Hedge fund returns are often explained using linear factor models such as Fung and Hsieh (2004). However, since most hedge funds live only for 3years, these linear regressions are subject to over-parameterization. I improve the out-of-sample accuracy of the linear factor model by combining cross-sectional and time series information for groups of hedge funds with similar investment strategies. The additional cross-sectional information allows more accurate estimates of risk exposures. I also propose a trading strategy based on this methodology for extracting substantially larger risk-adjusted returns.