Working Paper
Dynamic Selection in Algorithmic Decision-making
This version: September 2023
JEL: C13, C26, C44
Abstract
This paper identifies and addresses dynamic selection problems in online learning algorithms with endogenous data. In a contextual multi-armed bandit model, a novel bias (self-fulfilling bias) arises because the endogeneity of the data influences the choices of decisions, affecting the distribution of future data to be collected and analyzed. We propose an instrumental-variable-based algorithm to correct for the bias. It obtains true parameter values and attains low (logarithmic-like) regret levels. We also prove a central limit theorem for statistical inference. To establish the theoretical properties, we develop a general technique that untangles the interdependence between data and actions.
Acknowledgements
We are grateful for helpful conversations with Chunrong Ai, Xi Chen, Pingyang Gao, Bob Gibbons, John Klopfer, Danielle Li, Whitney Newey, Michael Song, Wing Suen, Chang Sun, Brian Viard, and seminar participants at Causal Data Science Meeting 2021, University of Hong Kong, UC Riverside, and CUHK Workshop: “Advances in Econometrics, Machine Learning, and Big Data.” All remaining errors are ours.
Related Research
Suggested Citation
Li, J., Luo, Y., & Zhang, X. (2023). Dynamic selection in algorithmic decision-making (CAMO Working Paper No. 2023-01). HKU Centre for AI, Management and Organization. https://camo.hku.hk/2023-01-dynamic-selection-algorithmic/
BibTeX
@techreport{li_luo_zhang_2023,
author = {Li, Jin and Luo, Ye and Zhang, Xiaowei},
title = {Dynamic Selection in Algorithmic Decision-making},
institution = {HKU Centre for AI, Management and Organization},
type = {{CAMO} Working Paper},
number = {2023-01},
year = {2023},
month = sep,
url = {https://camo.hku.hk/2023-01-dynamic-selection-algorithmic/}
}