Description

Decomposing Stock Factors

Li Guo Wolfgang Karl Härdle Yuqian Jin Qingfu Liu Chuanjie Wang 

This study leverages Empirical Mode Decomposition (EMD) to improve stock return predictability by dissecting market anomalies into distinct frequency components. Traditional anomaly research often misses mixed frequency information in firm characteristics. EMD isolates these signals into Intrinsic Mode Functions (IMFs) that capture different frequencies, enhancing the accuracy of predictive models. Using EMD, we develop a composite return predictor from 20 monthly anomalies, which significantly outperforms raw data models. The EMD based composite generates a notable return spread of 79 basis points per month, primarily driven by the long leg with 144 basis points per month. Its robustness is confirmed across various firm characteristics, factor models, transaction costs, and information environments, with stronger predictability for high sentiment periods and those with lower media coverage. EMD composite also predicts future cash flows, earnings surprises, and revenue surprises, highlighting its abil-ty to capture unrecognized fundamental information. This research emphasizes the critical role of frequency information in understanding and predicting cross sectional stock returns.