Surrogate Data for Non–stationary Signals

Most methods used in the field of linear and nonlinear time series analysis assume stationarity of the considered data. Non–stationarity is very likely to lead to wrong results. This is especially true for tests for nonlinearity. A common approach is to split the time series into segments which can be considered nearly stationary and perform individual tests. But for short time series or not too slowly varying non– stationarities these segments have to be made too short to meaningfully calculate a test statistic on them.