FORECAST ACCURACIES OF HYBRID OF BILINEAR AND EXPONENTIAL SMOOTH TRANSITION AUTOREGRESSIVE MODELS WITH GARCH MODELS

The study looks at the forecast accuracies of GARCH and Bilinear models on the one hand, and hybrids of Bilnear with GARCH (BL-GARCH) and ESTAR with GARCH (ESTAR-GARCH) models on the other hand. Necessary mathematical or theoretical frameworks were set up for the four models and were illustrated with inflation rate data from Nigeria (1994–2019). Stationarity tests conducted (Unit Root Test) show that at level (original data) the series is chaotic and volatile (non-stationary), but at the first difference the series is stationary. This, therefore, paved the way for the analysis. In and out of sample, the forecast accuracies of all the models used were checked and compared. Information criteria such as AIC, BIC, and FPE were used, where it was seen that ESTAR-GARCH performs best. Also, the performance measure indices used revealed that ESTAR-GARCH was very outstanding. The same goes for variance analysis (ESTAR-GARCH had the least variance). The results show that the two hybrid models give exceptional performances; also, the GARCH model's performance is fairly good but not comparable to the hybrids. Exponential GARCH models perform better than Bilnear-GARCH models and GARCH models. However, the ESTAR-GARCH and BL-GARCH models are comparable and could be interchangeably used by would-be forecasters, investors, academia, and other would-be users.