Premium
Forecasting Seasonal Time Series of Corporate Earnings: A Note *
Author(s) -
Jarrett Jeffrey
Publication year - 1990
Publication title -
decision sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.238
H-Index - 108
eISSN - 1540-5915
pISSN - 0011-7315
DOI - 10.1111/j.1540-5915.1990.tb01257.x
Subject(s) - box–jenkins , griffin , autoregressive integrated moving average , earnings , econometrics , series (stratigraphy) , time series , computer science , identification (biology) , forecast error , estimation , mean squared prediction error , statistics , economics , mathematics , finance , machine learning , paleontology , botany , management , archaeology , biology , history
The purpose of this paper is to compare the accuracy of various models for forecasting time series of corporate earnings. Previous research indicated that user‐identified time series (ARIMA) models were less useful for forecasting corporate earnings than prespecified models of the Watts‐Griffin and Brown‐Rozeff type. In this study, these research results are disputed. Specifically, prespecified models did not produce models having smaller forecast errors in the statistical sense than did user‐identified models for the same time series data for the same time period. The user‐identified models are those selected by the prescriptive methods of Box‐Jenkins identification, estimation, and diagnostic testing. Furthermore, the magnitude of the forecasting error may be understated for the prespecified models indicating that Box‐Jenkins models may even be more useful than indicated by measures of forecast error.