A Generalized Distributed Lag Model for Predicting Quarterly Earnings
提出一种引入宏观变量分布滞后的收益预测模型，与多种已有模型比较后发现它能缓和近年预测误差的爆炸性，但并非在所有情况下最优。
Despite much research, uncertainty still exists regarding the choice of appropriate time-series models for forecasting earnings and other purposes. One factor previously not explored is the role of nonaccounting information in forecasting earnings. In this paper, I propose an empirical methodology for forecasting earnings which allows the introduction of nonaccounting data via a distributed lag with macro (DLWM) variables model. Predictions from this model were compared with those of several models suggested in previous studies for a sample of capital goods and durable goods firms wherein I found that the DLWM methods moderated the explosive nature of earnings predictions errors encountered in recent years. However, the DLWM approach did not dominate other prediction methods in all cases. The conclusions of the study must be considered preliminary because (1) alternative specifications were not explored and (2) relatively crude methods were used to predict the macroeconomic variables themselves. Previous attempts to validate and compare various earnings predictions by using ex post error measurement have yielded inconsistent results across industries and time. These attempts used either of two approaches to the application of Box-Jenkins (BJ) techniques. One (the parsimonious approach) relies on earnings series to establish the structure of a general model, thus avoiding the problems encountered in fitting