Empirical Asset Pricing With Many Assets and Short Time Series

We construct tracking portfolios consisting of a large number of assets for macroeconomic factors using the L_2-boosting algorithm. We use these tracking portfolios as instruments to estimate factor risk prices. The same learning algorithm also provides the weights of a mean-variance efficient portfolio. With this additional input we compute the Hansen-Jagannathan distance to compare how alternative models fit the cross-section. We apply the method to 900 portfolio return series in the Kenneth French data library. While macro factors fail to explain most cross-sectional variation, we find that both consumption and inflation risk are priced