Institution profile

Boston University

Academic institutionnorthamerica · us
Official website
Research library702linked papers
Opportunities0open roles
Selected work

Representative Papers

Adaptive Neyman Allocation

Sep 15, 2023ACM Conference on Economics and Computation

This paper addresses the optimal sample allocation problem in multi-stage randomized controlled trials when group variances (treatment and control) are unknown. We propose Adaptive Neyman Allocation: a method that dynamically estimates inter-group variances from early-stage data and continuously optimizes subsequent-stage assignment ratios to minimize the variance of the treatment effect estimator. We introduce, for the first time, a competitive analysis framework into experimental design; theoretically, our algorithm achieves an Ω(√log M) competitive ratio over T total samples and M stages, approaching the information-theoretic lower bound. Our approach breaks from the conventional fixed 1:1 allocation paradigm by enabling variance-driven, stage-wise optimal allocation. Empirical evaluation on real-world A/B tests at a major social platform demonstrates its effectiveness: under finite-stage constraints, estimation accuracy improves by over 30% compared to standard designs.

12 citations2 influentialRead paper

Estimating Effects of Long-Term Treatments

Jul 07, 2023ACM Conference on Economics and Computation

Accurately estimating the causal effects of long-term product interventions—such as UI redesigns or recommendation algorithm updates—in digital platforms remains challenging, as conventional short-term A/B tests fail to capture delayed and evolving impacts. To address this, we propose the first causal inference framework specifically designed for estimating long-term treatment effects. Our approach disentangles time-varying confounding from lagged treatment effects by explicitly modeling treatment duration as a key covariate. It integrates structural time-series modeling, doubly robust estimation, and dynamic causal graphs to enable counterfactual effect estimation without requiring costly long-duration experiments. Evaluated on real-world platform data, our method reduces long-term effect estimation error by 42% and achieves high-fidelity predictions across core metrics—including user retention rate and click-through rate—thereby significantly improving both the reliability and efficiency of long-horizon strategy evaluation.

7 citationsRead paper

Dynamic heterogeneous distribution regression panel models, with an application to labor income processes

Feb 08, 2022Social Science Research Network

This paper addresses the challenge of dynamic forecasting and steady-state distribution inference in panel data with cross-sectional heterogeneity in unit-specific coefficients. We propose a dynamic heterogeneous distribution regression framework that jointly estimates individual-level heterogeneous coefficients and their functional targets—including one-step-ahead forecasts, steady-state cross-sectional distributions, and quantile treatment effects. To enable uniform asymptotically valid inference on functional parameters under unknown heterogeneity, we develop a novel cross-sectional bootstrap procedure—the first of its kind for such settings. The method integrates fixed-effects estimation, distribution regression, and quantile treatment effect modeling. Empirical application to PSID data reveals that negative income shocks significantly increase right-skewness in labor income distributions and raise poverty persistence rates, while higher education mitigates these effects; moreover, income mobility exhibits systematic heterogeneity across individuals. Simulation studies confirm the method’s robustness and reliability.

4 citations1 influentialRead paper

Behavioral Economics of AI: LLM Biases and Corrections

Jan 01, 2026Social Science Research Network

This study investigates whether large language models (LLMs) exhibit systematic behavioral biases in economic and financial decision-making and proposes strategies to mitigate them. Drawing on paradigms from cognitive psychology and experimental economics, the authors systematically apply classic human bias experiments to multiple versions and scales of LLMs, analyzing their behavior in preference- and belief-based tasks. The findings reveal that larger models more closely mirror human irrationality in preference tasks, yet demonstrate greater rationality in belief tasks. Importantly, the study shows that rationality-oriented prompt engineering can significantly reduce such biases. These results uncover a nuanced relationship between model scale and decision rationality and introduce an effective prompting-based correction method to enhance the reliability of LLMs in economic reasoning contexts.

2 citationsRead paper

Kernel Limit of Recurrent Neural Networks Trained on Ergodic Data Sequences

Aug 28, 2023arXiv.org

This work investigates the asymptotic behavior of recurrent neural networks (RNNs) as the number of hidden units, sequence length, and training steps jointly tend to infinity. Conventional mean-field analysis fails because the hidden-state update scales as $ mathcal{O}(1) $, violating standard assumptions. To overcome this, we propose a novel analytical framework integrating stochastic algebraic fixed-point equations with Poisson equation techniques, combined with Sobolev-space regularity analysis, Markovian sequence modeling, and neural tangent kernel (NTK) derivation. We rigorously prove that, under ergodic data assumptions, the RNN dynamics converge to a coupled system comprising an infinite-dimensional ordinary differential equation (ODE) and a stochastic algebraic equation (SAE) fixed point. Furthermore, we derive— for the first time—the limiting RNN neural tangent kernel (RNN-NTK), explicitly characterizing its structure for temporal modeling. This establishes a foundational theoretical framework for deep sequential models.

2 citationsRead paper
Recent publications

Latest Papers