top of page



Welfare Implications of Machine Learning Algorithms


Machine learning algorithms are increasingly used to make predictions in areas such as credit scoring and healthcare, where objectives beyond accuracy are relevant for their design and regulation. The following papers build conceptual frameworks to model (and in some cases, empirically quantify) different social and economic implications of large datasets and ML algorithms.

1. Algorithm Design: A Fairness-Accuracy Frontier (with Jay Lu, Xiaosheng Mu, and Kyohei Okumura), extended abstract at EC'22, latest draft: May, 2024 [+pdf] [+slides]

revision invited at Journal of Political Economy

2. Testing the Fairness-Accuracy Improvability of Algorithms (with Eric Auerbach, Max Tabord-Meehan, and Kyohei Okumura), latest draft: May, 2024 [+pdf]

3. The Value of Context: Human versus Black Box Evaluators (with Andrei Iakovlev), latest draft: Feb, 2024 [+pdf]

4. Algorithmic Fairness and Social Welfare, forthcoming at AEA Papers & Proceedings, 2024 (with Jay Lu) [+pdf]

5. Data and Incentives, Theoretical Economics, Vol. 19 (1), Pages 407-448, January 2024 (with Erik Madsen), extended abstract at EC'20 [+pdf] [+slides] [+talk]

Using Machine Learning to Inform Economic Modeling

Machine learning algorithms have demonstrated striking success in prediction, yet often fall short in interpretability and insight. Economic models, in contrast, provide abstractions and narratives that clarify the underlying forces behind behaviors. These papers ask whether black box methods can nevertheless be useful to an economic modeler whose goals extend beyond prediction, and if so, how.


Click here for overview slides that briefly summarize the papers listed as [1], [2], [3], and [5] below.

1. The Transfer Performance of Economic Models (with Isaiah Andrews, Drew Fudenberg, Lihua Lei, and Chaofeng Wu), latest draft: November, 2023 [+pdf] [+talk]

2. How Flexible is that Functional Form? Quantifying the Restrictiveness of Theories (with Drew Fudenberg and Wayne Gao), 2024, accepted at Review of Economics and Statistics [+pdf] [+slides]

"Exemplary AI and Computation Track Paper" at EC'21

3. Measuring the Completeness of Economic Models, Journal of Political Economy, Vol. 130 (4), Pages 956-990, April 2022 (with Drew Fudenberg, Jon Kleinberg and Sendhil Mullainathan), extended abstract at EC'17 [+pdf]

4. Machine Learning for Evaluating and Improving Theories, SIGEcom Exchanges, Vol. 18 (1), Pages 4-11, December 2020 (with Drew Fudenberg) [+pdf]

5. Predicting and Understanding Initial Play, American Economic Review, Vol. 109 (12), Pages 4112-4141, December 2019  (with Drew Fudenberg) [+pdf] [+slides] [+talk]

"Highlights Beyond EC" at EC'19

6. Inference of Preference Heterogeneity from Choice Data, Journal of Economic Theory, Vol. 179, Pages 275-311, January 2019 [+pdf]

Dynamic Information Acquisition of Correlated Information

A Bayesian agent allocates limited resources across different sources of information to inform an action taken at an endogenously chosen time. Classic solutions for optimal information acquisition exist when these sources of information are independent, but progress for correlated information has proven challenging. These papers show that a particular framework of dynamic information acquisition from correlated Gaussian sources is in fact very tractable, and explore the economic consequences of informational complementarities.

1. Dynamically Aggregating Diverse Information, Econometrica, Vol. 90 (1), Pages 47-80, January 2022 (with Xiaosheng Mu and Vasilis Syrgkanis), extended abstract at EC'21 [+pdf] [+online appendix] [+slides] [+talk]

2. Complementary Information and Learning Traps, Quarterly Journal of Economics, Vol. 135 (1), Pages 389-448, February 2020 (with Xiaosheng Mu), extended abstract at EC'18 [+pdf] [+online appendix] [+slides] [+talk]

3. Optimal and Myopic Information Acquisition, Proceedings of the 2018 ACM Conference on Economics and Computation, 2018 (with Xiaosheng Mu and Vasilis Syrgkanis) [+pdf] [+slides] [+talk]

bottom of page