Recursive Portfolio Selection with Decision Trees

A great proportion of stock dynamics can be explained using publicly available information. The relationship between dynamics and public information may be of nonlinear character. In this paper we offer an approach to stock picking by employing so-called decision trees and applying them to XETRA DAX stocks. Using a set of fundamental and technical variables, stocks are classified into three groups according to the proposed position: long, short or neutral. More precisely, by assessing the current state of a company, which is represented by fundamental variables and current market situation, well reflected by technical variables, it is possible to suggest if the current market value of a company is underestimated, overestimated or the stock is fairly priced. The performance of the model over the observed period suggests that XETRA DAX stock returns can adequately be predicted by publicly available economic data. Another conclusion of this study is that the implied volatility variable, when included into the training sample, boosts the predictive power of the model significantly.