Parametric and nonparametric aproach for trend detection in time series

Abstract

One of the most commonly used tools for detecting changes in time series is trend analysis. A number of parametric and nonparametric tests exist to detect the significance of trends in time series. The latter have been widely used mainly because of fewer number of assumptions needed in their implementation. The most often used test for detecting significant trends is Mann-Kendall test, that still requires sample data to be serially independent. To eliminate the effect of serial correlation on the Man-Kendall test different correction and pre-whitening methods have been introduced. This paper reviews the most commonly used approaches for trend detection in time series with or without presence of serial correlation. At the end these methods are applied to real datasets.