Sales Prediction for Fashion Retailer Based on State Space Model

Abstract

This study builds a novel model to predict sales for fashion retailers, an industry in which demand forecasting is notoriously challenging. Fashion sales are very volatile and are affected by new trends, highly seasonal patterns, and frequent promotional activities. Traditional forecasting methods often fail as they assume that sales patterns are stable over time. However, these patterns are always changing. In order to overcome this obstacle, the authors of this paper recommend the use of a state space model. This kind of model is very effective since it considers the actual sales level as a latent state which changes. It can also explain in a very organized manner how factors such as past sales, price changes, and holidays affect the demand to come. The main objective of this project is to develop and perform an experiment with this model using the real historical sales data of fashion retailers. We will examine the data and then apply the state space model to make sales predictions. To determine the model's success, we will gauge the forecasts' accuracy by computing the mean square error, which is a standard measure of prediction error. We will also compare its results to those obtained by a traditional method such as ARIMA to determine which is more accurate.