Generative AI for European asset pricing: alleviating the momentum anomaly
本文用生成式AI构建资产定价模型，发现欧洲市场存在大量因子而非少数集中因子，该模型在16年样本外测试中表现优于所有基准，年化夏普比率达3.68，并揭示了动量异常可通过条件贝塔更有效替代。
We challenge the notion of classical factor models that concentrated factors, particularly the anomalous momentum factor, dominate the European stock market. We use a generative artificial intelligence (generative AI) asset pricing model that incorporates the economic rationale of no-arbitrage and treats the European capital market as a complex system. This model outperforms all European benchmarks over 16 years out-of-sample, with an annualized Sharpe ratio of 3.68, a cross-sectional 𝑅2 of over 22%, and an explained variation of over 13%. Using interpretable AI techniques, we find that the model sees a zoo of factors in the European market rather than just a concentrated set. These excellent results stem from time-conditional modeling, which requires momentum, especially for tangency portfolio weights. Conditional betas can substitute momentum more efficiently. Overall, the risk-sharing mechanism for European assets is more complex than previously thought.