Time-Series Properties of Earnings: A Comparison of Extrapolative and Component Models
比较了单变量和多变量时间序列模型在预测会计收益上的表现，发现多变量模型能更好地捕捉收益成分间的内在关系，对会计和金融研究者有参考价值。
For several years now, accounting researchers have applied univariate time-series analysis to investigate the behavior of reported numbers (see Foster [1977]; Griffin [1977]; Watts and Leftwich [1977]; and Brown and Rozeff [1979]). Univariate methods examine the interstructure of a series, that is, the behavior of the series explained by its own past variability. When modeling the behavior of accounting data, however, the intrastructure between variables may also be relevant. Relationships exist between reported accounting numbers because economic events are interrelated and because the accounting measurement system imposes its own relationships. Consider the impact of investment in plant or equipment on earnings. A direct impact occurs through depreciation. An additional impact may occur through interest charges, depending on both the extent of debt financing and the cost of debt. Finally, since earnings is the aggregate of components such as revenue, depreciation, and interest charges, its behavior is also the result of component aggregation. Multivariate methods are able to model such intrastructure, while univariate