Detecting Level Shifts in Time Series

This article demonstrates the difficulty that traditional outlier detection methods, such as that of R. S. Tsay, have in identifying level shifts in time series. Initializing the outlier/level-shift search with an estimated autoregressive moving average model lowers the power of the level-shift detection statisti cs. Furthermore, the rule employed by these methods for distinguishing between level shifts and innovation outliers does not work well in t he presence of level shifts. A simple modification to Tsay's procedure is proposed that improves the ability to correctly identify level shift s. This modification is relatively easy to implement and appears to be quite effective in practice.