Portfolio Optimization through Machine Learning

Abstract

A major challenge in the field of quantitative finance is maximizing risk-adjusted returns. In this article, we present a method for developing a portfolio of stocks that attempts to balance risk and reward to achieve this goal. To do so, we divided the returns of each security in the portfolio by the risk of that security, determined by the negative volatility of the previous year’s prices, and used these scores to determine what percent of the portfolio should be allocated to each security. Our method outperformed the majority of the stocks included in the portfolio and outperformed even weighting of each stock. From the results of this experiment we concluded that it is possible to devise a portfolio that performs better than even allocation; however, the final optimized portfolio was not without risks and did not produce returns as large as the best possible portfolio. In summary, our method provides a reasonable starting point for investors interested in maximizing risk-adjusted returns, with the possibility for improvement in future work.