Sequential Detection of US Business Cycle Turning Points: Performances of Shiryayev-Roberts, CUSUM and EWMA Procedures

In this paper we consider the problem of sequential detecting change points in economic time series. We compare the performances of three well known procedures, Shiryayev-Roberts, CUSUM and EWMA, in the problem of early detection of the US business cycle turning points using leading indicators or some financial series. The comparison was done separately for detecting recessions and expansions during the period of 1955-2003. We found that in most cases the Shiryayev-Roberts procedure is superior to the other two in detecting turning points with leading indicators. At the same time the CUSUM procedure performs better in detecting turning points with stock price indices.