Regime Switching and Artificial Neural Network Forecasting

This paper provides an analysis of regime switching in volatility and out-of-sample forecasting of the Cyprus Stock Exchange using daily data for the period 1996-2002. We first model volatility regime switching within a univariate Markov-Switching framework. Modelling stock returns within this context can be motivated by the fact that the change in regime should be considered as a random event and not predictable. The results show that linearity is rejected in favour of a MS specification, which forms statistically an adequate representation of the data. Two regimes are implied by the model; the high volatility regime and the low volatility one and they provide quite accurately the state of volatility associated with the presence of a rational bubble in the capital market of Cyprus. Another implication is that there is evidence of regime clustering. We then provide out-of-sample forecasts of the CSE daily returns using two competing non-linear models, the univariate Markov Switching model and the Artificial Neural Network Model. The comparison of the out-of-sample forecasts is done on the basis of forecast accuracy, using the Diebold and Mariano (1995) test and forecast encompassing, using the Clements and Hendry (1998) test. The results suggest that both non-linear models equivalent in forecasting accuracy and forecasting encompassing and therefore on forecasting performance.