Further Descriptive and Predictive Evidence on Alternative Time-Series Models for Quarterly Earnings
重新评估了三种简约的季度盈余时间序列模型（Foster、Griffin-Watts、Brown-Rozeff）的拟合优度，并控制行业因素后寻找替代模型，为会计研究者选择预测模型提供参考。
Studies by Watts [1975], Foster [1977], Griffin [1977], and Brown and Rozeff [1979] have provided evidence suggesting the usefulness of three alternative, parsimonious time-series models for quarterly earnings data. These models, of the ARIMA class, have been identified with the researchers above and can be described as: (1) Foster's model (100) X (010) with drift, (2) Griffin-Watts' model (011 x (011), and (3) Brown-Rozeff's model (100) x (011) (hereafter referred to as F, GW, and BR, respectively).' Since these studies necessarily employed different samples of firms, time periods, and forecast horizons, and were conducted on alternative quarterly earnings numbers (i.e., eps vs. levels), it is probably unrealistic to expect a single parsimonious model to dominate across all studies. Moreover, uncontrollable sources of error due to sampling variation and potential small sample bias in parameter estimation could also have affected the reported descriptive and predictive findings in the above studies. In the present paper we reassess the validity of the above models for quarterly earnings using statistical goodness-of-fit criteria. We also search for alternative structures after controlling for a singular, potentially confounding variable-industry membership.2 If the diversity in parsi