An Evaluation of Univariate Time-Series Earnings Models and Their Generalization to a Single Input Transfer Function
评估了以往季度盈利时间序列模型是否因忽略市场盈利指数中的信息而存在设定偏误，并引入一个同时利用指数和盈利时间序列特性的传递函数模型。
The development of statistical models for accounting earnings has been an evolving process. Earlier studies used simple index and/or naive models (e.g., Ball and Brown [1968] and Brown and Kennelly [1972]) because of the lack of tools to deal with formal time series. Later, Dopuch and Watts [1972], Ball and Watts [1972], and others employed formal time-series models to annual earnings, and Collins and Hopwood [1980], Brown and Rozeff [1979], Foster [1977], Griffin [1977], and Watts [1975] did the same with quarterly earnings. While this research has produced some conclusions with respect to the time-series properties of earnings and how these may be used in predicting earnings, Brown and Rozeff [1978] and then Collins and Hopwood [1980] recently provided evidence that financial analysts can produce forecasts more accurate than those based on the ARIMA models alone. This suggests that other information may be useful in providing estimates of future earnings, when incorporated with knowledge of the time-series properties of earnings. The purpose of this study is twofold: (a) we first evaluate the models used in past studies of quarterly time series of earnings in order to determine whether they are systematically misspecified due to the omission of other information which is implicitly contained in a market earnings index-the evidence suggests that the misspecification exists, and (b) we introduce a transfer function model which utilizes the timeseries properties of both the index and earnings numbers. This model is obtained based on both theoretical derivation and inspection of data.