Hybrid forecasting models for S&P 500 index returns

Purpose - The purpose of this paper is to propose two hybrid forecasting models which integrate available ones. A hybrid contaminated normal distribution (CND) model accurately reflects the non-normal features of monthly S&P 500 index returns, and a hybrid GARCH model captures a serial correlation with respect to volatility. The hybrid GARCH model potentially enables financial institutions to evaluate long-term investment risks in the S&P 500 index more accurately than current models. Design/methodology/approach - The probability distribution of an expected investment outcome is generated with a Monte Carlo simulation. A taller peak and fatter tails (kurtosis), which the probability distribution of monthly S&P 500 index returns contains, is produced by integrating a CND model and a bootstrapping model. The serial correlation of volatilities is simulated by applying a GARCH model. Findings - The hybrid CND model can simulate the non-normality of monthly S&P 500 index returns, while avoiding the influence of discrete observations. The hybrid GARCH model, by contrast, can simulate the serial correlation of S&P 500 index volatilities, while generating fatter tails. Long-term investment risks in the S&P 500 index are affected by the serial correlation of volatilities, not the non-normality of returns. Research limitations/implications - The hybrid models are applied only to the S&P 500 index. Cross-sectional correlations among different asset groups are not examined. Originality/value - The proposed hybrid models are unique because they combine available ones with a decision tree algorithm. In addition, the paper clearly explains the strengths and weaknesses of existing forecasting models.