Are Anomalies Still Anomalies after Allowing for Change Points?
Prof. Lingxiao ZHAO
Date: 11 October 2023 (Wednesday)
Time: 10:30 am -12:00 pm
Host: Prof. Zhuo QIAO, Associate Professor in Business Economics
Online registration: https://umac.au1.qualtrics.com/jfe/form/SV_eVtS0y6ZQCik14q
We present a pure Bayesian method for identifying anomalies’ identities when the observations are allowed to belong to different regimes. Starting with a general CAPM framework for anomaly pricing, we introduce an anomaly pricing model that allows for multiple change-points. The estimation methodology for the model is described, along with the model comparison approach for these multiple change-point models. Utilizing 153 anomalies and 40 years of monthly observations for from 1973 to 2022, we demonstrate that, from a Bayesian model comparison perspective, the model with one change-point is substantially superior to the model without any change-points for over 96% of the anomalies. To determine the number of change-points, we calculate the marginal likelihoods for each anomaly with different numbers of change-points. Furthermore, we present the mispricing posterior estimation results for each anomaly in the last regime, considering various numbers of change-points. The results show that as we incorporate more change-points, the number of genuine anomalies diminishes. This conclusion is supported by both the Bayesian t-ratio measure and the Bayesian 95% confidence interval measure. When considering five regimes of anomalies, only approximately 25% of genuine anomalies remain, providing strong evidence of the existence of the replication crisis for the majority of anomalies when multiple change-points are considered.
Prof. Lingxiao Zhao is currently an Assistant Professor of Finance at Peking University HSBC Business School. She got her Ph.D. in Economics from Washington University in St. Louis, and her research primarily focuses on asset pricing and Bayesian econometrics. Her work published in the Journal of Finance and upcoming research to be featured in Management Science.
All are welcome!