Outlier time-series models and analysts' forecasting of GNP and corporate earnings per share

Abstract

The purpose of this note is to estimate an outlier analysis model of GNP and to develop composite models using forecasts by GNP forecasting services and the outlier analysis model.The outlier analysis model outperforms the GNP forecasting services in post-sample forecasting accuracy.However, the composite models incorporating the outlier analysis model do not always outperform the composite models based on only the GNP forecasting services.Outlier analysis is particularly effective in forecasting corporate earnings per share during the 1983-87 period in the US and for the 1985-92 period in Japan.The outlier-adjusted models produce forecasts that generate statistically significant excess returns.