TY - JOUR T1 - Deep Sequence Modeling: Development and Applications in Asset Pricing JF - The Journal of Financial Data Science SP - 28 LP - 42 DO - 10.3905/jfds.2020.1.053 VL - 3 IS - 1 AU - Lin William Cong AU - Ke Tang AU - Jingyuan Wang AU - Yang Zhang Y1 - 2021/01/31 UR - https://pm-research.com/content/3/1/28.abstract N2 - The authors predict asset returns and measure risk premiums using a prominent technique from artificial intelligence: deep sequence modeling. Because asset returns often exhibit sequential dependence that may not be effectively captured by conventional time-series models, sequence modeling offers a promising path with its data-driven approach and superior performance. In this article, the authors first overview the development of deep sequence models, introduce their applications in asset pricing, and discuss their advantages and limitations. They then perform a comparative analysis of these methods using data on US equities. They demonstrate how sequence modeling benefits investors in general through incorporating complex historical path dependence and that long short-term memory–based models tend to have the best out-of-sample performance.TOPICS: Big data/machine learning, security analysis and valuation, performance measurementKey Findings▪ This article provides a concise synopsis of deep sequence modeling with an emphasis on its historical development in the field of computer science and artificial intelligence. It serves as a reference source for social scientists who aim to use the tool to supplement conventional time-series and panel methods.▪ Deep sequence models can be adapted successfully for asset pricing, especially in predicting asset returns, which allow the model to be flexible to capture the high-dimensionality, nonlinear, interactive, low signal-to-noise, and dynamic nature of financial data. In particular, the model’s ability to detect path-dependence patterns makes it versatile and effective, potentially outperforming existing models.▪ This article provides a horse-race comparison of various deep sequence models for the tasks of forecasting returns and measuring risk premiums. Long short-term memory has the best performance in terms of out-of-sample predictive R2, and long short-term memory with an attention mechanism has the best portfolio performance when excluding microcap stocks. ER -