Thomas Chan

Vancouver, BC · thomas.yautung.chan@gmail.com

I am an econometrician engaging in both theoretical and applied work. My research spans two main areas: causal inference, including experiment designs and policy learning, and nonparametric estimation. In my job market paper, I examine how adaptive experiments can enhance the estimation of a variety of causal parameters, aiming to inform policy decisions that account for more complex objectives beyond average effects.

I expect to graduate in Spring 2026 and will be available for interviews in the 2025-2026 job market. You can find my CV here.

Job Market Paper

Adaptive Experiment Design for Estimating Causal Effects

[link]

This paper considers a sequential experiment setting in which units arrive over time and outcomes are rapidly observed. It proposes an adaptive procedure that updates treatment randomization based on accumulated data to efficiently estimate a target parameter from a general class. This class includes not only average and quantile treatment effects but also distributional effects, inequality measures, and other policy-relevant parameters. Treatment randomization in the procedure is shown to converge to an optimal scheme, yielding estimators that achieve minimum asymptotic variance or minimize a designer-specified loss function. As experiments increasingly inform real-world decisions with diverse objectives, this approach broadens the scope of adaptive experiment designs for practical policy-making. Theoretical guarantees of optimality are supported by empirical illustrations using data from the Oregon Health Experiment and simulation evidence.

Working Papers

Policy Learning with Compliance Guarantee

With Vadim Marmer and Kyungchul Song

We study optimal policy learning where a policy maker uses policy outcome data from a source population to design treatment assignments for a target population under budget constraint. Due to the budget constraint, the policy maker needs to consider both the treatment effects and individuals’ incentives for treatment participation to minimize wasted resources. The main challenge is that treatment participation incentives may differ between the two populations. We develop a maximin approach that maximizes the minimum expected treatment outcome across all possible incentive configurations. We find that this optimal policy learning problem transforms into one with stochastic dominance constraints, where optimal assignment prioritizes individuals most likely to comply with the treatment assignment.

Work In Progress

Asymmetric and Optimal Bandwidth Selection in Estimation for First Price Auctions

I analyze bandwidth selection in the estimator proposed by Guerre, Perrigne, and Vuong (2000). I extend the inference framework of Ma, Marmer, and Shneyerov (2019) to cases where the ratio of the first- to second-stage bandwidths converges to either zero or infinity. In such regimes, the asymptotic normality is governed by the stage with the slower bandwidth rate. Further analysis shows that minimizing the pointwise mean squared error requires the bandwidth ratio to converge to zero. This result is driven by a bias-variance tradeoff that arises across the two estimation stages under certain conditions

Contact