Abstract
Observations of investor behavior in financial markets suggest that investors are increasingly sensitive to non-financial criteria in the portfolio construction process. We therefore conduct two empirical studies on the impact of adopting an investment strategy that integrates environmental, social and governance (ESG) issues into asset selection on portfolio performance. The first study allowed us to compare the performance of three portfolios composed of U.S. market stocks. The composition of these portfolios was made via ESG filters. First, the best-in-class stocks, i.e. those with the highest ESG ratings, then those with average ratings and finally the lowest rated stocks and finally the lowest rated stocks. The implementation of the optimization is carried out in a mean-LPM theoretical framework that highlights our specific modeling of the portfolio dependency structure. The second study focuses on two portfolios of Eurozone stocks. One portfolio withthe best-in-class and the second one composed of stocks with homogeneous ratings. Unlike the first study, theoptimization is done in a worst case worst case mixture copula mean-CvaR theoretical framework. From these studies we concluded that this criterion is not incompatible with profit, but it would be necessary to reduce these requirements in terms of preference for stocks with high ESG ratings. In view of a potential extension with more efficient models, a third empirical study was conducted to highlight the power and relevance of machine learning models for forecasting financial time series. Or as an alternative a hybridization between artificial neural network model and ARIMA model to predict the trend of the time series.