<p>Traditional sparse factor models (e.g., Fama–French) struggle to explain cross-sectional returns in high-dimensional settings due to the ‘factor zoo’—a proliferation of anomalies with overlapping or noisy signals. We show that a principal component (PC)-based stochastic discount factor (SDF) using regularization techniques can aggregate characteristics into dominant risk sources, balancing parsimony and robustness. First, the SDF is estimated using a small sample of 25 portfolios double-sorted by size/book-to-market ratio, and it is found that only 2 principal component factors are needed to predict the cross-sectional returns well, which is consistent with the classical size premium and value premium. Then, the sample is further extended to 72 anomalous characteristics. The results show that the sparse PC-based SDF predicts the cross-sectional returns better than the sparse original characteristic-based SDF. We verify that sparse PC-based models outperform traditional sparse factor models even in emerging markets like China, where retail-driven trading and regulatory shifts amplify idiosyncratic risks.</p>