Three essays in financial economics ; Trois essais en économie financière

This thesis makes a significant contribution to the field of asset returns forecasting. The first part introduces an innovative portfolio optimization paradigm, called "Smart Alpha," aimed at enhancing equity portfolio diversification while effectively managing exposure to risk factors. Recognizing the growing interest in factor investing due to the underperformance of purely active or passive strategies, Chapter 1 presents an active strategy of stock selection, which avoids betting on a-priori factors but focuses instead on an approach that minimizes the exposure of the portfolio to systematic sources of risk, while maximizing its potential alpha. Empirical evidence drawn from European stocks demonstrates the superiority of the Smart Alpha strategy over the European market index and over popular European factor investing and smart beta strategies. The second part of the research focuses on the critical task of short to medium-term return forecasting. While long-term predictive models using financial ratios have been well established, evidence for predictability over shorter horizons remains limited in the academic literature. In Chapter 2, a new predictive regression model is proposed, capitalizing on the observed dynamics of stock returns following mean reversions in the US Shiller CAPE ratio. This model displays superior predictive power, particularly at short horizons, from one to several months, both in-sample and out-of-sample. It leverages business cycle variables, such as the term spread and credit spread, to further enhance its predictability, particularly out-of-sample. The results are robust with respect to the choice of the valuation ratio (CAPE, excess CAPE, or dividend yield), and countries (Canada, Germany and the UK). Chapter 3 extends this approach by incorporating an extensive array of business cycle variables, employing penalization techniques. This extension arises from the absence of a consensus regarding which business cycle variables present the best predictive performance for mean ...