Counterexamples and Sufficient Conditions: Comments on "Optimally-Transported Generalized Method of Moments"

summary

The gist

The paper comments on the optimally-transported generalized method of moments (OTGMM) estimator proposed by Schennach & Starck (2026a) and provides counterexamples to several theorems under their…

In short

This paper critiques Schennach & Starck's OTGMM estimator by providing counterexamples showing that existing theorems lack sufficient assumptions for consistency and asymptotic normality. It demonstrates that in many cases, the estimators select incorrect values or violate moment restrictions. The authors propose stronger conditions to restore validity for consistency and asymptotic properties.

Key concepts

OTGMM Estimator
This is a method used to estimate parameters by optimally transporting observed data moments to a theoretical distribution. It involves finding the best way (the transport metric) to map the observed sample moments onto a target distribution, aiming for an efficient estimation of the underlying model.
Consistency and Asymptotic Normality
These are crucial properties for statistical estimators. Consistency means that as more data is collected, the estimator converges to the true parameter value. Asymptotic normality describes how the estimator's errors behave in terms of probability distributions when large amounts of data are available.
Modified Moments
In some analyses, the original moments used in estimation are adjusted or 'modified'. The paper shows that under certain conditions, using these modified moments leads to different results for estimators like OTGMM and standard GMM, indicating a potential issue with the equivalence theorems.
Transport Metric
The transport metric defines how observed data points are moved or mapped onto a target distribution. The choice of this metric is critical; the paper notes that under misspecification, the resulting population value depends heavily on this chosen metric and which variables are allowed to be adjusted.

Terminology used across episodes

This episode discusses

The paper

Counterexamples and Sufficient Conditions: Comments on "Optimally-Transported Generalized Method of Moments" · Read on arXiv

University of Tokyo

We comment on the optimally-transported generalized method of moments (OTGMM) estimator proposed by Schennach & Starck (2026a) and give counterexamples to Theorems 2-6 under their stated assumptions. First, the assumptions used in the small-error analysis are insufficient for consistency in Theorem 2 and asymptotic normality in Theorem 3. Next, we consider the large-error analysis, in which Theorem 4 states that the OTGMM estimator is equivalent to a GMM estimator with modified moments. We show that in a scalar model, Theorem 4 selects a value that differs from the unique OTGMM minimizer and violates the OTGMM sample moment restriction. In an overidentified model satisfying the assumptions used in Theorems 5 and 6, the first component of the Lagrange multiplier has different probability limits under the OTGMM estimator and the GMM estimator with modified moments. Under misspecification, the population value selected by OTGMM depends on the transport metric and on which variables may be adjusted. We give a sufficient condition under which solutions of the modified moment equations also solve the original constrained problem at a given parameter value, and separate conditions for consistency and asymptotic normality of the OTGMM estimator. We also show that Assumption 16 does not imply the matrix bound used in the supplemental proofs and replace Assumption 16 with a matrix condition that yields the bound.

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: "Counterexamples and Sufficient Conditions".

Warren: Here's the gist of the paper. The paper comments on the optimally-transported generalized method of moments (OTGMM) estimator proposed by Schennach & Starck (2026a) and provides counterexamples to several theorems under their stated assumptions, showing that the assumptions are insufficient for consistency and asymptotic normality in many cases.

Ruth: Critique of Theoretical Guarantees The study demonstrates that the assumptions used in Theorems 2 and 3 are insufficient for consistency in Theorem 2 and asymptotic normality in Theorem 3.

Warren: A counterexample shows that changing two observed values so that the norm of each change converges to zero makes the sample moments hold at a parameter separated from the truth and at lower cost than at the true parameter.

Ruth: Furthermore, in an overidentified model satisfying Assumptions 5 and 6, the first component of the Lagrange multiplier has different probability limits under the OTGMM estimator and the GMM estimator with modified moments.

Paper summary: Ruth: So, to wrap up on this paper, "Counterexamples and Sufficient Conditions: Comments on Optimally-Transported Generalized Method of Moments," it’s really clear that the original assumptions for the OTGMM estimator don't guarantee consistency or asymptotic normality in many important scenarios.

Warren: The main point is that you can construct examples where changing some observed data points so their changes shrink to zero still allows the sample moments to hold at a parameter value far from the true one, and even cheaper than at the true one.

Iris: And in more complex models, they showed that comparing OTGMM with GMM estimators with modified moments leads to different probability limits for key components of the Lagrange multiplier.

Ruth: The implication for us is that we need much stronger conditions—like uniform bounds on how much moments can change when you adjust data, and a specific quadratic approximation for the transport cost—if we want to guarantee consistency and asymptotic normality.

Warren: They also clarified that when things are misspecified, the result isn't just about the transport metric; it’s dependent on which variables you are allowed to adjust, which is a limitation you need to keep in mind.

Iris: Overall, this paper doesn't invalidate OTGMM entirely but shows exactly where its guarantees break down and tells us what those stronger conditions actually look like for the results to hold.

Conclusion: Ruth: So, this paper by Schennach and Starck is really looking at their optimally transported generalized method of moments estimator.

Warren: They're showing where that method actually breaks down when you test its claims about consistency and asymptotic normality.

Iris: It seems like they found that the assumptions they set up for things to work correctly are just not strong enough for many real-world situations.

Ruth: Exactly, it’s not a one-size-fits-all setup; there are specific conditions we need to meet for the results to hold up.

Warren: They’re providing counterexamples where the estimator picks a value that isn't even the best possible one, and it messes with the underlying math of how those moments should behave.

Iris: And they’re clarifying that when things get messy, like when you have misspecification in your model, the result you get depends heavily on which variables you choose to adjust.

Ruth: So what does this mean for us? It means we can't just trust the initial theorems blindly; we have to be much more careful about the setup.

Warren: It means we need those stronger conditions they suggest, like bounding how much the moments can shift when you change some observed data points.

Iris: And they also pointed out that a test for whether there is error after theorem six needs its own special derivation because the math gets singular under certain conditions.

Ruth: This paper sets a clear boundary on what we can expect from this type of estimation, and it highlights exactly where the theory needs tightening before we can rely on these results.

Warren: It’s a warning to be very precise about your assumptions when you're building something like this estimator for real data.

More episodes

← Home