The Decision Tree Approach to Stock Selection

Abstract

A frequent question regarding quantitative investing is: What are good variables for stock selection? Traditional quantitative strategies are variations of screening techniques. Quantitative investment managers seek to narrow the investable universe to a manageable number of stocks that have desirable characteristics. The authors introduce an alternative approach to traditional methods of stock screening based on a statistical technique known as classification and regression tree (CART). CART allows screening factors to interact on a conditional basis. The end result is a hierarchical (tree) structure that assigns a probability of outperformance (or underperformance) for each stock. The authors apply two alternative CART strategies to the selection of technology stocks, and evaluate their performance. The models demonstrate significant improvement over the more ad hoc stock ranking techniques.