Sales forecasting is a key factor for success in the supply chain and inventory management for most industries. It is especially true in the textile and apparel industry where the lead time is very long compared with the lifespan of the products. However, many other constraints make the sales forecasting very complex in this competitive market, such as seasonal sales that are very sensitive to the weather, the very volatile consumer demand, the new collections that provide new products without any historical data, the huge variety of products, and the many exogenous factors that disturb the sales. For these reasons, sales forecasting of textile and apparel products has led to many works in the literature.
Generally, fast fashion retailing requires more specific sales forecasting systems and, consequently, makes the forecasting challenge more difficult.
In fashion industry, the consumer demand can be considered to be very volatile. Indeed, the offer is very large and the consumer need has to be satisfied immediately, or the consumer switches to another brand or store. Thus, companies have to keep a high service level and a very competitive price.
Generally, traditional brands ensure the low price of their products with supplies in low-cost countries that generate a long lead time. The service level is obtained with inventories in warehouses that are often difficult to manage and are expensive.
The market is strongly influenced by numerous factors which make the sales very variables. These factors are also called explanatory variables, are sometimes not controlled and even unknown. Some of them involve in an increase of the purchase decision, others modify the store traffic. Hence, the difficulty to exactly identify them and to quantify their influence.
Thus, to deal with these obstacles, companies have to rely on efficient and accurate sales forecasting systems. These systems should be perfectly suited to the requirements of the clothing market. First, it is important to know the product, the sales features, and how the retailer will use the forecasts. Then, the forecasting technique can be wittingly selected.
Sales Forecasting Techniques:
In the field of the sales forecasting techniques, statistical techniques are certainly the most used ones. These statistical methods include various well-known models that have formal statistical foundations: exponential smoothing, Holt Winters model, Box & Jenkins model, regression models, or auto regressive integrated moving average (ARIMA) models.
In spite of providing satisfactory results in some domains and being popularly used for their simplicity and fast speed, statistical techniques suffer a few problems when they are implemented for the clothing sector.
Indeed, these techniques are not suitable for the textile and apparel environment and more generally in any fashion sectors, especially because most time series methods require large historical data sets, a complex optimization of their parameters, and a certain experience of the user, and they are often limited to a linear structure. Thus, their performances are usually worse when they are compared to more sophisticated methods such as artificial intelligence (AI).
Recently, different reviews on sales forecasting techniques for fashion retailing have been proposed in the literature. A study of forecasting techniques over time shows that the implementation of AI methods (pure or hybrid) is growing in the last decade. These methods are able to respond to the main requirements of the fashion and apparel sector.
To be specific, fuzzy-based systems are used to deal with long-term forecasts. Some AI models are also developed for the complex issue of sales forecasting of new items without or with a limited amount of historical data. These models are based on clustering and classification techniques using decision trees, ANN, support vendor machine, or Grey model combined with ANN.
In many cases the best results are obtained with hybrid models, such as by mixing AI with times series techniques or AI techniques together. Indeed, the combination of techniques enables one to overcome the deficiencies of single models and to improve forecasting performance. Finally, more recently, AI techniques have been specifically used for fast fashion forecasting issues. However, the most used technique in the literature to respond to the fast fashion constraints is obviously the extreme learning machine (ELM). ELM has demonstrated better learning capabilities, such as processing time and generalization, compared to the ANN with a gradient-based learning algorithm. Other benefits are also given to ELM in the literature, especially its ability to avoid many difficulties associated with gradient-based learning methods, such as stopping criteria, learning rate, learning epochs, local minima, and the over-tuned problem. For these reasons, ELM has been widely used in fashion sales forecasting.
However, if ELM appears as faster and more efficient on limited data than the classical ANN, some drawbacks have been pointed out. To be specific, in some cases, it is blamed that ELM is unstable or requires more time than expected to achieve a reasonable forecast. However, these problems also arise with other pure statistical and AI methods. Consequently, different enhancements of ELM, often based on hybrid models, have been proposed to provide better forecasting systems to fashion retailers.
For instance, a new ELM has also been successfully implemented in the specific context of fast fashion forecasting. The proposed hybrid model based on ELM and the Grey model particularly responds to the limited data andf time constraints and provides satisfactory forecast results.
However, as to the best of our knowledge, no investigation has been performed on the implementation of this technique on the whole fast fashion supply process that includes the assortment policy and the two-level forecasting stages: the preseason forecast (long-term forecast) and the ongoing season forecast (short-term forecast). Thus, to quantify the real benefits provided by this method for fast fashion retailing.
Sales forecasting is very important for many retail operations. It is especially critical for the fashion retailing service industry in which product demand is very changeable and product’s lead time is short. In this article, I have conducted a comprehensive review of the literature on fast fashion retailing sales forecasting. I have explained the drawbacks of different kinds of analytical techniques for fast fashion retail sales forecasting. I have also explored the pertinent issues related to real-world applications of the fashion retail sales forecasting models.
- Information Systems for the Fashion and Apparel Industry Edited by Tsan-Ming Choi
- Volume 2013, Mathematical Problems in Engineering; Sales Forecasting for Fashion Retailing Service Industry: A Review by Na Liu, Shuyun Ren, Tsan-Ming Choi, Chi-Leung Hui, and Sau-Fun Ng