In seasonal goods markets, products have short lifecycles and retailers widely use dynamic pricing strategies (Namin et al., 2017). In the fashion goods market, for example, retailers use markdown pricing, where they announce a retail (full) price for each product at the start of its selling season and announce one or more permanent price cuts (markdowns) until the end of the product’s shelf life. In this market, procurement lead times are often much longer than product shelf lives (6–7 months vs. 3–4 months), and fashions change quickly from season to season. These institutional details give rise to unique challenges for the retailer in pricing its products. First, due to the length of procurement lead times relative to the length of the selling season, the retailer needs to place a single order for each product well ahead of its selling season, and cannot replenish its inventory during the selling season. Second, since the end of the selling season is well defined and the salvage values are very low or zero, it is in the retailer’s best interest to sell-off the whole inventory for each product before the end of its selling season. Consequently, the retailer aims to maximize revenues from selling a fixed inventory over a finite time horizon and faces the trade-off of pricing too high and ending-up with leftover inventory at the end of the season versus pricing too low and leaving money on the table.

What further complicates the fashion goods retailer’s pricing decisions is that the retailer faces considerable demand uncertainty (Sodero et al., 2021, Soysal and Chintagunta, 2020, Shen et al., 2013, Guo et al., 2011, Araman and Caldentey, 2011, Aviv and Pazgal, 2008). As discussed earlier, the fashion retailer places orders and sets initial prices for the current season’s products six to seven months before the start of the selling season, without being aware of consumers’ preferences and their evaluations of individual products. This uncertainty is not limited just to the total demand potential of each product, but it is also related to the pattern of demand observed over time within the selling season. Consumers’ evaluations of individual products and purchase patterns vary dramatically and while some products may become immediate hits and show strong and steady demand very early in the season, other products may take time to establish a market, and yet others may exhibit a more stable demand pattern over the selling season. Faced with such uncertainty regarding the demand potentials and within-season demand patterns of individual products (Lazear, 1986, Pashigian, 1988, Pashigian and Bowen, 1991, Namin et al., 2017, Soysal and Chintagunta, 2020), fashion retailers, after setting an initial price for each product, observe demand early in the season and adjust product prices. A fashion retailer might, for example, announce earlier/shallower markdowns for products that have a slow start to stimulate demand and to dispose inventory before the selling season ends (Soysal and Krishnamurthi, 2012, van Ryzin and Talluri, 2005). In this paper, we suggest that fashion goods retailers can use fashion product characteristics (Pashigian & Bowen, 1991) to better estimate the demand patterns that products will follow within the selling season, which will help in deciding the timing and depth of the retailer’s markdowns. For instance, how will the within-season demand pattern vary between a highly fashionable expensive coat in a bright color and a classical gray coat sold at a lower initial price? Should these two items follow the same markdown path in the season, and if not, how should the price paths differ?

To answer these questions, we build a demand model for fashion products, which takes into account differences in demand across products based on their characteristics. The motivation behind our modeling approach is that consumers’ preferences and purchase (product adoption) behavior might vary across different types of products. Mapping the within-season variation in demand to product characteristics would enable us to not only decipher the impact of product characteristics on consumer preferences and purchase behavior, but it would also enable us to incorporate these cutting-edge insights into firms’ pricing decisions.

We utilize a unique data set from a leading specialty fashion apparel retailer of women’s and men’s coats. We apply an advanced latent class technique, a finite mixture model (FMM), to empirically estimate demand while allowing demand to vary by product characteristics. In our setting, there are latent classes (segments) of products with different demand patterns, and class memberships are identified by variations in fashion product characteristics such as product color, style, type, gender, and product season. Our demand model is defined as a mixture of normal distributions, and we show that the distributions of unobserved heterogeneity in our data across men’s and women’s coats are best explained by separate normal distributions corresponding to two latent/hidden classes of products. We name these two classes as stable demand vs. sharply-deteriorating demand products, based on their response to demand factors such as price and time since product launch.

After estimating demand, we use the demand model estimates for each product class to examine the revenue impact of alternative markdown policies (Agrawal & Smith, 2009). Specifically, we measure the improvement in revenue due to incorporating fashion product attributes in demand estimation and markdown pricing compared to a policy where the same markdown is used for all products regardless of their attributes. We provide empirical evidence that the cross-product heterogeneity in within-season demand is not trivial, and not taking it into account when making pricing decisions would significantly affect the retailer’s bottom line. Our analysis shows that ignoring the cross-product heterogeneity would result in a 5.77% reduction in the retailer’s revenues, which is a very large effect considering that retailer margins are very thin in this industry. Furthermore, our analyses provide insights to the retailer in terms of crafting its markdown policy for fashion items with different product characteristics. More specifically, our analysis shows that retailers should employ middle-of-the-season markdowns for coats in the sharply-deteriorating demand class, which are more likely to come in dark colors, be introduced in the summer or fall, or have a high initial price. On the other hand, the retailer’s markdowns should be earlier in the season and shallower for coats in the more stable demand class.

While the theoretical literature on the pricing of fashion goods is extensive, there have been only a limited number of empirical studies of markdown pricing by fashion retailers. We believe that our empirical analysis of a fashion retailer’s markdown pricing policy and demand patterns based on product characteristics is unique. We contribute to the current literature by: (1) developing a comprehensive empirical model that allows within-season fashion product demand to vary as a function of product characteristics; (2) empirically estimating this model using a rich and unique data set from a leading fashion apparel retailer and an advanced latent class analysis technique; (3) conducting counterfactual policy analyses based on the demand model estimates, showing that in order to maximize revenue, the retailer should implement different markdown policies for different classes of products; and (4) providing an expected magnitude of improvement in revenue for the case of conducting price markdowns based on fashion product characteristics.

Our empirically validated and unique pricing policy recommendations have significant managerial implications and can be used by retailers in the fashion industry to improve pricing and revenues. Our FMM demand estimation results can be used in forecasting the future demand for fashion products based on their characteristics, thereby reducing demand uncertainty. In other words, using fashion product characteristics as the input, our findings enable the retailer to predict whether each fashion item belongs to a stable or sharply-deteriorating demand class of products, even before it is launched to the market. In the age of big data, not only the demand uncertainty facing the seasonal goods retailers but also the need to make and implement real-time pricing decisions is on the rise. The FMM model can be used with existing software and converges quickly on a retailer’s aggregate sales data. Therefore, the model is suitable for use with large volumes of high velocity data. In sum, by determining the link between product characteristics and within-season demand variation, our model helps retailers to incorporate rich insights about customer preference and purchase behavior into pricing decisions.

The remainder of the paper is organized as follows. We begin with a summary of the relevant literature, followed by elaborating on our theory and a representation of our model. Next, we present our data and estimation processes, provide details on the counterfactuals, and include managerial implications. We conclude the paper by providing a summary of our work and describing future avenues of research.