What Is a Digital Cookie Worth?
37 Pages Posted: 2 Apr 2016 Last revised: 6 Jul 2021
Date Written: March 31, 2016
Tracking a user’s online browsing behavior to target them with relevant ads has become pervasive. There is an ongoing debate about the value of such tracking and the associated loss of privacy experienced by users. We inform this debate by quantifying the value of using different kinds of potentially intrusive information in targeted advertising. We collect a large proprietary dataset with over 1.3 million individual impression-bid-level observations. The data has detailed cookie information, as well as the bids placed by the firm for serving ad impressions. We also know whether a user saw the ad and whether a purchase occurred. First, we find that using more information from cookies increases the accuracy of prediction of purchases, but at a decreasing rate. We also find that firm’s bidding decision (how much to bid for an ad) can be accurately predicted by cookie information. We then estimate the effect of an ad on a user’s purchase probability. In particular, we examine whether users who have a high baseline purchase probability are also more likely to be influenced by ads. We find that on average ads do not have a statistically significant impact on purchase probabilities of consumers. However, individuals who have a high baseline purchase probability, do respond positively to ads and ads can increase their purchase probability by up to 2.7 percentage points. To overcome potential endogeneity in ad placement, we use an instrumental variable and find that these results are robust. Finally, we simulate different policy regimes by restricting different kinds of user information from being used for targeted advertising and quantify the impact such restrictions have on sales. We find that restricting more intrusive variables for targeting lowers ad effectiveness and leads to fewer potential purchases.
Suggested Citation: Suggested Citation