A Hybrid Neural Network GARCH Approach to Forecasting Zimbabwean Inflation Volatility

What the paper says

Volatility of economic indices like inflation, stocks, exchange rates etc has generated a lot of interest among researchers especially after the Options Markets’ celebrated Black Scholes option pricing formula relied on the assumption of constant underlying asset volatility, itself unobserved on the market. In that sphere (the options market) researchers concluded that, as functions of parameters like strike price, underlying asset volatility is actually not constant and one locus is volatility smile. We posit a priori that Zimbabwe went through periods of economic crisis observed (according to data available) from 1980 and thus modellers face challenges using available data for many technical reasons which to the best of our knowledge has resulted in very little research done in this sphere. This paper provides additional research by modelling and forecasting the inflation volatility present in Zimbabwe, using traditional GARCH models hybridized with Artificial Neural Networks (ANN) and Recurrent Neural Networks (RNN). There are several important conclusions drawn from our results. First, out of the GARCH models, the EGARCH generally performed the best. Second, both the ANN and RNN hybrid models outperformed the traditional GARCH models by a significant margin. Finally, the hybrid ANN models provided more accurate forecasts during volatile periods when compared with hybrid RNN models.