I have some questions with model compares ion and forecasting. The row data (quarterly traffic accident numbers) is not stationary but it is stationary at first difference. we can model and forecast ARIMA model by taking the first difference of the data.In order to see the volatility of the data we can do with GARCH model by taking relative ratio of the data as return series. since the stationary data(1st difference)is not full fill the assumptions of GARCH model. 1. Is it possible to compare the performance between ARIMA and GARCH models for estimation and forecasting using different series to choose the best one. the series is 1.1st difference series 2.return(relative ratio)series for ARMA-GARCH model.