An adaptive evolutionary strategy for long–short portfolio construction

This contribution aims to investigate portfolio optimization problems, that qualifies amongst the most discussed topics in the FinTech domain. In particular, we tackle the uniperiodal portfolio selection problem, that aims at finding the optimal composition of a portfolio over a given time horizon, and in this framework the portfolio should bear low turnover and transaction costs in order to be affordably rebalanced: this can be attained by constraining the number of assets in a portfolio, and the amount of wealth invested in a specific asset or asset class. In the same direction, it is also possible to set a minimum margin for loan-financed transactions of securities, as required by Regulation T in the U.S., making the problem hard to be solved by exact methods, and for which approximated algorithms seem to be a viable tool for providing a near-optimal solution. In this framework, the aim of this paper is to present a new mathematical approach in the FinTech domain, by presenting an adaptive evolutionary algorithm approach for the portfolio optimization problem that is compliant with the Regulation T. Empirical results show that our approach can be used to construct effective long–short portfolios that achieve better risk-return trade-offs compared to standard portfolios with long-only formulations.