Finite-Sample Properties of Model Specification Tests for Multivariate Dynamic Regression Models

Akihiko Noda; Koichiro Moriya

arxiv: 2601.21272 · v2 · submitted 2026-01-29 · 💰 econ.EM · q-fin.PR· q-fin.ST

Finite-Sample Properties of Model Specification Tests for Multivariate Dynamic Regression Models

Koichiro Moriya , Akihiko Noda This is my paper

Pith reviewed 2026-05-16 10:21 UTC · model grok-4.3

classification 💰 econ.EM q-fin.PRq-fin.ST

keywords model specification testmultivariate dynamic regressionDurbin estimatorbootstrap Wald testexogeneity conditionsfinite-sample propertiesFama-French models

0 comments

The pith

A generalized Durbin estimator for multivariate dynamic regressions remains consistent under the weakest exogeneity condition and supports reliable bootstrap Wald tests.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a model specification test for multiple-equation systems that allows cross-equation error correlations and dynamic dependence between regressors and errors. Conventional tests assume stronger exogeneity that often fails in such settings and renders them invalid. The authors introduce a generalized Durbin estimator that accommodates intercepts along with these dependences and prove it stays consistent under the weakest exogeneity condition. They derive the asymptotic distribution to build Wald tests and show via Monte Carlo experiments that the bootstrap version substantially improves finite-sample size control. An application to Fama-French multifactor models finds the new test leaves the null unrejected in cases where FGLS-based tests reject it.

Core claim

We propose a generalized Durbin estimator for multiple-equation systems with an intercept, cross-equation error correlations, and dynamic regressor-error dependence. This estimator remains consistent under the weakest exogeneity condition. We derive its asymptotic distribution and construct Wald tests for model specification. Monte Carlo experiments confirm that the bootstrap-based Wald test substantially improves finite-sample size control. An application of the bootstrap-based Wald test to the Fama-French multifactor models leaves the null hypothesis unrejected in cases where competing FGLS-based tests reject it.

What carries the argument

The generalized Durbin estimator for multiple-equation systems, which adjusts for dynamic regressor-error dependence and cross-equation error correlations to achieve consistency under minimal exogeneity.

Load-bearing premise

The bootstrap procedure for the Wald test is valid under the dynamic regressor-error dependence and cross-equation correlations assumed in the Monte Carlo design and the Fama-French application.

What would settle it

A Monte Carlo experiment with dynamic dependence stronger than in the paper's design where the bootstrap Wald test fails to maintain correct size at the nominal level would falsify the finite-sample improvement claim.

read the original abstract

We propose a new model specification test for multiple-equation systems with cross-equation error and dynamic regressor--error dependences. Conventional tests often rely on exogeneity conditions strong enough to ensure consistency of the OLS estimator. These exogeneity conditions are violated when regressors and errors are dynamically dependent, rendering conventional model specification tests invalid. To address these limitations, we clarify the relationship among alternative exogeneity conditions, characterize the consistency of competing multiple-equation estimators, and propose a generalized Durbin estimator for multiple-equation systems with an intercept, cross-equation error and regressor--error dependences. We show that our estimator remains consistent under the weakest exogeneity condition. We then derive its asymptotic distribution and construct Wald tests. Our Monte Carlo experiments confirm that the bootstrap-based Wald test substantially improves finite-sample size control. An application of the bootstrap-based Wald test to the Fama--French multifactor models leaves the null hypothesis unrejected in cases where competing FGLS-based tests reject it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a generalized Durbin estimator for multivariate dynamic systems that stays consistent under weak exogeneity and pairs it with a bootstrap Wald test that improves size in simulations, but the bootstrap lacks a clear consistency proof.

read the letter

The core contribution is extending the Durbin estimator to multiple equations while allowing cross-equation error correlation and dynamic regressor-error feedback. They show consistency under the weakest exogeneity condition, derive the asymptotic distribution, and build Wald tests. The bootstrap version is the practical piece: Monte Carlo evidence indicates it controls finite-sample size better than standard FGLS-based alternatives, and the Fama-French application shows it not rejecting when competitors do. That addresses a real issue in applied work with multivariate time series or factor models where strong exogeneity fails. The Monte Carlo design and the application are straightforward and relevant. The main limitation is that bootstrap validity is asserted from the simulations rather than established through explicit arguments such as Edgeworth expansions or resampling theory under the dynamic dependence they allow. The abstract and available details do not spell out the full regularity conditions or derivation steps for the asymptotics, so readers will need to check those sections carefully. This is aimed at econometricians who routinely check multivariate dynamic regressions and want a test that does not require strong exogeneity. It is solid enough on the estimator and the empirical illustration to warrant referee time, though the bootstrap theory will likely need strengthening in revision.

Referee Report

2 major / 2 minor

Summary. The paper proposes a generalized Durbin estimator for multivariate dynamic regression models that is consistent under the weakest exogeneity condition permitting dynamic regressor-error dependence and cross-equation correlations. It derives the asymptotic distribution of this estimator to construct Wald tests for model specification, shows that conventional tests fail under weaker exogeneity, and demonstrates via Monte Carlo experiments that a bootstrap-based Wald test substantially improves finite-sample size control. An application to Fama-French multifactor models is included where the bootstrap test does not reject the null in cases where FGLS-based tests do.

Significance. If the bootstrap procedure is valid under the paper's maintained conditions, the work would offer a practical specification test for systems where standard exogeneity assumptions are violated, with Monte Carlo evidence and an empirical application providing concrete support for improved size control. The clarification of exogeneity conditions and consistency characterizations across estimators is a useful contribution to the literature on dynamic multivariate models.

major comments (2)

[Monte Carlo experiments and bootstrap construction] The central recommendation of the bootstrap-based Wald test rests on Monte Carlo evidence of improved size control under dynamic regressor-error dependence, but no section provides a proof of bootstrap consistency (e.g., via Edgeworth expansion or validity of the resampling scheme under the weakest exogeneity condition). This is load-bearing because the asymptotic distribution of the Wald statistic is derived, yet the bootstrap step that justifies finite-sample use lacks theoretical grounding beyond the specific MC design.
[Asymptotic distribution derivation] The derivation of the asymptotic distribution of the generalized Durbin estimator is stated, but the key steps, regularity conditions, and handling of cross-equation error correlations under dynamic dependence are not visible or detailed, making it difficult to verify the Wald statistic's limiting distribution.

minor comments (2)

[Estimator definition] Clarify notation for the generalized Durbin estimator in the presence of an intercept to avoid ambiguity with standard Durbin-Watson forms.
[Empirical application] The Fama-French application would benefit from reporting the exact Wald statistic values and p-values for both bootstrap and competing tests to allow direct comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our results. We address each major point below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

Referee: The central recommendation of the bootstrap-based Wald test rests on Monte Carlo evidence of improved size control under dynamic regressor-error dependence, but no section provides a proof of bootstrap consistency (e.g., via Edgeworth expansion or validity of the resampling scheme under the weakest exogeneity condition). This is load-bearing because the asymptotic distribution of the Wald statistic is derived, yet the bootstrap step that justifies finite-sample use lacks theoretical grounding beyond the specific MC design.

Authors: We acknowledge that the manuscript does not contain a formal proof of bootstrap consistency under the weakest exogeneity conditions. The Monte Carlo design covers the relevant cases of dynamic dependence and cross-equation correlation, showing clear improvements in size control. In the revision we will add a dedicated subsection that sketches the bootstrap consistency argument by appealing to the established asymptotic normality of the generalized Durbin estimator and citing related results on bootstrap validity for Wald statistics in dynamic multivariate models. This addition will provide the requested theoretical grounding without altering the core Monte Carlo evidence. revision: partial
Referee: The derivation of the asymptotic distribution of the generalized Durbin estimator is stated, but the key steps, regularity conditions, and handling of cross-equation error correlations under dynamic dependence are not visible or detailed, making it difficult to verify the Wald statistic's limiting distribution.

Authors: We agree that the main text would benefit from greater transparency on the derivation. The complete proof, including all regularity conditions and the treatment of cross-equation error correlations under dynamic dependence, appears in the appendix. We will revise the main text to include a concise step-by-step outline of the key arguments and explicitly list the regularity conditions. The appendix will be expanded with additional intermediate steps to facilitate verification of the limiting distribution of the Wald statistic. revision: yes

Circularity Check

0 steps flagged

No significant circularity; consistency and asymptotics derived from estimator construction under stated assumptions

full rationale

The paper defines the generalized Durbin estimator for multivariate systems allowing dynamic regressor-error dependence and cross-equation correlations, then shows its consistency under the weakest exogeneity condition via direct argument from the estimator's construction. Asymptotic distribution is subsequently derived to build the Wald statistic. Bootstrap performance is assessed through Monte Carlo experiments under the paper's maintained design, which constitutes external simulation validation rather than a fitted input renamed as prediction. No self-definitional loops appear (e.g., no quantity defined in terms of itself), no load-bearing self-citations reduce the central result to prior unverified work by the same authors, and no ansatz is smuggled via citation. The derivation chain remains self-contained against the explicit exogeneity and dependence assumptions; Monte Carlo confirmation does not create circularity because it tests finite-sample behavior outside the asymptotic derivation. This yields a normal non-finding of score 0.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard econometric regularity conditions for consistency and asymptotic normality of estimators under weak exogeneity, plus bootstrap validity in finite samples.

axioms (2)

domain assumption Standard regularity conditions for asymptotic normality of the generalized Durbin estimator under weak exogeneity.
Invoked to derive the asymptotic distribution and construct the Wald test.
domain assumption Bootstrap consistency for the Wald statistic under the maintained dynamic dependence structure.
Used to justify improved finite-sample size control in Monte Carlo experiments.

pith-pipeline@v0.9.0 · 5476 in / 1170 out tokens · 23107 ms · 2026-05-16T10:21:31.039235+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Proposition 2... BD ⊊ GEXOG ⊊ EBD... generalized Durbin estimator... consistent under EBD

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.