Ensemble learning for stochastic volatility jump diffusion models

Abstract

The pricing of securities is indispensable in ensuring the efficiency of financial markets. This is notably important in frontier markets where financial systems are evolving, and asset dynamics exhibit unique characteristics. Accurate pricing ensures market efficiency by enabling the accurate reflection of all available information in asset prices. Incorporation of salient features and characteristics of these assets requires the use of accurate models. Extensive research on these properties in frontier markets is imperative. With the increase in the financial instruments being introduced and developed in these markets, a deeper understanding of the underlying security is cardinal. This study applies a model that accurately describes these features in the Kenyan market. A stochastic volatility jump diffusion model is used to characterize the behaviour of market indices. The Bates model, an extension of the Heston model is applied to market index data from the Nairobi Securities Exchange. Estimation of the model parameters is done through Markov Chain Monte Carlo techniques. The model was not only able to capture stylized facts of stochastic volatility but also the discontinuous price movements by incorporating jumps. The calibration of the Bates model on this market brings to light the presence of jumps on the market indices. A Random Forest ensemble algorithm is then utilized in asset return prediction. The results demonstrate that this ensemble method yields a forecast with very low root mean squared errors. This forecasting technique is thereby able to track the return series closely. The predicted values are close to the actual values. The method is shown to be accurate in the prediction of these returns.