Abstract
The paper is aimed at analyzing the difficulties that arise when checking time series of field measurements for stationarity and identifying the main characteristics of the series. This paper provides a comprehensive review of methods for assessing stationarity in time series, highlighting the inherent challenges in primary analysis. The author emphasizes the limitations of graphical methods (including (P)ACF), statistical tests (such as ADF, PP, KPSS, ZA tests and CUSUM), and procedures like the variance ratio test, particularly when dealing with complex or real-world data. A novel, simple procedure is proposed to detect changes in the mean and variance of a series by analyzing the symmetry and distribution of aggregated samples of mean and variances of time series subsamples. The method is tested on synthetic data, including white noise, trend, seasonality, and heteroscedastic series, demonstrating its ability to visually identify non-stationary behaviors through constructed scatterplots and histograms. While not a formal stationarity test, the procedure offers a preliminary tool for characterizing time series behavior, complementing existing methodologies by focusing on fundamental statistical properties.

This publication has 25 references indexed in Scilit: