Testing for Linear and Nonlinear Gaussian Processes in Nonstationary Time Series

Abstract

Surrogate data methods have been widely applied to produce synthetic data, while maintaining the same statistical properties as the original. By using such methods, one can analyze certain properties of time series. In this context, Theiler's surrogate data methods are the most commonly considered approaches. These are based on the Fourier transform, limiting them to be applied only on stationary time series. Consequently, time series including nonstationary behavior, such as trend, produces spurious high frequencies with Theiler's methods, resulting in inconsistent surrogates. To solve this problem, we present two new methods that combine time series decomposition techniques and surrogate data methods. These new methods initially decompose time series into a set of monocomponents and the trend. Afterwards, traditional surrogate methods are applied on those individual monocomponents and a set of surrogates is obtained. Finally, all individual surrogates plus the trend signal are combined in order to create a single surrogate series. Using this method, one can investigate linear and nonlinear Gaussian processes in time series, irrespective of the presence of nonstationary behavior.

Using missing ordinal patterns to detect nonlinearity in time series data

“…Hence, the NMP of a nonstationary series maybe different from those of its IAAFT surrogates, resulting in a potential false rejection of the null hypothesis. Improved surrogate methods have been proposed to overcome this drawback [38,39].…”

A comprehensive study of the delay vector variance method for quantification of nonlinearity in dynamical systems

“…In this regard, a theoretically well-established technique for detecting the nature or nonlinearity of time series is the surrogate data method [ 10 ], which was originally motivated by statistical hypothesis testing and presents an indirect way of detecting nonlinearity in a signal [ 11 ]. In recent literature, there is extensive research on signal linearity analysis based on surrogate data [ 12 , 13 , 14 ]. Failure to detect nonlinearity may result from an inappropriate choice of test statistic and there are also problems with artefacts occurring in the process of generating surrogate datasets [ 15 ].…”