Parameters on the boundary in predictive regression

Edoardo Zanelli; Giuseppe Cavaliere; Iliyan Georgiev

arxiv: 2409.12611 · v1 · submitted 2024-09-19 · 💰 econ.EM · stat.ME

Parameters on the boundary in predictive regression

Giuseppe Cavaliere , Iliyan Georgiev , Edoardo Zanelli This is my paper

Pith reviewed 2026-05-23 20:32 UTC · model grok-4.3

classification 💰 econ.EM stat.ME

keywords bootstrap inferencepredictive regressionparameter on boundarywild bootstrapnon-stationarityGranger causalityconstrained estimation

0 comments

The pith

A data-dependent shift of the bootstrap parameter space removes boundary-induced randomness from predictive regression inference.

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

The paper studies bootstrap testing in predictive regressions where the parameter of interest may lie on the boundary of its space, as occurs with no predictability or sign-restricted predictability under a smooth inequality constraint. Standard fixed-regressor wild bootstrap produces limiting distributions that are random because they mix boundary effects with non-stationarity of the predictor. The authors modify the scheme by shifting the bootstrap parameter space with a data-dependent function that cancels the boundary contribution, proving that the adjusted bootstrap statistics converge to the correct null limit driven only by predictor non-stationarity. This matters for reliable inference on predictability or Granger causality when parameters sit at the edge. The method is first shown in a location model and then extended to the regression case.

Core claim

Constrained estimation in predictive regressions yields bootstrap statistics whose limit is random due to both boundary location of the true parameter and non-stationarity of the predictor; a data-dependent shift of the bootstrap parameter space eliminates the boundary portion of the limiting randomness, so that validity of bootstrap inference holds with non-stationarity as the sole remaining source.

What carries the argument

The data-dependent shift of the bootstrap parameter space, which adjusts the constrained estimation to cancel boundary-induced limiting randomness while retaining the effects of predictor non-stationarity.

If this is right

Bootstrap critical values remain valid even if the true parameter vector lies on the boundary.
The procedure covers both the case of no predictability and the case of sign-restricted predictability.
Validity is established under non-stationarity of the predicting variable.
The same shift technique applies to other inference problems that involve parameters on the boundary of a smooth constraint.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar shifts could be explored in bootstrap methods for other models with inequality restrictions common in economic applications.
The approach may extend to settings with additional forms of dependence beyond the fixed-regressor framework used here.
Testing the finite-sample behavior of the shift when the smoothness of the constraint is only approximate would be a natural next check.

Load-bearing premise

The inequality constraint on the parameter space is smooth and the data-dependent shift isolates non-stationarity as the only source of limiting randomness.

What would settle it

A Monte Carlo experiment or analytic check that compares the limit distribution of the shifted bootstrap statistic against the original statistic when the true parameter lies on the boundary and the predictor is integrated.

read the original abstract

We consider bootstrap inference in predictive (or Granger-causality) regressions when the parameter of interest may lie on the boundary of the parameter space, here defined by means of a smooth inequality constraint. For instance, this situation occurs when the definition of the parameter space allows for the cases of either no predictability or sign-restricted predictability. We show that in this context constrained estimation gives rise to bootstrap statistics whose limit distribution is, in general, random, and thus distinct from the limit null distribution of the original statistics of interest. This is due to both (i) the possible location of the true parameter vector on the boundary of the parameter space, and (ii) the possible non-stationarity of the posited predicting (resp. Granger-causing) variable. We discuss a modification of the standard fixed-regressor wild bootstrap scheme where the bootstrap parameter space is shifted by a data-dependent function in order to eliminate the portion of limiting bootstrap randomness attributable to the boundary, and prove validity of the associated bootstrap inference under non-stationarity of the predicting variable as the only remaining source of limiting bootstrap randomness. Our approach, which is initially presented in a simple location model, has bearing on inference in parameter-on-the-boundary situations beyond the predictive regression problem.

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 data-dependent shift to the bootstrap parameter space that removes boundary randomness in predictive regressions, leaving only non-stationarity as the limiting factor.

read the letter

The main contribution is a modification to the fixed-regressor wild bootstrap where the bootstrap parameter space gets shifted by a data-dependent function. This strips out the randomness coming from the boundary location of the parameter, so the bootstrap limit matches the null distribution under non-stationarity of the predictor alone. They lay it out first in a location model before extending to the regression case, and they claim proofs of validity for the adjusted procedure when the constraint is a smooth inequality. That targeted fix addresses a real issue: standard constrained bootstrap can produce a random limit when both boundary effects and non-stationarity are present. The construction looks like a genuine incremental step rather than a restatement of earlier work on boundary inference or wild bootstrap for non-stationary regressors. It is useful for handling cases like no predictability or sign-restricted predictability in Granger-causality tests. The soft spots are limited. Everything rests on the constraint being smooth; if it is not, the shift may not isolate non-stationarity cleanly. The abstract asserts the proofs but does not show the derivations, so any regularity conditions on the shift function or the non-stationarity process would need checking in the full text. The paper notes possible use beyond predictive regressions, but the results stay tied to this setup. This is for econometricians and time-series people who run bootstrap inference on predictive regressions with possible boundary parameters. A reader already working on constrained estimation or non-stationary predictors would get direct value from the specific adjustment. It deserves peer review because the problem is well-posed and the proposed solution is concrete enough to evaluate on its own terms.

Referee Report

2 major / 2 minor

Summary. The paper considers bootstrap inference for predictive (Granger-causality) regressions in which the parameter of interest may lie on the boundary of the parameter space, where the boundary is defined via a smooth inequality constraint (e.g., allowing for zero or sign-restricted predictability). It shows that constrained estimation produces bootstrap statistics whose limiting distribution is random and therefore does not match the null limit of the original statistic, owing to both boundary location and non-stationarity of the predictor. The authors propose a modification of the fixed-regressor wild bootstrap that shifts the bootstrap parameter space by a data-dependent function chosen to remove the boundary component of the limiting randomness, leaving only non-stationarity; they prove validity of the resulting procedure, first in a location model and then for the predictive regression.

Significance. If the validity arguments hold, the contribution supplies a concrete, implementable bootstrap method for a class of inference problems that arise routinely in empirical macroeconometrics and finance. The explicit separation of boundary-induced randomness from non-stationarity-induced randomness, together with the initial verification in the location model before extension, is a methodological strength. The work also supplies a template that may be useful for other boundary-constrained problems beyond predictive regression.

major comments (2)

[§3] §3 (location-model case): the proof that the data-dependent shift removes all boundary randomness while preserving the non-stationarity component relies on the smoothness of the inequality constraint and on the shift being o_p(1) in a suitable sense; the manuscript should state the precise regularity conditions on the shift function that guarantee this isolation, because violation would make the limiting bootstrap distribution again random.
[§4] §4 (extension to predictive regression): the argument that non-stationarity of the predictor is the sole remaining source of limiting randomness after the shift must be shown to survive the transition from the location model; in particular, the paper needs to verify that the fixed-regressor wild bootstrap weights do not re-introduce boundary effects once the parameter space is shifted.

minor comments (2)

The abstract and introduction would benefit from a short numerical illustration (even in the location model) showing the difference between the unshifted and shifted bootstrap distributions.
Notation for the data-dependent shift function should be introduced once and used consistently; currently the same symbol appears to be overloaded in the location-model and regression sections.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading, positive assessment, and constructive suggestions. We address the two major comments point by point below.

read point-by-point responses

Referee: §3 (location-model case): the proof that the data-dependent shift removes all boundary randomness while preserving the non-stationarity component relies on the smoothness of the inequality constraint and on the shift being o_p(1) in a suitable sense; the manuscript should state the precise regularity conditions on the shift function that guarantee this isolation, because violation would make the limiting bootstrap distribution again random.

Authors: We agree that an explicit listing of the regularity conditions on the shift function would strengthen the exposition. The current proof invokes C^1 smoothness of the constraint and o_p(1) convergence of the shift, but does not isolate these as a standalone assumption. In the revision we will add a remark immediately after the statement of the location-model result that enumerates the required conditions: (i) the inequality constraint g is continuously differentiable with ∇g(θ0) ≠ 0 on the boundary, and (ii) the data-dependent shift δ_n satisfies δ_n = o_p(1) uniformly in a shrinking neighborhood of the boundary point. This will make transparent why boundary randomness is eliminated while the non-stationarity term is preserved. revision: yes
Referee: §4 (extension to predictive regression): the argument that non-stationarity of the predictor is the sole remaining source of limiting randomness after the shift must be shown to survive the transition from the location model; in particular, the paper needs to verify that the fixed-regressor wild bootstrap weights do not re-introduce boundary effects once the parameter space is shifted.

Authors: The proof in Section 4 already establishes that the shifted bootstrap statistic inherits the same limiting behavior as the location-model case under the fixed-regressor wild bootstrap, because the weights are independent of the regressor and the shift is constructed from the original-sample estimator. Nevertheless, we acknowledge that an explicit verification step would be helpful. In the revision we will insert a short paragraph after the statement of the main theorem that confirms the weights remain orthogonal to the shifted boundary constraint: the mean-zero property of the wild weights and their independence from the data-generating process ensure that no additional boundary-induced term appears in the limiting distribution. This step directly extends the location-model argument without requiring new assumptions. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper derives a modified fixed-regressor wild bootstrap by applying a data-dependent shift to the bootstrap parameter space, isolating non-stationarity of the predictor as the sole source of limiting randomness, with validity proved under a smooth inequality constraint. This construction is first shown in a location model and extended to predictive regressions. No quoted equations or steps reduce any claimed prediction or validity result by construction to fitted parameters, self-citations, or ansatzes from the same authors; the proof is presented as independent once the shift is applied. The central claim does not rely on load-bearing self-citation or renaming of known results.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review based solely on the abstract; full list of regularity conditions, moment assumptions, and non-stationarity definitions not available.

axioms (1)

domain assumption The parameter space is defined by a smooth inequality constraint.
Explicitly stated in the abstract as the definition of the boundary.

pith-pipeline@v0.9.0 · 5745 in / 1230 out tokens · 33197 ms · 2026-05-23T20:32:56.511194+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.

We discuss a modification of the standard fixed-regressor wild bootstrap scheme where the bootstrap parameter space is shifted by a data-dependent function in order to eliminate the portion of limiting bootstrap randomness attributable to the boundary
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear

?

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

the parameter space is defined by means of a smooth inequality constraint

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.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Georgiev and E

Cavaliere, G., I. Georgiev and E. Zanelli (2024): Parameter on the boundary in predictive regression, Econometric Theory, forthcoming

work page 2024
[2]

Georgiev, A.M.R

Demetrescu, M., I. Georgiev, A.M.R. Taylor and P.M.M. Rodrigues (2023): Extensions to IVX methods of inference for return predictab ility, Journal of Econometrics 237 (Issue 2, Part C). 4 Table S1: Empirical rejection probabilities (ERPs) of bootstrap tes ts under the null. Nominal level: 0 .05 θ0 = (0, 0)′ θ0 = (− 0.75, 0.75)′ θ0 = ( − 1.50, 1.50)′ b1 b2...

work page 2023

[1] [1]

Georgiev and E

Cavaliere, G., I. Georgiev and E. Zanelli (2024): Parameter on the boundary in predictive regression, Econometric Theory, forthcoming

work page 2024

[2] [2]

Georgiev, A.M.R

Demetrescu, M., I. Georgiev, A.M.R. Taylor and P.M.M. Rodrigues (2023): Extensions to IVX methods of inference for return predictab ility, Journal of Econometrics 237 (Issue 2, Part C). 4 Table S1: Empirical rejection probabilities (ERPs) of bootstrap tes ts under the null. Nominal level: 0 .05 θ0 = (0, 0)′ θ0 = (− 0.75, 0.75)′ θ0 = ( − 1.50, 1.50)′ b1 b2...

work page 2023