A simple stock screener framework for portfolio optimization

Abstract

This thesis proposes a stock portfolio optimization method that is simple, scalable, and efficient compared to other proposed strategies from the literature, while significantly outperforming the market. We discuss the survivor bias effect that affects datasets composed of historical information on stock prices and how that can distort results and hinder the proper evaluation of any portfolio optimization strategy. Our approach uses a screening tool to select stocks out of a large pool. The screener’s parameters are optimized on a training dataset. We then construct a portfolio which weights stocks so as to minimize the correlation of the selected stocks. We also incorporate a "trigger" mechanism for identifying downturns in stock prices in a way that informs our trading decisions. Using multiple testing periods of 14, 17 and 20 years, our strategy surpassed the S&P500 index and outperformed many similar studies. Overall, this work shows that a simpler, more fundamental approach can oftentimes perform better than complex models.