TIME SERIES ANALYSIS: FORECASTING SALES PERIODS IN WHOLESALE SYSTEMS

Abstract

The main goal of time series analysis is explaining the correlation and the main features of the data in chronological order by using appropriate statistical models. It is being used in various aspects of life and work, as well as in forecasting future product demands, service demands, etc. The most common type of time series data is the one whose observations are taken in equally distributed time intervals (daily, weekly, monthly, etc.). However, in this paper, we analyze a different kind of time series which represents product purchase moments. Thus, since there are not any regular observation periods, this irregular time series must be transformed in some way before traditional methods of analysis can be applied.  After the data transformation is complete, the next step is modeling the nonstationary time series using commonly known models such as ARIMA and PNBD, which have been chosen for their fairly easy and successful forecasting processes. The goal of this analysis is timely product advertising to a customer in order to increase sales.

Unlike some other models that consider the relationship between two or more different phenomena, time series models, including ARIMA, Pareto/NBD and Poisson models, examine the impact of historical values of a single phenomenon on its present and future value. This approach enables the study of the behavior of a given phenomenon over time and produces good results, especially if a large amount of historical data is available.