Stochastics and StatisticsForecasting demand for single-period products: A case study in the apparel industry
Introduction
In the apparel industry, three prominent developments contribute to the complexity of forecasting: shortening product life-cycles, increasing product variety, and globalization of sourcing and manufacturing. The impact of each of these developments will be discussed in detail.
Ninety-five percent of SKUs (Stock Keeping Units) change every selling season (Gutgeld and Beyer, 1995). Because product life-cycles are short, there are no historical demand data that can be used to obtain a priori demand forecasts, i.e. before any demand has been realized. Furthermore, the number of in-season replenishment opportunities after observing demand and updating forecasts accordingly is limited and the risk of product obsolescence is high.
Due to global competition, faster product development, technological advances, increasingly flexible manufacturing systems, and more demanding consumers, an unprecedented number and variety of products are competing for demand (Fisher et al., 1994). As a result, the volume of sales per SKU is very low (Gutgeld and Beyer, 1995), and demand for SKUs within the same product line can vary significantly (Abernathy et al., 2000). Thus, even if aggregate demand can be predicted with some certainty, it is very difficult to predict how that demand will be distributed over the many products that are offered. The complexity of production planning and ordering increases accordingly.
The bulk of products is produced in South East Asia, and hence the lead time to Western retailers is long. The typical lead time from fabric manufacturers is 3 months (Gutgeld and Beyer, 1995). The specific mail order company that we study faces lead times of 6–14 weeks. Long lead times dictated by powerful suppliers and strong competition from other retailers trying to secure enough production capacity force retailers to commit to initial order quantities long, usually several months, before the start of the selling season. We remark that there are a few well-known apparel/fashion retailers like Zara and Hennes & Mauritz that have deviating strategies based on local manufacturing. However, for the large majority of apparel retailers, who have a low-cost focus and often sell private label products, the situation is as we described.
So, for most products that an apparel retailer sells in any season, a demand forecast is needed well before the start of the season, when no historic demand data is available. An apparel retailer’s success hinges to a large extent on the accuracy of those pre-season forecasts on which the initial orders are based. These initial orders comprise the bulk of the total volume ordered (Fisher and Raman, 1999). Additional in-season replenishment opportunities, if available, are essentially emergency replenishment opportunities and only serve to prevent shortages resulting from possible initial underestimation of demand. We refer interested readers to Mostard and Teunter (2006) for a further discussion and analysis of inventory control issues. In this paper, the focus is on forecasting.
To obtain maximum accuracy of initial forecasts, there are two common practices for gathering relevant information. First, a so-called preview is common in the apparel industry (see, e.g., Tang et al., 2004, Fisher and Rajaram, 2000, Chambers and Eglese, 1986). During a two to five week period before the start of the selling season (the preview period), customers can pre-order products at a small discount. Second, many apparel retailers use a committee of experts (e.g. purchasers, planners) to provide forecasts for individual products (see, e.g., Mantrala and Rao, 2001, Fisher et al., 2000, Raman, 1999).
In this paper, we propose new forecasting methods based on advance demand information, and perform a case study to compare them to existing ones based on advance demand information and also to methods based on expert judgments. Numerical results are obtained using data from a large mail order/Internet retailer based in the Netherlands. This company currently bases its forecasts on advance demand information. Based on a data set of around seven hundred SKUs and for two successive summer seasons, we compare the accuracy of the various methods based on advance demand information. For a smaller subset of around one hundred SKUs, we also obtained forecasts from a number of company experts. For this subset, we compare methods based on these expert judgments to methods based on advance demand information.
The remainder of the paper is organized as follows: We will first give an overview of the relevant literature in Section 2 and outline our contributions. In Section 3, we describe the different forecasting methods. Section 4 introduces the case company and available data in more detail. The empirical results are described in Section 5. Finally, we present our conclusions, discuss limitations, and provide directions for further research in Section 6.
Section snippets
Literature review: forecasting demand for single period (fashion) products
Fashion products in general are characterized by high demand uncertainty, high stockout costs and a high risk of obsolescence (Lee, 2002). Although the specific mail order company that we study can be classified as an apparel company rather than a fashion company, it shares these characteristics. This is evidenced by the fact that the company frequently has significant leftovers of individual SKUs which cannot be carried over to the next season and need to be sold at high markdowns. Customer
Forecasting methods
We will discuss methods that forecast based on advance demand analysis in Section 3.1, and then continue in Section 3.2 with expert judgment methods. For all methods, we let denote the set of SKUs in an upcoming selling season for which demand forecasts are needed, and N denote the number of SKUs in .
Application: a large mail order/Internet retailer
The case company is a mail order/Internet apparel retailer operating only in the Netherlands. It divides each year into two selling seasons, spring–summer (December–June) and autumn–winter (June–December). One main catalogue is issued per season, and several smaller catalogs appear throughout the year, containing special collections or special offers aimed at specific groups of customers. A total of around 80,000 SKUs are offered, distributed over three collections: apparel, furniture and
Empirical results
In this section, we will compare the five forecasting methods using the case study data described in Section 4. Methods 1–3 will first be compared for the full data set (assortment groups 1–3) and in their forecasting accuracy for Season 2 (based on Season 1) and Season 3 (based on Season 2). Then, for Assortment group 1 and Season 3, Methods 1–3 will also be compared to Methods 4 and 5 based on expert judgment.
We used three different performance measures of forecast accuracy: mean absolute
Conclusion and recommendations
For a large set of SKUs and in two successive selling seasons, we have compared the accuracy of three quantitative forecasting methods based on advance (preview) demand information. The methods all use a top-down approach, and first forecast the aggregate total demand for a group of SKUs by scaling up the aggregate preview demands. They differ in the subsequent division of that aggregate forecast over the individual SKUs; proportional to preview demand (Method 1), equal (Method 2), or top-flop
References (36)
- et al.
Fashion retail forecasting by evolutionary neural networks
International Journal of Production Economics
(2008) - et al.
Forecasting demand for mail order catalogue lines during the season
European Journal of Operational Research
(1988) Pre-season stocking and pricing decisions for fashion retailers with multiple information updating
International Journal of Production Economics
(2007)Applying Bayesian methodology with a uniform prior to the single period inventory model
European Journal of Operational Research
(1997)- et al.
Production under periodic demand update prior to a single selling season: A decomposition approach
European Journal of Operational Research
(2008) - et al.
The distribution-free newsboy problem with resalable returns
International Journal of Production Economics
(2005) - et al.
The newsboy problem with resalable returns: A single period model and case study
European Journal of Operational Research
(2006) - et al.
The efficacy of using judgmental versus quantitative forecasting methods in practice
Omega,the International Journal of Management Science
(2003) - et al.
A neural clustering and classification system for sales forecasting of new apparel items
Applied Soft Computing
(2007) - et al.
Control your inventory in a world of lean retailing
Harvard Business Review
(2000)
Forecasting demand variation when there are stockouts
Journal of the Operational Research Society
Database models and managerial intuition: 50% model + 50% manager
Management Science
Use of preview exercises to forecast demand for new lines in mail order
Journal of the Operational Research Society
Estimation of forecast errors for seasonal-style-goods sales
Management Science
Accurate retail testing of fashion merchandise: Methodology and application
Marketing Science
Reducing the cost of demand uncertainty through accurate response to early sales
Operations Research
Cited by (44)
Demand forecasting for fashion products: A systematic review
2024, International Journal of ForecastingRetail forecasting: Research and practice
2022, International Journal of ForecastingCitation Excerpt :Before each season, there is a need for a forecast of total product sales, whether distributed via the high-street or online, and lead times may be lengthy. For many years, a variant of the problem arose through the use of mail-order catalogs (see e.g. Mostard et al., 2011), though the increased prevalence of online has left the catalog as merely an adjunct to online marketing. Typically, analogous products are used, though more advanced methods can incorporate product attributes as well as their sales.
Robust decisions for regulated sustainable manufacturing with partial demand information: Mandatory emission capacity versus emission tax
2022, European Journal of Operational ResearchResearch manuscript: The Bullwhip Effect in rule-based supply chain planning systems–A case-based simulation at a hard goods retailer
2021, Omega (United Kingdom)Citation Excerpt :Indeed, some orders may not follow the predefined logic when demand planners believe the actual demand will differ from the programmed forecast. In Mostard et al.’s study [80], the expert-based forecasting methods surpass all of the algorithm-based ones considered. In PA, demand planners in charge of a product category can autonomously modify the forecast insofar as they believe it will better represent true demand.
Determination of shipping timing in logistics warehouse considering shortage and disposal in textile industry
2019, Procedia ManufacturingNews-based forecasts of macroeconomic indicators: A semantic path model for interpretable predictions
2019, European Journal of Operational ResearchCitation Excerpt :Hence, it is likely that these forecasts Numerous studies have analyzed professional forecasts for their predictive performance and ability to identify potential biases (Blanc & Setzer, 2015; Mostard, Teunter, & de Koster, 2011). In the field of macroeconomic forecasting, a recent study (Dovern & Weisser, 2011) finds that the distribution of a forecasts accuracy varies significantly across indicators, forecasters and nations.