Factor Zoo Revisited: Multiple Testing, Hierarchical Modeling, and Out-of-Sample Evidence

Factor Zoo Revisited: Multiple Testing, Hierarchical Modeling, and Out-of-Sample Evidence

• Autor
• Paulo Roberto Fonteles Guimarães
• Co-autores
• Herbert Kimura
• Resumo

• The proliferation of return predictors has raised concerns about multiple testing, p-hacking, and out-of-sample performance. While most evidence comes from developed markets, Emerging Markets (EM) provide a demanding out-of-sample environment, where many factors were not originally proposed and face harsher inferential conditions. This paper revisits existence, magnitude, and robustness of asset-pricing discoveries through multiple-testing adjustments and hierarchical modeling.

  We examine three questions: whether alphas persist after multiple-testing correction, whether hierarchical theme-based modeling improves inference, and whether these approaches enhance out-of-sample portfolio performance. Using an EM equity panel, we consider frequentist controls and Bayesian hierarchical models, which are embedded in walk-forward backtests.

   

  Results show that only a limited set of themes delivers robust alphas. Hierarchical models reveal economically meaningful theme-level effects while supporting more parsimonious inference. Portfolio-wise, ignoring multiple testing harms performance, while overly conservative frequentist corrections raise risk. FDR-based and hierarchical Bayesian approaches provide a more balanced, economically meaningful framework.

• Palavras-chave
• actor zoo, asset pricing, multiple testing, partial pooling, emerging markets, hierarchical models, Bayesian statistics, False Discovery Rate, Family-Wise Error Rate
• Modalidade
• Comunicação oral
• Área Temática
• Apreçamento de Ativos

Voltar Download