Abstract:  In this work we present a way to select macroeconomic factors determinant for the movements of return of individual assets. We use available market data and the factors are found using an unsupervised learning technique based on projection pursuit methodology. The indices selected are based on monthly return data spanning 2000 to 2004 and the resulting model is tested on an independent set of returns for a different time period (2005 - 2007). We compare our approach with the Fama and French (1993) three factor model and with the Ross (1976) Arbitrage Pricing Theory. We discover that the procedure performs surprisingly well when compared with the traditional methods, as assessed by standard statistical measures.

This is joint work with Aleksey Myasnikov.