Testing the Exclusion Restriction in IV Models Using Non-Gaussianity: A LiNGAM-Based Approach

TL;DR

This paper introduces a novel method combining IV analysis with LiNGAM to test the exclusion restriction in instrumental variable models by leveraging non-Gaussian data properties, providing multiple tests and empirical validation.

Contribution

It proposes a new framework that exploits non-Gaussianity to test the exclusion restriction in IV models, which was previously untestable under standard assumptions.

Findings

Five tests effectively assess exclusion restriction violations.

Monte Carlo simulations show controlled Type I error rates.

Empirical application demonstrates practical utility.

Abstract

Instrumental variable (IV) methods rely critically on the exclusion restriction, which is untestable in exactly-identified models under standard assumptions. We propose a framework combining IV analysis with the LiNGAM method to test this restriction by exploiting non-Gaussianity in the data. Under non-Gaussian structural errors, the exclusion violation parameter is point-identified without additional instruments. Five complementary tests (bootstrap percentile, asymptotic normal, permutation, likelihood ratio, and independence-based) are introduced to assess the restriction under varying data conditions. Monte Carlo simulations and an empirical application to the Card (1995) dataset demonstrate controlled Type I error rates and reasonable power against economically relevant violations.