Mislearning of Factor Risk Premia under Structural Breaks: A Misspecified Bayesian Learning Framework

Battle Count: 73

Highly relevant for quantitative trading. The paper provides a novel state variable (mislearning intensity Delta_t) that predicts long-horizon factor performance and risk characteristics. The uncertainty premium mechanism suggests factor timing opportunities. The distinction between low-friction (uncertainty premium) and high-friction (instability/drawdown) environments is directly actionable for portfolio construction. The passive ownership interaction provides a market-structure conditioning variable. The anomaly family heterogeneity results inform factor selection and risk management. However, the predictive power is primarily at 12-month horizons, limiting short-term trading applications.

Moderate to high complexity. Requires implementing: (1) Kalman filter for stable state-space model, (2) Markov-switching model for break-aware benchmark, (3) predictive density estimation and likelihood ratio computation, (4) rolling mislearning measures, (5) predictive regressions with multiple horizons and controls, (6) cross-sectional decomposition analysis, (7) IVOL screening, (8) passive ownership interaction tests. The mathematical framework is well-specified but the estimation of Markov-switching models for 212 anomalies requires careful implementation. The anomaly family classification adds additional complexity.

The paper uses publicly available data (CRSP/Compustat factor returns, Global-Q data library, ICI fund data). The methodology is well-documented with full mathematical proofs in appendices. However, no GitHub repository is mentioned, and the anomaly family classification uses custom name-based rules that may be difficult to replicate exactly. The Kalman filter and Markov-switching estimation procedures are standard but the specific implementation details (hyperparameters, initialization) are not fully specified.