cases, not consistent. For that reason, we prefer to apply the normal distribution.

We employ the US Treasury Bill rate as the risk-free interest rate, obtained from Kenneth R. French’s website.

It is important to note that this index family has approximately USD 6 billion in assets under management in a variety of financial products including mutual funds, separate accounts, structured products and exchange-traded funds, and future contracts as reported in the 2012 results of the annual Dow Jones Sustainability Indexes review (http://​www.​sustainability-index.​com).

Although nowadays the field of socially responsible indexes includes several families such as the Calvert Group, E.Capital, Ethibel, FTSE4Good, Humanix, Jantzi, KLD Analytics, and Vigeo, they are not included in our study because they are not available or they are available for shorter time horizons. In addition, these families differ in their construction methodologies, weighting criterion and component selection procedures, and so their inclusion in the analysis could make it difficult to extract concluding remarks.

We use logarithmic returns multiplied by 100 to facilitate the convergence of the empirical models.

Eastern Time (GMT-4).

Data snooping occur whether the same sample is used for both estimation and allocation.

To that end, we consider the benchmark indexes for North America, Europe, and Asia-Pacific provided by Dow Jones STOXX.