A Time-Series Analysis of Nonseasonal Quarterly Earnings Data
发现对非季节性企业使用季节性ARIMA模型会人为引入季节性、增加模型复杂度并降低预测能力，提出了一种简单有效的筛选方法以提升描述准确性和预测效果。
Research on the time-series behavior of quarterly earnings data has resulted in the identification of several different seasonal autoregressiveintegrated-moving-average (ARIMA) models as best fits for all firms.1 In this paper we are concerned about the fit of these models for a sample of firms which do not exhibit seasonal behavior. Specifically, each of the proposed parsimonious ARIMA models imposes seasonal differencing and/or seasonal moving-average parameters on all firms irrespective of whether or not they are seasonal. For nonseasonal firms this has the effect of inducing seasonality when none in fact exists, results in the identification of a needlessly complex model, and, as our results indicate, diminishes predictive ability. The relative ease with which nonseasonal firms may be culled from seasonal ones suggests that our procedure is a cost-effective means to enhance both descriptive validity and predictive ability in a quarterly earnings context.2 We find that the use of a simple, stationary autore-