Evaluation of Some Heteroscedastic Models of Generator Plants Noise Production

Abstract

Regression analysis requires the homoscedasticity condition to fit and predict time series observations. Basically, most real life data have a tendency to exhibit changing in variances which violates the homoscedasticity assumption. As such, it is appropriate to consider outlines that allow the variance to be subjected to the history of the data. This study therefore compare some het eroscedastic models in fitting and forecasting generator noise from one location to another. A stationarity and heteroscedasticity procedures are considered on the data across the locations using Augmented Dickey Fuller (ADF) and Breusch-Pagan (BP) tests respectively. Also, the four heteroscedastic models of different orders are used to model the data. Thereafter, the inadequacies of the models selected are determined for future forecasting in which all generating plants noise data were found to be stationary and heteroscedastic. GARCH(1,1) and GARCH(1,2) are used to fit and forecast the data. It was observed that forecasted values are stationary over time across the locations