Publication: Statistical method for detecting trend and seasonality in non stationary time series data
Loading...
Date
2012-01-01
Authors
Zaharudin, Muhammad Danial
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This research investigated statistical methods that can be used to detect trend and seasonality in time series data. It also improved the work that had been done by Muhammad Farid [7]. Muhammad Farid [7] had developed a framework and software for forecasting time series data that have trend and seasonality. However, the framework did not include methods for detecting trend and seasonality. In this research, two methods had been suggested for trend; simple moving average and regression analysis. Five methods had been suggested for seasonality; average graph, autocorrelation plot, correlation analysis, lag plot, and detrended graph. The latter was used to confirm seasonality in the case of conflicting results from the first four methods. Though all of the suggested methods are documented in the literature, the use of average graph and correlation analysis for seasonality are suggested for the first time in this research. Thirteen time series data sets from various fields such as agriculture, power utility, finance and sales were used to evaluate the methods. Excel-based templates for executing all of the methods except for autocorrelation plot were developed. Minitab software was used to generate the autocorrelation plot.
The methods produced graphs and statistical summaries such as autocorrelation
coefficients and correlation coefficients. This research also shows how to interpret the
graphs and the coefficients. In the final part of this research, the methods were
integrated into Muhamad Farid’s framework.