Seasonal Time Series Forecasting using SARIMA and Holt Winter’s Exponential Smoothing

Abstract

The purpose of this study is to compare SARIMA and Holt-Winter’s Exponential Smoothing methods in an attempt to generate customer transaction forecasting in Store X with high accuracy.This study will compare the results of sales forecasting with time series forecasting model of Seasonal Autoregressive Integrated Moving Average (SARIMA) and Holt Winter’s Exponential Smoothing method. SARIMA model still accurate when the time series data is only in a short period, this model is accurate on short period foracasting but less accurate on long period forecasting. Meanwhile Holt Winter’s Exponential Smoothing accurate on forecasting seasonal time series data, either it’s pattern shows trend or not. Both models are compared with forecasting data showing seasonal patterns. The data used is the data of clothing retail store sales from 2013 to 2017. Accuracy level of each model is measured by comparing the percentage of forecasting value with the actual value. This value is called Mean Absolute Deviation (MAD). Based on the comparison result, the best model with the smallest MAD value is SARIMA model (1,1,0) (0,1,0)12 with MAD value 5.592. From the comparison results can be concluded that the SARIMA model is feasible to be used as a model for further forecasting.