Papers
Short notes on selected papers — what the problem was, what the paper does, and what it establishes — written to be read before the paper rather than instead of it. The complete list of publications is on the Research page.
AI, LLMs, and Economic Decision Making
- AI Economist Agent: An Agentic Framework for Model-Grounded Economic Analysis with RAG, Knowledge Graphs, and Large Language Models
An agentic framework that lets a language model write economist-style reports while every quantitative claim is produced by an explicit model computation and linked back to a retrieved source.
Adaptive Experimental Design
- Active Adaptive Experimental Design for Treatment Effect Estimation with Covariate Choices
If the experimenter can choose who enters the experiment as well as what they receive, then jointly optimizing the covariate density and the propensity score reaches a strictly lower semiparametric efficiency bound than optimizing the propensity score alone. - Adaptive Experimental Design for Policy Learning
When the deliverable of an experiment is a policy rather than a single recommended arm, the natural criterion is worst-case expected simple regret; the proposed PLAS strategy is minimax rate-optimal for it. - Adaptive Generalized Neyman Allocation: Local Asymptotic Minimax Optimal Best Arm Identification
In the small-gap regime, the Generalized Neyman Allocation attains the worst-case lower bound on the probability of misidentification exactly, constant terms included.
Best-Arm Identification and Treatment Choice
- Adaptive Experimental Design for Policy Learning
When the deliverable of an experiment is a policy rather than a single recommended arm, the natural criterion is worst-case expected simple regret; the proposed PLAS strategy is minimax rate-optimal for it. - Adaptive Generalized Neyman Allocation: Local Asymptotic Minimax Optimal Best Arm Identification
In the small-gap regime, the Generalized Neyman Allocation attains the worst-case lower bound on the probability of misidentification exactly, constant terms included. - Minimax and Bayes Optimal Best-Arm Identification
A single fixed-budget strategy that is simultaneously asymptotically minimax optimal and Bayes optimal for the simple regret, with upper bounds matching the lower bounds including constants. - The Role of Contextual Information in Best Arm Identification
Covariates observed during the experiment reduce the number of samples needed to identify the arm with the best marginalized mean reward, and a context-aware Track-and-Stop attains the instance-specific lower bound.
Debiased Machine Learning and Riesz Regression
- A Unified Framework for Debiased Machine Learning: Riesz Representer Fitting under Bregman Divergence
Fit the Riesz representer by minimizing a Bregman divergence. Squared distance recovers Riesz regression, KL recovers tailored loss minimization, and suitable divergence/link pairs make covariate balancing and Neyman orthogonalization automatic. - Learning Causal Models from Conditional Moment Restrictions by Importance Weighting
Conditional moment restrictions are converted into unconditional ones by importance weighting with a conditional density ratio estimator, which makes flexible models — including neural networks — usable for nonparametric instrumental variable problems.
Causal Inference and Off-Policy Evaluation
- Active Adaptive Experimental Design for Treatment Effect Estimation with Covariate Choices
If the experimenter can choose who enters the experiment as well as what they receive, then jointly optimizing the covariate density and the propensity score reaches a strictly lower semiparametric efficiency bound than optimizing the propensity score alone. - Adaptive Experimental Design for Policy Learning
When the deliverable of an experiment is a policy rather than a single recommended arm, the natural criterion is worst-case expected simple regret; the proposed PLAS strategy is minimax rate-optimal for it. - PUATE: Semiparametric Efficient Average Treatment Effect Estimation from Treated (Positive) and Unlabeled Units
Efficiency bounds and efficient ATE estimators for the case where the data contain a treated group and an unlabeled group whose treatment status is unknown. - Learning Causal Models from Conditional Moment Restrictions by Importance Weighting
Conditional moment restrictions are converted into unconditional ones by importance weighting with a conditional density ratio estimator, which makes flexible models — including neural networks — usable for nonparametric instrumental variable problems. - Off-Policy Evaluation and Learning for External Validity under a Covariate Shift
Off-policy evaluation and learning when the population you will deploy on differs from the population that generated the logs: the paper derives the efficiency bound under covariate shift and gives doubly robust, efficient estimators. - Asymptotically Unbiased Synthetic Control Methods by Density Matching
Standard synthetic control weights are biased because donor outcomes are correlated with the error term; matching distributions rather than fitting observed outcomes yields an asymptotically unbiased estimator and a full counterfactual distribution.
Density Ratio Estimation, Weak Supervision, and Anomaly Detection
- A Unified Framework for Debiased Machine Learning: Riesz Representer Fitting under Bregman Divergence
Fit the Riesz representer by minimizing a Bregman divergence. Squared distance recovers Riesz regression, KL recovers tailored loss minimization, and suitable divergence/link pairs make covariate balancing and Neyman orthogonalization automatic. - PUATE: Semiparametric Efficient Average Treatment Effect Estimation from Treated (Positive) and Unlabeled Units
Efficiency bounds and efficient ATE estimators for the case where the data contain a treated group and an unlabeled group whose treatment status is unknown. - Learning from Positive and Unlabeled Data with a Selection Bias
PU learning normally assumes labeled positives are distributed like unlabeled positives. When labeling is selective that assumption fails; the classifier is then only partially identified, and what can be learned is a scoring function that preserves the order of the class posterior. - Non-negative Bregman Divergence Minimization for Deep Direct Density Ratio Estimation
Bregman-divergence density ratio estimation overfits badly with deep models because the empirical objective can be driven down without learning the ratio; a non-negative correction removes that failure mode, with a generalization error bound and gains in inlier-based outlier detection. - Learning Causal Models from Conditional Moment Restrictions by Importance Weighting
Conditional moment restrictions are converted into unconditional ones by importance weighting with a conditional density ratio estimator, which makes flexible models — including neural networks — usable for nonparametric instrumental variable problems. - Off-Policy Evaluation and Learning for External Validity under a Covariate Shift
Off-policy evaluation and learning when the population you will deploy on differs from the population that generated the logs: the paper derives the efficiency bound under covariate shift and gives doubly robust, efficient estimators.
Portfolio Optimization and Financial Engineering
- Conformal Predictive Portfolio Selection
Wrap any return predictor in conformal prediction intervals, then choose the portfolio from the intervals rather than from point forecasts — a model-agnostic way to let prediction uncertainty enter the allocation.