A Mean Squared Error Criterion for Time Series Data Windows
提出一种周期图均方积分误差准则，用于评估时间序列数据窗（锥形窗）的性能，并发现锥形化在谱峰尖锐时极为有效，且最优锥形点数随样本量增大趋于常数。
A periodogram mean integrated squared error criterion is proposed for assessing the performance of time series data windows, or tapers. The criterion is evaluated numerically for several known autoregressive processes to determine the optimal amount of tapering in Tukey's cosine window. It is found that tapering can be extremely beneficial if the spectrum contains sharp peaks. Also, in all cases studied, the optimal number of points tapered converges to a constant as the sample size increases. This result is at odds with the traditional strategy of tapering a fixed percentage of the data. Finally, a method is proposed for estimating the optimal degree of tapering on the basis of data from an unknown process. Some Monte Carlo evidence is presented to demonstrate the effectiveness of the proposed estimation technique.