The prediction performance of independent factor models
Laiwan Chan · 2003
In the literature, independent component analysis (ICA) has been proposed to construct factor models in finance. According to the basic principle, the factors extracted using ICA are expected to be independent of each other. This factor model is hence called the independent factor model, in contrast to the traditional factor models which assumes uncorrelated factors. We analyze and compare the performance of the independent factor model and the traditional factor model based on the prediction ability of the factors. Two examples are given to show that the independent factor model would reduce loss if we have good predictability on one of the factors. On the contrary, the uncorrelated factor model may not benefit from an accurate factor prediction.