IV-ICL: Bounding Causal Effects with Instrumental Variables via In-Context Learning

2026-05-12

AI Summary

This work addresses the challenge of partial identification of causal effects under unobserved confounding, where existing methods either rely on cumbersome manual derivations or suffer from the high computational cost and prior sensitivity of Bayesian inference. The paper proposes IV-ICL, the first approach to integrate amortized Bayesian inference with inclusive KL divergence optimization into instrumental variable causal inference. IV-ICL directly models the marginal posterior distribution of causal effects via context learning and constructs effect bounds by extracting its quantiles, thereby covering the entire identified set. Additionally, the authors introduce a novel evaluation benchmark based on randomized controlled trials that provides ground-truth causal guarantees for instrumental variables. Experiments demonstrate that IV-ICL substantially outperforms state-of-the-art semiparametric, Bayesian, and plug-in methods across multiple benchmarks, yielding more reliable intervals and achieving 20–500× faster inference.

Abstract

The instrumental-variables (IV) setting is standard for partial identification of causal effects when unobserved confounding makes point identification impossible. Existing approaches face methodological bottlenecks: closed-form bound estimands are required -- e.g., Balke-Pearl equations in binary IV -- and even when available, designing accurate estimators requires manual effort tailored to each estimand. While direct Bayesian inference of the causal effects, instead of the bounds, circumvents these challenges, it is often computationally intensive and suffers from high prior sensitivity or under-dispersed posteriors. As a remedy, we introduce IV-ICL, an amortized Bayesian in-context learning method that learns the marginal posterior distribution of the causal effects directly and derives bounds as its quantiles. Unlike standard variational inference that optimizes exclusive KL divergence, amortized Bayesian inference minimizes the expected inclusive KL, a mass-covering objective. We empirically observe that optimizing inclusive KL can recover the entire identified set across diverse data-generating processes, while exclusive-KL (e.g. with variational inference) of the same Bayesian formulation collapses onto a single mode and fails to cover the identified set. We evaluate IV-ICL on synthetic and semi-synthetic IV benchmarks and show it produces intervals that are more reliably valid and more informative compared to efficient semi-parametric, Bayesian, and plug-in baselines, at 20-500x lower inference time. Beyond methodology, we propose a procedure to convert randomized controlled trials into IV benchmarks with provably preserved ground-truth causal effects that enables a more realistic evaluation of partial-identification methods.

Problem

• instrumental variables
• partial identification
• causal effects
• unobserved confounding
• bounding

Innovation

• instrumental variables
• in-context learning
• amortized Bayesian inference
• causal bounds
• inclusive KL divergence