Empirical Asset Pricing with Bayesian Regression Trees

Drew D. Creal现任美国伊利诺伊大学香槟分校经济学系教授。此前，他曾于圣母大学和芝加哥大学Booth商学院担任副教授。他的研究主要聚焦于高维模型、时间序列、贝叶斯方法、宏观经济政策以及金融资产定价。Creal教授现任Journal of Econometrics、Journal of Business and Economic Statistics、Journal of Applied Econometrics、Journal of Financial Econometrics以及Econometrics Journal的副主编。他的研究成果发表于多个顶级期刊，包括Journal of Finance、The Review of Financial Studies、Review of Economics and Statistics、Journal of Econometrics、Journal of Business and Economic Statistics、International Economic Review、Quantitative Economics等。

Portfolio sorts are a popular technique used in finance to study the cross-section of expected returns. However, existing methods are typically limited to including one or two variables at a time, making it difficult to disentangle which characteristics are the most important. We address this problem by developing a new Bayesian factor model with regression tree priors that selects across a large space of characteristics simultaneously. We apply our methods to an unbalanced panel of currency returns. We find that the interest rate differential and FX volatility are the primary drivers of currencies’ betas. Portfolios constructed from the model exhibit a higher Sharpe ratio than the carry trade and can be achieved using information on the VIX, aggregate capital ratio of financial intermediaries, and the global interest rate differential.