I conduct research in the areas of Industrial Organization, Inverse Reinforcement Learning, and Reinforcement Learning.
Inverse Reinforcement Learning with Conditional Choice Probabilities https://arxiv.org/abs/1709.07597
We make an important connection to existing results in econometrics to describe an alternative formulation of inverse reinforcement learning (IRL). In particular, we describe an algorithm using Conditional Choice Probabilities (CCP), which are maximum likelihood estimates of the policy estimated from expert demonstrations, to solve the IRL problem. Using the language of structural econometrics, we re-frame the optimal decision problem and introduce an alternative representation of value functions due to (Hotz and Miller 1993). In addition to presenting the theoretical connections that bridge the IRL literature between Economics and Robotics, the use of CCPs also has the practical benefit of reducing the computational cost of solving the IRL problem. Specifically, under the CCP representation, we show how one can avoid repeated calls to the dynamic programming subroutine typically used in IRL. We show via extensive experimentation on standard IRL benchmarks that CCP-IRL is able to outperform MaxEnt-IRL, with as much as a 5x speedup and without compromising on the quality of the recovered reward function.
Applied Micro Research
Published in Vol. 55, No. 4, November 2014, International Economic Review
This paper studies repeated entry and bidding decisions in construction procurement auctions. We find evidence in the data that suggest the presence of significant cost savings from entering contracts of the same type. We estimate a dynamic auction model to measure the gains to experience for bidders. We allow for endogenous entry, synergies in entry and unobserved auction heterogeneity. We find that a bidder can halve entry costs by focusing on specific contract types. An auctioneer can increase competition by awarding contracts of the same type in sequence. As a result, procurement costs foreach contract can be lowered by 7%, a saving of $110,000.
This paper develops an approach for identifying and estimating the distribution of valuations in ascending auctions where an indeterminate number of bidders have an unknown number of bidding opportunities. To finesse the complications for identification and estimation due to multiple equilibria, our empirical analysis is based on the fact that bidders play undominated strategies in every equilibrium. We apply the model to a monthly financial market in which local banks compete for deposit securities. This market features frequent jump bidding and winning bids well above the highest losing bid, suggesting standard empirical approaches for ascending auctions may not be suitable. We find that frictions are costly both for revenue and allocative efficiency.
In this paper I empirically investigate prediction markets for binary options (Arrow-Debreu securities) for political events.
2019-present Amazon Economist in Display Advertising Monetization
2017-2019 Amazon Economist in Digital, Advertising and Corporate Development team
2010-2017 Assistant Professor of Economics, Carnegie Mellon University, Tepper School of Business