![IconGems-Print](images/IconGems-Print.png)
507 – Business, Time Series, and Spatial Analysis Methods
Nonparametric Smoothing of Time Series
James A. Shine
US Army Corps of Engineers (retired)
Smoothing of data helps in understanding the data-generating process, but, by its nature, it often obscures interesting observations. In time series data, one kind of interesting observations is a changepoint, that is, a point in the time series at which the data-generating process undergoes a change. The change may be in the general trend of the data, or it may be a change in volatility. These kinds of changes are of particular interest in economic and financial data, so it is important that a smoothing method not obscure the changepoints. The nonparametric alternating trends smoothing (ATS) technique is based on explicit identification of changepoints in the trends of the means in a time series. ATS performs favorably in changepoint detection when compared to other nonparametric smoothing methods such as moving averages. A disadvantage of the ATS approach, however, is that the underlying model assumes that changing trends change in sign.