Abstract
In recent years, progresses in data mining and business analytics have fostered the advent of recommender systems, behavioral advertising, and other ways of using consumer data to personalize offers and products. We investigate the incentives for sellers to invest in systems that allow the tracking of consumers and then to truthfully report whether potential buyers will enjoy yet untried products. We find that there are two types of equilibria: For some parameter values, sellers will target all potential buyers, hence their targeted ads or purchase recommendations provide no benefit to the consumer. But for other values, ads and recommendations will be accurate. In particular, the incentive for the seller to provide accurate ads and recommendations will be inversely related to the difference between the cost of producing the good and its average market evaluation.
Similar content being viewed by others
Notes
In this context, data mining (Fayyad et al. 1996) typically refers to the analysis of vast amounts of diverse types of data, searching for interesting patterns and correlations. Collaborative filtering (Resnick et al. 1994) refers to “filtering” information through the collaboration of multiple entities—for instance, predicting a consumer’s preferences using the collected preferences of many other consumers. Data mining and collaborative filtering can be used in recommender systems, which try to predict an agent’s rating of an item, in order to provide useful item recommendations. See also Kohavi and Provost (2001).
We focus on a single product recommendation per period. This approach is justified not just by the fact that certain recommender systems work that way (for instance, Amazon.com may send its customers emails recommending a specific new product), but also because, even when the merchant is offering multiple products, each recommendation can be analyzed in isolation from the others. For instance, the consumer may choose to try one among the products being offered; her reaction to that single product will influence her future interactions with that merchant, as detailed in the model, even if the merchant had recommended other products as well.
As usual, one can resolve the indifference in favor of the seller, since the seller could charge a price slightly smaller than \(v\) and create a strict decision to purchase.
We ignore the equality case, as it has probability zero of occurring.
We need to restrict the set of allowable strategies. Otherwise strategies such as “buy only if the price is greater than \(Eu/2\) and less than or equal to \(Eu\)” would be possible.
References
Allen, F. (1984). Reputation and product quality. RAND Journal of Economics, 15(3), 311–327.
Ansari, A., Essegaier, S., & Kohli, R. (2000). Internet recommendation systems. Journal of Marketing Research, 37(3), 363–376.
Avery, C., Resnick, P., & Zeckhauser, R. (1999). The market for evaluations. American Economic Review, 89(3), 564–584.
Beales, H. (2010). The value of behavioral targeting. Network Advertising Initiative. http://www.networkadvertising.org/pdfs/Beales_NAI_Study.pdf.
Benabou, R., & Laroque, G. (1992). Using privileged information to manipulate markets: Insiders, gurus, and credibility. Quarterly Journal of Economics, 107(3), 921–958.
Bennett, J., & Lanning, S. (2007). The Netflix prize. In Proceedings of KDD Cup and Workshop.
Blume, A. (1998). Contract renegotiation with time-varying valuations. Journal of Economics and Management Strategy, 7(3), 397–433.
Che, Y.-K. (1996). Customer return policies for experience goods. Journal of Industrial Economics, 44(1), 17–24.
Crawford, V., & Sobel, J. (1982). Strategic information transmission. Econometrica, 50(6), 1431–1452.
Dellarocas, C. (2003). The impact of online opinion forums on competition and marketing strategies. Mimeo, MIT Sloan School of Management.
Ely, J., & Valimaki, J. (2003). Bad reputation. Quarterly Journal of Economics, 118(3), 785–814.
Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). From data mining to knowledge discovery in databases. AI Magazine, 17(3), 37.
Goldfarb, A., & Tucker, C. (2011). Privacy regulation and online advertising. Management Science, 57(1), 57–71.
Kohavi, R., & Provost, F. (2001). Applications of data mining to electronic commerce. In Data Mining and Knowledge Discovery (pp. 1–7). Boston, MA: Kluwer.
Krishna, A., & Zhang, Z. J. (1999). Short or long-duration coupons: The effect of the expiration date on the profitability of coupon promotions. Management Science, 45(8), 1041–1056.
Krishna, V., & Morgan, J. (2001). A model of expertise. Quarterly Journal of Economics, 116(2), 747–775.
Lewis, T. R., & Sappington, D. E. M. (1994). Supplying information to facilitate price discrimination. International Economic Review, 35(2), 309–327.
Liebeskind, J., & Rumelt, R. P. (1989). Markets for experience goods with performance uncertainty. RAND Journal of Economics, 20(4), 601–621.
Maskin, E., & Tirole, J. (1990). The principal-agent relationship with an informed principal, I: The case of private values. Econometrica, 58(2), 379–409.
Maskin, E., & Tirole, J. (1992). The principal-agent relationship with an informed principal, II: Common values. Econometrica, 60(1), 1–42.
McCulloch, R. E., Rossi, P. E., & Allenby, G. M. (1996). The value of purchase history data in target marketing. Marketing Science, 15(4), 321–340.
McFadden, D. L., & Train, K. E. (1996). Consumers’ evaluation of new products: Learning from self and from others. Journal of Political Economy, 104(4), 683–703.
Miller, N., Resnick, P., & Zeckhauser, R. (2005). Eliciting informative feedback: The peer-prediction method. Management Science, 51, 1359–1373.
Moorthy, S., & Srinivasan, K. (1995). Signaling quality with a money-back guarantee: The role of transactions costs. Marketing Science, 14(4), 442–466.
Morgan, J., & Stocken, P. (2003). An analysis of stock recommendations. RAND Journal of Economics, 34(1), 183–203.
Morris, S. (2001). An instrumental theory of political correctness. Journal of Political Economy, 109(2), 231–265.
Ottaviani, M., & Sørensen, P. (2006a). Reputational cheap talk. Rand Journal of Economics, 37(1), 155–175.
Ottaviani, M., & Sørensen, P. (2006b). The strategy of professional forecasting. Journal of Financial Economics, 81(2), 441–466.
Prendergast, C. (1993). A theory of “yes men”. American Economic Review, 83(4), 757–770.
Resnick, P., & Sami, R. (2007). The influence limiter: Provably manipulation-resistant recommender systems. In Proceedings of the 2007 ACM Conference on Recommender Systems (pp. 25–32). ACM.
Resnick, P., & Varian, H. (1997). Recommender systems. Communications of the ACM, 40(3), 56–58.
Resnick, P., Iacovou, N., Suchak, M., Bergstrom, P., & Riedl, J. (1994). Grouplens: An open architecture for collaborative filtering of netnews. In Proceedings of the 1994 ACM Conference on Computer supported cooperative work (pp. 175–186). ACM.
Rossi, P. E., & Allenby, G. M. (1998). Marketing models of consumer heterogeneity. Journal of Econometrics, 89(1–2), 57–78.
Schafer, J. B., Konstan, J., & Riedl, J. (1999). Recommender systems in e-commerce. In Proceedings of the 1st ACM Conference on Electronic commerce (pp. 158–166). ACM.
Schlee, E. E. (2001). Buyer experimentation and introductory pricing. Journal of Economic Behavior and Organization, 44(3), 347–362.
Shaffer, G., & Zhang, Z. J. (1995). Competitive coupon targeting. Marketing Science, 14(4), 395–416.
Shapiro, C. (1983). Premiums for high quality products as returns to reputation. Quarterly Journal of Economics, 98(4), 659–679.
Sobel, J. (1985). A theory of credibility. The Review of Economic Studies, 52(4), 557–573.
Victor, P., Cornelis, C., De Cock, M., & Pinheiro da Silva, P. (2009). Gradual trust and distrust in recommender systems. Fuzzy Sets and Systems, 160(10), 1367–1382.
Yan, J., Liu, N., Wang, G., Zhang, W., Jiang, Y., & Chen, Z. (2009). How much can behavioral targeting help online advertising?. In Proceedings of the 18th International Conference on World Wide Web (pp. 261–270). ACM.
Zhang, Z. J., Krishna, A., & Dhar, S. (2000). The optimal choice of promotional vehicles: Front-loaded or rear-loaded incentives? Management Science, 46(3), 348–362.
Acknowledgments
The author is grateful to Hal Varian for the inspiration for the model, and to Veronica Marotta as well as the editors and anonymous reviewers for extremely helpful comments. This research is partly supported by the Singapore National Research Foundation under its International Research Centre Singapore Funding Initiative and administered by the IDM Programme Office.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Proof of Proposition 1
Let \(\pi _s\) be the posterior belief of the buyer about its type upon receiving the signal. There are three cases to consider:
-
Case 1
\(\pi _s v > c\). The seller must set the price \(p_1 \le \pi _s v\) in order to induce purchase. If \(p_1 > c\), then the seller would profit from signaling to both types since
$$\begin{aligned} \pi _s (p_1 - c) < p_1 - c. \end{aligned}$$But then the signal is uninformative, so \(\pi _s = \pi \). Since \(\pi v < c\), this contradicts the premise of this case. If \(p_1 \le c\), it will be profitable to signal only type \(v\). The seller wants to choose the largest such price, so it picks \(p_1 = c\). This is consistent with any \(\pi _s\) such that \(\pi _s v > c\). The seller would rationally want to signal all type \(v\) consumers, since this would maximize second-period sales. Hence a rational buyer must believe \(\pi _s=1\).
-
Case 2
\(\pi _s v < c\). Note that this includes the case \( \pi _s = \pi \), in which the buyer regards the seller’s signal as uninformative. In order to induce purchase, the seller must set \(p_1 \le \pi _s v\). Since \(p_1 \le \pi _s v < c\),
$$\begin{aligned} \pi _s (p_1 -c ) > p_1 -c, \end{aligned}$$so it is not in the seller’s interest to sell to both types. In this case the signal is completely informative, so the buyer should revise its posterior probability to \(\pi _s=1\). Since \(v > c\), this contradicts the premise. Hence this case cannot arise.
-
Case 3
\(\pi _{s}v=c\). Consider the strategy where the seller sets a price \(p_{1}\le \pi _{s}v=c\) and signals both types. The signal would then be uninformative, so \(\pi _{s}=\pi \) which contradicts \( \pi _{s}v=c\) since \(\pi v<c\). On the other hand, if the seller only signals type \(v\), then the signal is informative so \(\pi _{s}=1\), which implies \(v=c\) , which is a contradiction. \(\square \)
Proof of Proposition 2
The profit maximizing strategy for the seller is to sell to both types at the highest price that they are willing to accept. This price is \(\pi v\) for both types. If the seller’s strategy was to offer the good at a higher price \(p_{1}>\pi v\) to all, then the expected utility of the buyer would be negative, and no purchase would result. If the seller’s strategy was to offer the good only to the \(v\) type at a price \( p_{1\_v}=v>\pi v\), then every buyer upon receiving the signal would accept, in which case the best response of the seller would be to offer that price to all types, thereby making the signal uninformative and the expected utility of the buyer negative once again. Hence, no price above \(\pi v\) is sustainable in equilibrium. On the other hand, no price \(p_{1}<\) \(\pi v\) and no combination of prices \(p_{1\_0},p_{1\_v}\) (with \(p_{1\_0}\ne p_{1\_v}\)) for the \(v\) and \(0\) types are profit maximizing for the seller. The linear combinations \(p_{1\_0}=\frac{\pi }{1-\pi }e,p_{1\_v}=v-e\) (where \(e=\left[ 0,v\right] \) and \(v-e\ne \frac{\pi }{1-\pi }e\)) give both types nonnegative expected utilities but guarantee a lower profit for the seller than selling at a common price \(p_{1}=\pi v\). In the knife-edge case where \(\pi v=c\), as before the seller makes zero profit so it is again an equilibrium to signal only type \(v\), and for type \(v\) to buy. \(\square \)
Proof of Proposition 3
We have already shown that the seller will never charge the same price to both types when \(\pi v<c\). The seller will neither sell to both types at different prices. The following linear combinations of prices, \(p_{1\_0}=\frac{\pi }{1-\pi }e,p_{1\_v}=v-e\) (where \(e=\left[ 0,v \right] \) and \(v-e\ne \frac{\pi }{1-\pi }e\), and the seller practically gives away the good to one type of customer), give the buyer a nonnegative expected utility, but guarantee the seller only a profit of \(2\pi v-\pi c-c\), which is less than selling only to the \(v\) type at \(p_{1}=c\). Only the strategy of selling to the \(v\) type at \(p_{1\_v}=c\) satisfies Bayes’s Law. Let \(\pi _{s}\) be the posterior belief of the buyer about its type upon receipt of the signal. When \(\pi _{s}v>c\), if the seller sets \(p_{1}>c\), then it could sell to both types, thereby making the signal uninformative and \( \pi _{s}=\pi \), which contradicts \(\pi v<c\). When \(\pi _{s}v>c\) and the seller sets \(p_{1}=c\) and sells to all, the same contradiction arises. Only if the seller sets \(p_{1}=c\) and sells to the \(v\) type and \(\pi _{s}v>c\), then the signal is informative and \(\pi _{s}=1\), which satisfies \(\pi _{s}v>c \). \(\square \)
Proof of Proposition 4
Imagine that \(p\) is set to \(v\). To understand whether the customers can trust that the seller will only recommend a good sold at this price to those who will like it, we must compare the profit from giving good recommendations, to the profit from recommending the good to all. In the first case the present discounted value of the profit, over an infinite horizon, is \(\pi \left( p-c\right) +\frac{\delta }{1-\delta } \pi \left( p-c\right) \). In the second case, if the customer’s trigger strategy is never to buy again from a merchant that recommended a bad product to him, and the supply of customers is finite, the seller’s profits is \(\left( p-c\right) +\frac{\theta }{ 1-\theta }\left( p-c\right) \), where \(\theta =\frac{\pi }{1+r}\) because at each period \(1-\pi \) customers will abandon the seller after discovering that they have been recommended a good that they do not like. Hence, if \(\pi \left( p-c\right) +\frac{\delta }{1-\delta }\pi \left( p-c\right) >\left( p-c\right) +\frac{\theta }{1-\theta }\left( p-c\right) \), the honest strategy for the seller is a best response to the customer choosing to purchase. We can rewrite the inequality as:
by setting \(\frac{\delta }{1-\delta }=\frac{1}{r}\) and \(\frac{\theta }{ 1-\theta }=\frac{\pi }{1+r-\pi }\) and \(p=v\). Given that it is assumed that \( v>c\), and \((1+r)>\pi \) (see denominator in Eq. 5), then this condition is satisfied simply as long as \({3r\pi +3\pi -2\pi ^{2}-1-r}>0\). If this is the case, then the honest strategy is sustainable regardless of whether \(\pi v\lessgtr c\) (but as long as \(v>c\)).\(\square \)
Rights and permissions
About this article
Cite this article
Acquisti, A. Inducing Customers to Try New Goods. Rev Ind Organ 44, 131–146 (2014). https://doi.org/10.1007/s11151-013-9406-8
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11151-013-9406-8