Conditional forest through the trees: building cross-sections of stock returns applications to the brazilian market

• Ano de defesa: 2024
• Autor(a) principal: Bragança, Bernardo Scarpelli Cabral de
• Orientador(a): Pereira, Pedro L. Valls
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Inglês: Ap-trees; SDF; GARCH
• Link de acesso: https://hdl.handle.net/10438/35262

This study employs a recent methodology called Asset Pricing Trees (Ap-trees) to construct a diverse universe of managed portfolios based on characteristics and utilizes these portfolios as a foundation to recover the Stochastic Discount Factor (SDF). However, unlike the original methodology, we relax the assumption of iid returns. Specifically, in the step of weighting the portfolios to recover the SDF, instead of using the sample counterpart of the variance-covariance matrix, we employ a Factor GARCH model to parsimoniously model the variance process. The one-step ahead forecasts from this model are used as inputs to recover the stochastic discount factor in each period of the test sample. Subsequently, we compare the results with the base model using the Sharpe index and find superior outcomes when using the conditional variance matrix compared to the unconditional one. Additionally, we observe that the contribution of each characteristic to the SDF varies over time, and the conditional model, which takes this into account, achieves better out-of-sample results.