Time-Varying Anomaly Premia: Stable Fact or Disappearing Act? with Niels Groenborg, Kingway Lin, and Allan Timmermann
Abstract: We model the dynamics of expected returns for anomaly hedge portfolios in a setting that incorporates information from the full anomaly `zoo' (over 200 anomaly signals). Our panel forecasting model permits both cyclical forms of expected return variation and permanent decay effects related to publication-induced learning. We find strong support for both sources of variation. Our results provide relatively comprehensive evidence regarding which predictive signals most effectively capture time-variation in anomaly premia, and how this varies across different anomaly signal types. We highlight implications for portfolio-timing and for parsing the factor zoo from a dynamic perspective.
High on High Sharpe Ratios: Optimistically Biased Factor Model Assessments with Sara Easterwood
Abstract: Sharpe ratios and other performance metrics associated with prominent factor models are optimistically biased due to conditioning the selection of factors or information sets used to construct factors on preexisting research outcomes. We evaluate models using alternative approaches that mitigate this implicit form of look-ahead bias. Popular multifactor model Sharpe ratios fall dramatically and often fail to exceed that of the market Sharpe ratio. Recently proposed models that are distilled from large sets of cross-sectional return predictors are subject to similar biases. We conclude that Sharpe ratios associated with popular models are unlikely to violate ``good deal bounds'' and asset pricing improvements relative to the capital asset pricing model (CAPM) are more modest than suggested in the literature.
Abstract: We model the dynamics of expected returns for anomaly hedge portfolios in a setting that incorporates information from the full anomaly `zoo' (over 200 anomaly signals). Our panel forecasting model permits both cyclical forms of expected return variation and permanent decay effects related to publication-induced learning. We find strong support for both sources of variation. Our results provide relatively comprehensive evidence regarding which predictive signals most effectively capture time-variation in anomaly premia, and how this varies across different anomaly signal types. We highlight implications for portfolio-timing and for parsing the factor zoo from a dynamic perspective.
High on High Sharpe Ratios: Optimistically Biased Factor Model Assessments with Sara Easterwood
Abstract: Sharpe ratios and other performance metrics associated with prominent factor models are optimistically biased due to conditioning the selection of factors or information sets used to construct factors on preexisting research outcomes. We evaluate models using alternative approaches that mitigate this implicit form of look-ahead bias. Popular multifactor model Sharpe ratios fall dramatically and often fail to exceed that of the market Sharpe ratio. Recently proposed models that are distilled from large sets of cross-sectional return predictors are subject to similar biases. We conclude that Sharpe ratios associated with popular models are unlikely to violate ``good deal bounds'' and asset pricing improvements relative to the capital asset pricing model (CAPM) are more modest than suggested in the literature.
