A Forest Full of Risk Forecasts for Managing Volatility

A Forest Full of Risk Forecasts for Managing Volatility

• Autor
• Andre Portela Santos
• Resumo

• We propose a hybrid methodology that integrates forest-based machine learning with long-memory models of stock return volatility. Our approach exploits cross-sectional information in a panel of stocks and allows time-varying parameters that  are modeled as nonparametric functions of both firm-specific information and changing market conditions. Empirical results on a panel of 1,131 stocks reveal that covariates such as the VIX, idiosyncratic volatility, and momentum have heterogeneous and economically significant effects on the time-varying parameters. In terms of forecasting accuracy, our hybrid approach outperforms a broad range of time-series models and other machine-learning methods over multiple horizons and volatility regimes. Economically, our enhanced risk forecasts yield higher utility for volatility-managed stock investments and deliver minimum-variance portfolios with significantly lower return volatility and higher Sharpe ratios.

• Palavras-chave
• Accumulated local effects, cross-sectional heterogeneity, HAR model, local linear forest, volatility regimes
• Modalidade
• Comunicação oral
• Área Temática
• Econometria Financeira

Voltar Download