Calendar Effects in Monthly Time Series: Modeling and Adjustment
本文提出了一种新的月度时间序列日历效应调整方法，通过月长校正、幂变换和稳健回归估计日历成分，并利用残差图、谱分析等检验模型有效性，适用于经济统计数据的预处理。
Abstract Most monthly time series that represent a total of some variable for each month contain calendar effects due to changing month length, day-of-the-week effects, and holidays. It is important to remove the calendar variation to allow an effective assessment of the variation due to important factors. New procedures for calendar adjustment are presented in this article. A plausible model for the daily data is used to derive a model for the monthly data in which the power transformed, month-length corrected data are equal to trend plus seasonal plus calendar plus irregular. The procedure for fitting the calendar component in the monthly model is (a) divide the aggregated monthly data by month length and multiply by 30.4375, the average month length; (b) choose a power transformation; (c) remove trend and seasonal components; (d) estimate the calendar parameters by robust regression. Since the model is only a hypothesis it is important to check its validity. This can be done by various residual plots in the regression analysis and by using spectrum analysis and time-domain graphical methods to detect residual calendar effects in the adjusted series. This approach is compared to the X-11 calendar estimation procedures.