Best Feasible Conditional Critical Values for a More Powerful Subvector Anderson-Rubin Test

summary

The gist

For subvector inference in the linear instrumental variables model under homoskedasticity but allowing for weak instruments, Guggenberger, Kleibergen…

In short

The paper proposes a new conditional critical value function for the subvector Anderson-Rubin (AR) test, conditioning on the second-smallest eigenvalue of a data-dependent matrix instead of the largest one used by Guggenberger et al. This new approach maintains correct size and achieves strictly higher power than the original GKM test when there is more than one parameter not under test.

Key concepts

Subvector Anderson-Rubin (AR) Test
This is a statistical test used in linear instrumental variables models to perform inference on a subset of parameters. It relies on analyzing the minimum eigenvalue of a data-dependent matrix to determine if the instruments are weak or strong.
GKM Critical Value Function
Proposed by Guggenberger, Kleibergen, and Mavroeidis (2019), this function uses data-dependent critical values based on the largest eigenvalue of the relevant matrix to set rejection thresholds for the AR test.
Second-Smallest Eigenvalue Conditioning
The authors investigate conditioning the critical value on the second-smallest eigenvalue. They argue this specific eigenvalue is a better indicator of weak identification in this context, leading to a more powerful test when multiple parameters are not under test.

Terminology used across episodes

This episode discusses

The paper

Best Feasible Conditional Critical Values for a More Powerful Subvector Anderson-Rubin Test · Read on arXiv

Jesse Hoekstra, Frank Windmeijer

Department of Statistics and Nuffield College, University of Oxford, UK

For subvector inference in the linear instrumental variables model under homoskedasticity but allowing for weak instruments, Guggenberger, Kleibergen and Mavroeidis (2019) (GKM) propose a conditional subvector Anderson and Rubin (1949) (AR) test that uses data-dependent critical values that adapt to the strength of the parameters not under test. This test has correct size and strictly higher power than the test that uses standard asymptotic chi-square critical values. The subvector AR test is the minimum eigenvalue of a data-dependent matrix. The GKM critical value function conditions on the largest eigenvalue of this matrix. Instead, this paper considers a data-dependent critical value function that conditions on the second-smallest eigenvalue, as this eigenvalue is the appropriate indicator of weak identification. We find that the data-dependent critical value function of GKM also applies to this conditioning and show that this test has correct size and power strictly higher than the GKM test when the number of parameters not under test is larger than one. This is the best feasible test, where we consider feasible to be within the class of conditioning on a single eigenvalue. Our proposed procedure further applies to the subvector AR test statistic that is robust to an approximate Kronecker product structure of conditional heteroskedasticity as proposed by Guggenberger, Kleibergen and Mavroeidis (2024), carrying over its power advantage to this setting as well.

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Best Feasible Conditional Critical Values for a More Powerful Subvector Anderson-Rubin Test".

Warren: Here's the gist of the paper. For subvector inference in the linear instrumental variables model under homoskedasticity but allowing for weak instruments, Guggenberger, Kleibergen, and Mavroeidis (2019) (GKM) propose a conditional subvector Anderson and Rubin (1949) (AR) test that uses data-dependent critical values that adapt to the strength of the parameters not under test.

Ruth: This test has correct size and strictly higher power than the test that uses standard asymptotic chi-square critical values.

Warren: Pruning redundant reasoning tokens The paper investigates a data-dependent critical value function for the subvector AR test that conditions on the second-smallest eigenvalue of a data-dependent matrix, showing it maintains correct size and strictly higher power than the GKM test when the number of parameters not under test is larger than one.

Ruth: The authors find that this approach applies to the subvector AR test statistic that is robust to an approximate Kronecker product structure of conditional heteroskedasticity, carrying over its power advantage to this setting as well.

Paper summary: Ruth: So to recap, the main thrust of this paper is about improving subvector inference in linear instrumental variables models when we allow for weak instruments. They start with the GKM proposal, which uses data-dependent critical values to adapt to how strong the parameters you aren't testing are.

Warren: That GKM test already showed that it has correct size and strictly higher power than standard chi-square tests, but they are looking at a refinement of that idea.

Iris: This paper investigates using a data-dependent critical value function that conditions on the second smallest eigenvalue of the data-dependent matrix instead of the largest one. They argue this second smallest eigenvalue is the more appropriate signal for weak identification problems.

Ruth: They find that this conditioning works with the GKM function, and crucially, it shows that this new test has correct size and power strictly higher than the GKM test when there is more than one parameter not under test.

Warren: The authors also show they can apply this to a specific statistic that handles conditional heteroskedasticity in an approximate Kronecker product structure, which keeps that power advantage even in those complicated error settings.

Iris: They define the feasible set as conditioning only on a single eigenvalue, and they prove this is enough for controlled size when p is greater than two.

Ruth: The key finding here is that by conditioning on the second smallest eigenvalue instead of the largest, you get a test with better power when there are multiple parameters involved in identifying your model.

Warren: It matters because if you're running a simulation or an experiment and you have several variables but only some are strongly identified, this method helps you detect that weakness more effectively than the original GKM approach.

Iris: The authors also confirm that when the p minus two largest eigenvalues decrease to less than infinity, the rejection probability decreases, which is how they maintain size control under all circumstances.

Conclusion: Ruth: So, looking at the title, "Best Feasible Conditional Critical Values for a More Powerful Subvector Anderson-Rubin Test," it really sums up the paper's achievement: they found a better way to set those critical values.

Warren: It boils down to this: when you're dealing with subvectors and weak instruments, focusing on the second smallest eigenvalue gives you a test that is both statistically correct in size and more powerful than previous methods.

Iris: The implication for us is that if we are building models where we don't know everything about all the parameters, this method tells us exactly which piece of information—the second smallest eigenvalue—is the most useful for identifying problems with weak instruments.

Ruth: It's not just about having a better number; it’s about having a procedure that is robust enough to handle the kind of complex error structures we often see in real-world data analysis.

Warren: So, this paper suggests that if you are using these types of tests, and you have more than one parameter involved, switching your focus to the second smallest eigenvalue is a practical way to boost your test's performance without sacrificing statistical rigor.

More episodes

← Home