Evaluation of Forecasts Performance of ARIMA-GARCH-type Models in the Light of Outliers

The carry-over effect of biased estimates of ARIMA-GARCH-type models parameters on forecasting accuracy is investigated in the presence of outliers by exploring the daily returns of share price series of three major banks in Nigerian. The banks considered are Diamond, United bank for Africa and Union. The data were collected from the Nigerian Stock Exchange and spanned from January 3, 2006 to December 30, 2016, comprises 2713 observations and were divided into two portions. The first portion which ranges from January 3, 2006 to November 24, 2016, comprises 2690 observations was used for model formulation and the second portion which ranges from November 25, 2016 to December 30, 2016, consisting of 23 observations was used for out-of-sample forecasting performance evaluation. The parametric bootstrap technique was used in computing the forecasts while Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Error (ME) were the methods of forecast evaluation considered. The findings of this study showed that in the presence of outliers, the forecasts were found to be biased as indicated by ME and the accuracy reduced as shown by MSE, RMSE and MAE. However, adjusting for the outliers, only marginal improvement on the forecasts was observed, reason being that all the outliers were treated as innovations and they occurred before the forecasts origin.