Sales forecasting in fashion retail industry with classical and machine learning methods

Abstract

This study includes the usage of classical econometric methods and machine learning models for short-term demand forecast of a fashion retail firm with hundreds of shops. Forecasting performances of different methods are compared. Retail clothing sector has an important place among all sectors both in Turkey as well as all over the world. According to the data of the global textile and clothing sector exports, Turkey has become one of the most important countries in this sector with 12 billion USD dollars in exports by 2018. This sector, which includes different functions from planning to production, logistics to merchandising, is a dynamic sector in which rapid changes occur due to impact of fashion and trend. Rapid adaptation to changes in this dynamic sector can contribute to the profitability of companies. At this point, demand forecasting plays an important role for companies to organize their activities and take necessary measures in advance. Demand forecasting affects many points in this sector, from production planning to shop planning. Long-term demand forecast is an important input for many activities for the company, from labor investment to managing financial resources, opening new stores, and determining the collection mix. The most important activities influenced by the short-term demand forecast in this sector that are logistics and merchandising activities. While the unpredictable demand can cause a bottleneck or idle capacity in logistics warehouses. On the merchandising side, less inventory than demand case can cause loss of sales while inventory more than demand can causes to inventory cost in the store and disruption of merchandising operations due to physical space constraints. In this study, demand estimation is performed for 2 different styles of 30 different stores at the daily level. This daily estimate is a requirement for the decision of how many products should be shipped to which store.