Mixed pricing in online marketplaces

Federal Trade Commission v. Staples, Inc., 970 F. Supp. 1066 (D.D.C. 1997) grants the FTC’s motion for a preliminary injunction.

“Websites Vary Prices, Deals Based on Users’ Information”, Jennifer Valentino-Devries, Jeremy Singer-Vine and Ashkan Soltani, Dec 24 2012.

“Office Depot, Inc. Announces First Quarter 2014 Results”, Office Depot Press Release, May 6 2014.

“FTC Challenges Proposed Merger of Staples, Inc. and Office Depot, Inc.”, FTC Press Release, Dec 7 2015.

See https://​www.​ftc.​gov/​sites/​default/​files/​documents/​closing_​letters/​office-depot-inc.​/​officemax-inc.​/​131101officedepo​tofficemaxstatem​ent.​pdf.

Additional credit should be given to computer science researchers who expose such practices, such as Mikians et al. (2012) which shows significant variation in prices resulting from location, cookie data, and equipment. Mikians et al. (2013) presents results of a fascinating experiment that gathered price data from users with a certain browser toolbar, and found widespread price differences on many popular websites.

Later in 2013, the firm redesigned its website and there is currently no detectable effect of the contents of the zip code field in the cookie. While it is possible that they have discontinued this pricing strategy, potentially in response to the news coverage it received, it is also plausible that they simply store location data on their servers instead of in cookies, making it impossible to manipulate the latter.

Three outliers that show high prices far more than others are unusual office supply purchases: store-brand aspirin, acetaminophen and bath tissue. For the remaining three outliers, we have far fewer price scrape observations, since they were put on a uniform sale for a large portion of our time period and we drop all prices marked as sale prices.

In making this comparison, we had to drop one of the five products for Staples. The particular Avery address label we chose was both on sale and on rebate when the data was collected, and the mystery shoppers did not uniformly include both in the online price they recorded, making it impossible to determine whether both the sale and the rebate were available to online shoppers from all zip codes or whether they were simply not included in the displayed price.

Our need to calibrate the low type’s price responsiveness under the special case of α = 0 highlights a short-coming of the “Breadth” data. In the theoretical model, the optimal low price and mixing frequencies are interrelated as shown in Eq. 6. Ideally, we would jointly estimate the bargain hunter’s price responsiveness and the type distribution α that rationalize the two observed outcomes of low price and price mixing frequency. We are unable to follow this approach, as it would require variation in the low price to pin down β _L . Our alternative Depth data set, while covering a larger number of products, shares this shortcoming. It furthermore does not allow us to investigate determinants of the observed variation in mixing probabilities across zip codes as its geographic reach is limited.

As a robustness check, we used our calibrated Logit preferences, the empirical estimates of the search intensity ρ _z and the share of loyal customers α _z to jointly solve the system of first-order conditions characterizing the optimal low price and optimal price mixing frequency in Eq. 6. In finding the optimal low price, we only impose the empirically observed constraint that it is constant and weigh each zip code by its population as a measure of market size M. The resulting optimal uniform low price is only 1.68 % higher than the observed low price that we employ in estimating ρ _z and α _z , suggesting that the calibrated preferences yield internally consistent pricing strategies.

We omit non-standard zip codes (i.e. post office boxes, military bases) and any zip codes with missing demographic data.

The confidence interval is (0.9644,0.9762). Here, we constrain the constant to be zero; a regression with a constant yields a coefficient of 0.823.

For maps of the store locations of both chains prior to the merger, see http://​mikejking.​com/​office-depot-office-max-store-closures/​, accessed 12/5/2015.