This paper explores scalable approaches to sensitivity analysis for instrumental variable (IV) estimation under joint relaxations of monotonicity and independence assumptions. A Monte Carlo simulation framework is developed to evaluate the robustness of IV estimates across different market conditions, using realistic financial data-generating processes. The results show standard IV estimators exhibit significant bias under assumption violations, while the proposed sensitivity analysis approach provides valid coverage.
Key findings
Standard IV estimators show significant bias under assumption violations.
Proposed sensitivity analysis maintains valid inference despite wider bounds.
Breakdown frontier is convex, indicating tradeoffs between independence and monotonicity violations.
Stressed markets reduce robustness compared to calm markets.
Limitations & open questions
The study focuses on financial markets, limiting generalizability to other fields.
The simulation parameters may not cover all possible market conditions.