An evaluation of univariate time-series models of quarterly earnings per share and their generalization to models with autoregressive conditionally heteroscedastic disturbances

Abstract

This study evaluates time-series models of quarterly earnings per share (EPS) in order to determine whether there are any changes in the residual variance to be modeled by the GARCH procedure. The results of statistical analyses indicate the presence of GARCH effect in the residuals generated from ARIMA models for quarterly EPS. Furthermore, based on Akaike's information criterion, modeling the GARCH effect appears to be desirable. However, the results of forecast accuracy comparisons provide no evidence that the ARIMA-GARCH specification results in more accurate forecasts than the conventional ARIMA models.