An analysis of the accuracy of long-term earnings predictions

This paper provides information on the long-term predictive ability of annual earnings numbers. We obtained a sample of 486 calendar, year-end firms that had complete quarterly earnings-per-share (eps) before extraordinary items available from 1978 to 1998. Firm-specific, quarterly, autoregressive-integrated-moving-average (ARIMA) time-series models were used to generate one through five year-ahead annual eps predictions across the 1994-1998 holdout period. Analysis of mean absolute percentage errors indicates: (1) firm-specific ARIMA models outperform so-called, common-structure, "primier" ARIMA models, (2) forecast errors from the firm-specific ARIMA time-series models ranged from 0.358 to 0.547 for one through five year-ahead annual eps predictions, (3) long-term earnings forecast accuracy is linked to firm size and earnings persistence, and (4) further research is needed to develop more powerful, long-term earnings prediction models suitable for use in conjunction with the abnormal earnings valuation model.