EconBase
← All papers

Distributionally Robust Policy Learning with Wasserstein Distance

Daido Kido

arXiv 10 May 2022 · Econometrics

arXiv:2205.04637 · PDF · DOI · OpenAlex · Extracted main text

Abstract

The effects of treatments are often heterogeneous, depending on the observable characteristics, and it is necessary to exploit such heterogeneity to devise individualized treatment rules (ITRs). Existing estimation methods of such ITRs assume that the available experimental or observational data are derived from the target population in which the estimated policy is implemented. However, this assumption often fails in practice because of limited useful data. In this case, policymakers must rely on the data generated in the source population, which differs from the target population. Unfortunately, existing estimation methods do not necessarily work as expected in the new setting, and strategies that can achieve a reasonable goal in such a situation are required. This study examines the application of distributionally robust optimization (DRO), which formalizes an ambiguity about the target population and adapts to the worst-case scenario in the set. It is shown that DRO with Wasserstein distance-based characterization of ambiguity provides simple intuitions and a simple estimation method. I then develop an estimator for the distributionally robust ITR and evaluate its theoretical performance. An empirical application shows that the proposed approach outperforms the naive approach in the target population.

Citation extraction

47
references
86
in-text mentions
47
distinct cited
1
self-citations
12,416
main-text words

appendix boundary found by none_found · 100% of the source is main text. Read the extracted text to check this.

Most heavily cited references

The works this paper leans on most, across its whole bibliography — not restricted to papers in our corpus. Ranked by composite intensity, which combines how often a work is mentioned, how many sections mention it, and how much of that falls in the main text rather than the appendix.

ReferenceIntensityMentionsSectionsMain text
1Kitagawa, T. and Tetenov, A (2018) Who Should Be Treated? Empirical Welfare Maximization Methods for Treatment Choice1.00063100%
2Hotz, V. J., Imbens, G. W., and Mortimer, J. H (2005) Predicting the efficacy of future training programs using past experiences at other locations0.92843100%
3Zhou, Z., Athey, S., and Wager, S (2022) Offline Multi-Action Policy Learning: Generalization and Optimization0.92843100%
4Mo, W., Qi, Z., and Liu, Y (2021) Learning Optimal Distributionally Robust Individualized Treatment Rules0.87482100%
5Zhao, Y., Zeng, D., Tangen, C. M., and Leblanc, M. L (2019) Robustifying trial-derived optimal treatment rules for a target population0.87462100%
6Si, N., Zhang, F., Zhou, Z., and Blanchet, J (2021) Distributional Robust Batch Contextual Bandits0.81142100%
7Uehara, M., Kato, M., and Yasui, S (2020) Off-Policy Evaluation and Learning for External Validity under a Covariate Shift0.73732100%
8Adjaho, C. and Christensen, T (2022) Externally Valid Treatment Choice0.64441100%
9Blanchet, J. and Murthy, K (2019) Quantifying Distributional Model Risk via Optimal Transport0.58531100%
10Si, N., Zhang, F., Zhou, Z., and Blanchet, J (2020) Distributionally Robust Policy Evaluation and Learning in Offline Contextual Bandits0.58531100%

Showing the top 10 of 47 scored citations.

Cited by, within the corpus

arXiv econ.EM papers that cite this one, ranked by how heavily they lean on it.

Citing paperIntensityMentionsSections
1Externally Valid Policy Choice1.00074
2Treatment Choice, Mean Square Regret and Partial Identification0.73732
3Quantifying Distributional Model Risk in Marginal Problems via Optimal Transport0.64441
4Stochastic treatment choice with empirical welfare updating0.64422
5Optimal treatment assignment rules under capacity constraints0.64422
6Decision Theory for Treatment Choice Problems with Partial Identification0.51121
7Robust Bayes Treatment Choice with Partial Identification0.51121
8Orthogonal Policy Learning Under Ambiguity0.40511
9Locally Asymptotically Minimax Statistical Treatment Rules Under Partial Identification0.40511
10Who With Whom? Learning Optimal Matching Policies0.40511