Introducing shrinkage in heavy-tailed state space models to predict equity excess returns

Area of Science:

• Econometrics
• Financial Modeling
• Bayesian Statistics

Background:

• Forecasting excess returns of the S&P 500 index is crucial for investment strategies.
• Existing models often struggle with non-Gaussian features and overparameterization.

Purpose of the Study:

• To develop and evaluate a flexible Bayesian econometric state space model for forecasting S&P 500 excess returns.
• To incorporate non-Gaussian features, shrinkage priors, and heavy-tailed innovations for improved forecast accuracy.

Main Methods:

• Utilized a flexible Bayesian econometric state space model.
• Employed global-local shrinkage priors to control overparameterization.
• Incorporated heavy-tailed state innovations and leptokurtic stochastic volatility.

Main Results:

• The proposed Bayesian model demonstrated superior performance compared to common competitors.
• The model achieved better point and density forecasts for S&P 500 excess returns.
• Shrinkage priors and heavy-tailed innovations effectively managed model complexity and potential breaks.

Conclusions:

• The flexible Bayesian state space model offers a robust framework for financial return forecasting.
• Accounting for non-Gaussian features and employing advanced Bayesian techniques enhances predictive accuracy.
• This approach provides a valuable tool for quantitative analysts and researchers in finance.