Welfare Properties of Recommender Systems: Theory and Results from a Randomized Experiment

57 Pages Posted: 21 Oct 2016

See all articles by Pedro Ferreira

Pedro Ferreira

Carnegie Mellon University - H. John Heinz III School of Public Policy and Management; Carnegie Mellon University - Department of Engineering and Public Policy

Xiaochen Zhang

Carnegie Mellon University - Department of Engineering and Public Policy

Rodrigo Belo

Nova School of Business and Economics, Universidade Nova de Lisboa

Miguel Godinho de Matos

Católica-Lisbon School of Business and Economics; Carnegie Mellon University

Date Written: October 20, 2016

Abstract

Recommender systems assign products to slots in ways that improve consumers’ experience when choosing what to buy. They usually lead to more sales, which increases both consumer surplus and profit. However, firms may also choose which recommender system to use to maximize profit. Furthermore, consumers tend to exhibit a lower price elasticity of demand towards products placed in salient slots. We show that a firm using this knowledge may increase its profit hurting both consumer surplus and total welfare. We use data from a large scale field experiment ran using the video-on-demand system of a large telecommunications provider to measure the price elasticity of demand for movies placed in salient and non-salient slots on the TV screen. During this experiment, the firm randomized the slots in which movies were recommended to consumers as well as their prices. This readily allows for identifying the effects of price and slot on demand and thus compute consumer surplus. We find empirical evidence that indeed consumers are less price elastic towards movies placed in salient slots. Using the outcomes of this experiment we simulate how consumer surplus and welfare change when the firm implements the recommender system that maximizes profit. We also show that, at least in our setting, this system still yields higher consumer surplus than some recommender systems often used in practice, such as lists of most sold, most rated and highest rated products. Our results question whether recommender systems embed mechanisms that extract excessive surplus from consumers, which may need to be better scrutinized.

Suggested Citation

Ferreira, Pedro and Zhang, Xiaochen and Belo, Rodrigo and Godinho de Matos, Miguel, Welfare Properties of Recommender Systems: Theory and Results from a Randomized Experiment (October 20, 2016). Available at SSRN: https://ssrn.com/abstract=2856794 or http://dx.doi.org/10.2139/ssrn.2856794

Pedro Ferreira (Contact Author)

Carnegie Mellon University - H. John Heinz III School of Public Policy and Management ( email )

Pittsburgh, PA 15213-3890
United States

Carnegie Mellon University - Department of Engineering and Public Policy ( email )

Baker Hall 129
5000 Forbes Ave
Pittsburgh, PA 15213
United States

Xiaochen Zhang

Carnegie Mellon University - Department of Engineering and Public Policy ( email )

Baker Hall 129
5000 Forbes Ave
pittsburgh, PA 15213
United States

Rodrigo Belo

Nova School of Business and Economics, Universidade Nova de Lisboa ( email )

Campus de Carcavelos
Carcavelos, 2775-405
Portugal

Miguel Godinho de Matos

Católica-Lisbon School of Business and Economics ( email )

Palma de Cima
Lisbon, Lisboa 1649-023
Portugal

Carnegie Mellon University ( email )

Pittsburgh, PA 15213
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
356
Abstract Views
2,148
Rank
155,328
PlumX Metrics