Supermarketing

Supermarkets are frequently replacing products on the shelves with their own versions. These "store brand" alternatives have become very common in supermarket chains in many countries. In my local supermarkets (Malmö, Sweden) it is actually difficult to do grocery shopping without purchasing something of the store brand. While many are happy about the seemingly lower prices, there are some potential problems with the increased shelf space dominance of store brand products. Store brand products can unfairly extend the market dominance from the market for supermarkets to each product segment in which the supermarket introduces the store brand. In addition, and importantly, store brand products reduce the incentive for other brands to innovate. A prerequisite for these problems is that the supermarket has a dominant market position so that consumers have few alternative supermarkets to go to. This page provides some descriptive statistics on how a dominant supermarket chain gives preferential treatment to its store brand products.

Of course, the supermarket could simply foreclose all other brands. While many brands have indeed had some of their products removed from the shelves, such foreclosure may irritate customers and competition authorities. The supermarket can instead keep other brands on the shelves while nudging customers into purchasing the store brand products. How? Well, if you have worked in the industry, like I have, you know that the placement in the store is crucial for the success of a product. Hence, the supermarket can devote the best shelf space to the store brand products. I collected some data to investigate this further.

As you can imagine, manual data collection from the physical stores on shelf spaces and products would require an outrageous amount of effort and time. Instead I turned to the digital world, where data come in plenty and (somewhat) easily. I was interested in the major Swedish supermarket chain that has a market share of more than 50%. Luckily, many of the chain's stores across the country have their own online stores where customers can shop online for pick-up or home delivery. I selected 50 stores of various sizes and locations and started collecting data on all products offered in each store, and how they were displayed. I collected data weekly from July, 2018, through April, 2019. I thus have data on all products in these stores over 40 weeks. On average, a store offers 11 107 different products in 536 different product categories. 24% of the products are store brand products.

Product visibility - list positions

Similarly to shelf space locations, when products are listed on a website, the list position is crucial for sales performance. Customers are more likely to select a product that is listed at the top of the page than a product listed further down. Therefore, I collected information about each product's list position in its product category. When you enter a product category, e.g. "fresh pasta", the products are listed three on each row and you typically have to scroll to see more than the first row (at the time of data collection, and it depends on your screen resolution). You can see all products in the category by scrolling down the page. I define each product's list position starting from the top left (list position 1) and through all the products in the category (this is actually the order in the HTML code). For example, the leftmost product on the third row has list position seven. You can see below how the stores list the products depending on the product brand.

Each bar shows the proportion of products of the corresponding brand type that are displayed at that list position in the product category. The data cover all products in the 50 stores in the beginning of March, 2019. As an example of how to read the figure, the leftmost blue bar shows that almost 8% of the store brand products are displayed at the top left position in their corresponding product categories. For other brands' products, barely 5 % hold the top left position. Because the blue bars are higher than the orange bars in the left part of the figure, a larger proportion of the store brand products are displayed at the top positions than other brands' products are. Other brands' products are thus more likely to be found further down after scrolling past the store brand products. Now you might think that perhaps the store brand products are displayed at the top because of other attributes that customers like and makes the supermarket want to sell them. The next two figures show the equivalent comparison but instead divides the products based on whether they are organic or whether they are domestically produced.

The left figure shows that organic products are actually less likely to be displayed at the top. The right figure shows a similar pattern as for the store branding. Domestic products are much more likely to be displayed at the top than imported products. You might note the downward trend - bars are generally lower to the right. This is because many product categories have only a few products. In fact, 45% of the product categories contain less than ten products. One can think that list position nine is pretty good in a category with 60 products, but not so god in a category with ten products. I therefore scale the list position of each product by the number of products in the category. This is typically called the percentile rank. Thus, a product at list position five in a category that contain ten products has percentile rank 0.5, while a product at list position five in a category with 50 products has percentile rank 0.1. The next figure shows the distributions of percentile ranks for store brand products and other brands' products respectively.

To answer that, I compare list positions of store brand products and other brands products, while adjusting for organic label, domestic production, whether or not the product is on promotion, the price, and the time that the product has been sold by the store. In statistical terms, I run an Ordinary Least Squares regression with the list position on the left-hand side and the indicator for store brand on the right-hand side with the above listed control variables, and category-by-store fixed effects. By this estimation, the "adjusted" difference between store brand and other brands' products in the same category is -2.86 list positions (.044 in percentile rank). This means that store brand products are typically displayed three positions higher up (have a lower list position number) than the other brands' products that are similar on other characteristics. In other words, a store brand product is typically displayed one row higher up than a comparable product from a different brand.

As is typically the case with observational data, there is some information that is unobserved. When I say a "comparable" product, that means that the products are comparable in the observed characteristics that I adjust for. Potentially, there are characteristics that I do not observe that determine list position and that is different for store brand products compared to other products. For example quality. If store brand products are generally of higher quality, perhaps consumers should be nudged into buying them. Alternatively, if mark-ups are higher for store brand products, then it is profitable for the supermarket to nudge consumers into buying them.

Product introductions

It is also interesting to follow new products over time. Then we can see whether there is any difference in list position and evolution of prices between store brand products and other products. For that purpose, I define a product introduction as the first time a product appears in a store. Typically, a new product is introduced in most stores simultaneously, but because the stores decide for themselves, new products can be introduced at different times in different stores. The sample includes 50 086 different products that are observed at least once. Of these products, 13 126 products that did not exist in any store during the first week of data collection. On average, each store introduces 1 169 products during the 40 weeks. Some of these might of course be seasonal products and not completely new products. 19 percent of the new products are store brand products.

To see how list positions vary over time for new products, I compute each product's relative list position (pct rank as before) in each week after the product is introduced. This enables us to follow the visibility of each new product in each store. The following figure shows how visible the new products are by week after introduction, on average for store brand products and other brands' products.

Perhaps surprisingly, new products are not highly visible. On average, all new products appear in the bottom half of the page. In the first week, store brand products are actually listed further down than other brands' products are at introduction. After the first week, however, both types of products slowly climb up the list positions. The store brand products climb somewhat faster, and after five weeks they have more favorable list positions than the new products from other brands do. The store brand products then continue to be listed more favorably, as the figures above show.

Another interesting feature to follow after product introduction is the price. To facilitate the analysis, I normalize prices by dividing each introduced product’s current price by the price it had in the first week it appeared in the store. Thus, if a product is introduced at price P which then never changes, the normalized price would be 1 in every week after its introduction. The next figure shows the normalized prices in averages for each brand type respectively.

As the normalized prices never go below 1 (on average), prices typically increase after introduction. The increase is most distinct from the first week to the second. The increase could be because the supermarkets introduce products at relatively low prices to induce customers to try them. Then prices increase over time. The steady increase over time shown in the figure is not because all product prices increase incrementally week to week, but because the prices of many products increase at some point but at different points in time after introduction. The more weeks after introduction we consider, the more products have experienced a price increase, and therefore the averages increase slowly and steadily. In general, however, store brand products' prices increase more than the prices for other brands' products. Store brand products are thus introduced with larger discounts. Because customers oftentimes do not reflect on prices for goods they by regularly, setting a low introduction price attracts more adopters that may like the new product and decide to buy it again. Next time they buy it, they will not notice that the price is somewhat higher. This figure suggests that the stores lure more adopters to their own products at the time of introduction compared to products from other brands.

Obviously, the prices of competing products are important. A new product's price may increase over time but if it does not increase above substitute products, the customer may not notice (or at least not switch back). Therefore, it is interesting to see how the prices of recently introduced products change in relation to the prices of other products. For that purpose, I compute each product's price rank within the product category within the store in each week. The rank thus describes the products' position when all products in the category are ordered by price. As before with list position, number one has the lowest price, number two the second lowest price, and so on. To take into account that the number of products in the category differs, I compute the percentile rank. Thus, if the percentile rank is close to zero, the product is among the cheaper products in the category. If the percentile rank is close to one, the product's price is higher than most other products in the category. The following figure illustrates how the percentile ranks of new products' price generally vary over time, for store brand products and other products separately.

In contrast to what we might believe given the general price increases shown in the previous figure, price ranks generally decrease over time after introduction. This means that the new products become cheaper relative to other products in the category over time. If a new product's price increases over time, or at least does not decrease, there are two possible ways for the price rank to decrease:

1. The price of initially cheaper products increases above the price of the new product.

2. The supermarket stops selling one or several products that were initially cheaper than the new product. Alternatively, the supermarket introduces additional products of higher price.

In any of the two cases, the main conclusion is the same: it becomes harder to find alternatives that are cheaper than the newly introduced product. The decrease in relative price for new products is more distinct for store brand products. Hence, although store brand products generally increase more in price after introduction, their price position improves more than other brands' products do. This might be because when the store introduces its own brand, it more extensively increases the prices of substitutes. It can also be that the store is more likely to remove a low price substitute after introducing a store brand product compared to after introducing a product from a different brand.

The difference in levels between the two lines reflects that store brand products are typically cheaper than other brands. These data do not say anything about quality. It is of course possible that the store brand products are of equivalent quality but at lower prices. What the data can show is, however, that a product of the store brand is typically introduced with a relatively large discount. The price then increases, and the product is no longer as cheap as it was in the beginning. You might not notice the price increase because compared to substitute products, the new product is actually becoming relatively cheap.

This combination of increasing nominal price and decreasing relative price make some customers that do not switch pay more anyway. The price of the product these customers buy may increase, or the products may even disappear completely from the shelves. Because it is hard to keep track of all the prices of all the products that you occasionally purchase (and their substitutes), customers are unlikely to notice these price changes.

Should we worry?

Theoretical models of vertical integration (integration between a supplier and the retailer) and market foreclosure does not provide a clear answer about the impact on consumer welfare. There is, however, good reason to caution against the introduction of store brand products. One problem with a dominant supermarket introducing its own brand is that it reduces the incentive for other brands to innovate. Consider a soda brand that wants to introduce a new flavor. To reach consumers, it has to convince the dominant supermarket that this new flavor will be popular. In addition, there is a time span between notifying the supermarket until the product is on the shelves (six months is not unusual). Traditionally, competing soda brands had no knowledge of the new flavor before it appeared on the shelves, which gave the first soda brand some time of monopoly on the new flavor before competing soda brands could imitate and distribute the new flavor. Now, however, when the supermarket produces soda itself, they will know about the new flavor as soon as the first soda brand tries to convince them to sell it. Not only can the supermarket start imitating the new flavor earlier than competing soda brands traditionally could, but they can even refuse to launch the first soda brand's new flavor and then imitate it. This situation has lower incentives for innovation by food producers than if stores do not sell store brand products.

A well-known benefit of vertical integration is the elimination of double marginalization. Double marginalization is when both the supplier and the retailer set higher prices than what would optimize their joint profits because they do not take into account the effect of one's own price on the other firm. This is bad for both the firms and the consumers, who end up paying higher prices and purchasing fewer goods. Vertical integration can eliminate this problem. However, while vertical integration eliminates double marginalization for the integrating firms, it exacerbates the problem for the non-integrated firms, i.e. the rivals. When the foreclosed suppliers have fewer retailers through whom to sell, those retailers' market power increases. In the end, the introduction of store brand products and foreclosure of other brands in one supermarket may lead to higher prices in other supermarkets.

Furthermore, firms that function as "market regulators" are expected (by European competition authorities) to ensure fair competition in the market they "regulate". In this case, it means that the dominant retailer ensures fair competition between the suppliers. When the dominant retailer also operates as a supplier, there is a conflict of interest. The dominant retailer obviously has incentives to distort competition in favor of its own supply. This is typically called self-preferencing. This has recently become a hot topic in competition policy as the tech giants increasingly both operate a platform and simultaneously compete with actors on the platform. Google has been fined by the EU Commission for giving its price comparison website, Google Shopping, preferential treatment among its general Google search results. Amazon has been accused of displaying its own products favorably to the detriment of independent sellers and has even been forbidden to sell its own products on the platform in India.

The problem with store brands is that when the supermarket has substantial market power, consumers cannot easily switch to a different supermarket. Many supermarkets are local monopolies, and many customers are bound to their local supermarket by travel costs. That means that the supermarket has more room for misbehavior before losing customers. There is however some competitive pressure from other supermarkets, not least the up-and-coming pure online supermarkets. That competition, and potential scrutiny from competition authorities, is why the supermarkets use subtle tactics to steer customers to the store brand products without much contemplation. As these tactics are subtle, they are hard to notice. Thanks to the internet and thorough data collection and analysis, we can see more clearly what is actually going on.

Technical note on the dynamic figures

When moving along the lines in these figures, one thing that changes together with the number of weeks is the product composition. For example, the average list position ten weeks after introduction is computed using only the products that I observe for at least ten weeks after introduction. Because I have 40 weeks of data, products that are introduced in the later part of the data collection period are observed for fewer weeks. Therefore, when considering longer time after introduction, such as 20 or 30 weeks, I cannot use the products that were introduced in the end of the data collection period. They are, however, included in the measures for the earlier weeks after introduction.

If products that are introduced later are consistently different from the previously introduced products in the aspects that the figures measure, that is a problem. What may look like a trend over the relative time after introduction may instead reflect a static difference between products depending on when they were introduced. To make sure the figures shown on this page are not subject to such bias, I reproduce each figure up to relative week 5, 10, 15, 20, 25, and 30 separately. For each number of relative weeks, I discard all products observed less than that many weeks. The resulting trends are more or less identical to the trends shown in the corresponding figures on this page. Hence, the compositional change with the number of weeks after introduction does not affect the conclusions. Please let me know if you would like to see these additional figures and I'll send them right over.