Identifying idiosyncratic stock return indicators from large financial factor set via least angle regression

Identifying important indicating factors for expected level of stock return has been one of the central problems in modern finance. Researchers have worked on different candidate sets of indicators from different perspectives, but there has not been a consensus reached on which factors to be included in the model. In this paper, based on relative complete information from a large set of factors from US financial reports, we use least angle regression (LARS) to select a sparse and relatively stable set of indicators for predicting stock return. The use of LARS is consistent with the theoretically well developed economic theory arbitrage pricing model. The empirical results offer new insights of the well-known indicators from the previous studies and discover new important factors.