The effect of product variety in multiproduct retail pricing: the case of supermarkets

ABSTRACT

The search literature offers several explanations as to how a multiproduct retailer’s offering of a larger variety may affect consumers’ search behaviour and hence prices. Among them are (1) higher sales to be generated when lower prices discourage consumers’ further search; and (2) more pricing power due to improved consumer expectations about consumer-product match valuations. The former concerns independent categories and derives a negative relationship while the latter concerns substitutes within categories and derives a positive relationship. To examine the effect of a retailer’s variety offerings on the price levels using scanner data from the supermarket industry, we devise metrics of assortment width and depth, where the former measures variety in category (more independent) offerings and the latter, variety in brand (more substitutable) offerings within categories. We find that the effect of an additional offering to each dimension of assortment decreases in the other dimension, leading the effect of assortment width to be negative while the effect of the depth to be positive for the majority of the stores used in our analysis—a result that reconciles the theories that purport to draw opposite conclusions.

KEYWORDS:

• Multiproduct retail pricing
• product variety
• supermarkets
• consumer search

JEL CLASSIFICATION:

• L81
• L20

Acknowledgment

Hyunchul Kim is grateful for the financial support by the BK21 FOUR Project of the Department of Economics at Sungkyunkwan University.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ The widespread use of online shopping would most likely complicate our analysis. The effect of an additional product offering may differ not only across products depending on its availability for online purchase but also across stores depending on whether they are patronized by more or less of online-shopping savvy consumers. While how the value of physical availability is affected by the rise of online shopping in the supermarket industry is certainly interesting, it is beyond the purpose of this study.

² Because a product is recorded only when it’s sold, it is omitted from our measurement of product variety if it was placed on the store’s shelf, but never sold over the whole year of 2006. While this may add to the concern about hidden factors affecting both product variety and prices, we believe the effect should be minimal as we include in our tally all of the products ever sold over the year. And, the omission of a product never sold over a year, if any, may be justified because it is unlikely to affect the prices of other products.

³ This shallowness in the assortment depth within brands sources from the fact that the depth $d_j^c=\sum _b{\omega }^{bc}{r}_j^{bc}$ is a weighted average of $r_j^{bc}$ while the weight ${\omega }^{bc}$ and $r_j^{bc}$ are negatively correlated in our data. The simple average of $r_j^{bc}$ reveals that conditional on a store carrying a brand, it carries 25% of the products of that brand. The negative correlation between ${\omega }^{bc}$ and $r_j^{bc}$ is likely due to the following phenomenon: a popular brand offers a variety of packaging sizes, flavours, etc. under the brand (${\omega }^{bc}$ and $n^{bc}$ are indeed strongly positively correlated), but sales and purchases are concentrated only on a small fraction of the products, and stores tend to carry those products only.

⁴ The event of TPR and the event of display or feature are highly correlated. The average probability that a TPR of 5% or larger is placed on a product in any given week is 0.25, and when such a TPR is placed on a product, the probability of it being advertised is 0.20. The probability of a product being advertised without a TPR of 5% or larger is only 0.03.

⁵ In addition, as we assign a single price (modal price) to each product and include all of the products ever sold during the whole sample year, our analyses are free from seasonal or incidental variations such as one caused by a decrease in the prices and an increase in the number of products in display during a stock clearing sale.

⁶ DeHoratius and Raman (2007) show a change in store manager incentive design is related to inventory shrinkage using data from a consumer electronics store chain. A high level of inventory shrinkage or general lack of attention to inventory management may discourage offering a large variety of products.

⁷ One possible exception is floor areas. Stores in areas with low housing and rental prices may have large floor areas that enable them to offer a large variety of products and may also set low prices faced with more price-sensitive consumers. As such, the omission of floor areas may generate a downward bias on the relationship between variety offerings and prices.

⁸ While economic and business research tends to presume consumers desire and benefit from variety, several behavioural marketing studies have shown using experimental data this may not be always the case (Diehl and Poynor 2010; Gourville and Soman 2005; Chernev 2003). Kuksov and Villas-Boas (2009) show that too many or two few product offerings may lead consumers to search or choose less when evaluation of each alternative is costly.

⁹ The exact number of stores with the described effect of the rival store’s depth is 12,484 and that for the rival store’s width is 10,452. But, only 9,275 stores belong to both groups—each dimension of the rival store’s variety offerings affecting the store’s prices and advertising activities differently.

¹⁰ Simultaneously, a categorization based on more categories tends to shift some variations in $D$ to variations in $W$, which results in an increase in correlation between assortment width and depth. See Appendix A1 for explanation.

¹¹ We do not use quantities or sales revenue as these are highly endogenous, more so, we believe, than the number of purchase trips or the number of sales week. By counting purchase trips and sales weeks, we may underweight products that are often purchased in bulk for storage or stockpiling purposes while overweighting more easily perishable products. Nonetheless, counting purchase trips and sales weeks likely better captures the importance of a product category or a brand in relation to the decision as to whether to visit a store that does and does not stock the category or the brand.

¹² If we only include trips to the stores that can be classified as grocery stores, the average number of trips drops to 2.7 trips per week.

¹³ For the calculation of trip frequencies, we augment the 2006 trip data with the 2007 trip data. The purpose is to maximize the number of RMS stores covered. When comparing the list of stores appearing in either of the 2006 HMS or the 2006 RMS data to the list of stores in the 2007 HMS data, we find the difference is trivial, pointing to rare entry and exit events that could alter the panel households’ shopping patterns.

¹⁴ See Frey and Dueck (2007).