Files
Abstract
Building a strategy that can consistently outperform the market index is not an easy task. The goal of this paper is to create a trading strategy that beats the performance of the SPY during the period of 2010-2022. The first step is to predict the relative stock performance of companies in the S&P 500, S&P 400, and S&P 600 indices compared to the overall market performance between two earnings report periods. Insider trading information, technical indicators, and accounting features are used to train machine learning models, providing a more comprehensive profile of each company compared to relying solely on accounting data. Classification models, such as random forest and gradient boosting, are used to make the prediction. The significance of some of the features indicates their predictive power. The second step is to build a long and short trading strategy based on the prediction results and compare it with the performance of a buy and hold SPY strategy. The results show that the random forest strategies outperform SPY during most of the testing periods, and the strategies generate a positive Alpha. The paper provides insights into combining fundamental information and technical indicators to predict future stock directions. Quantitative finance researchers can also use these features to find market signals and develop trading strategies.