REVIEW 2 major objections 2 minor 24 references
Partial Wavelet Canonical Coherence for Nonstationary Signals with High Dimensional Confounders
T0 review · 2 major / 2 minor · reviewed 2026-06-25 · grok-4.3
Pith's one-line read A new wavelet-based partial canonical coherence isolates direct associations between nonstationary multivariate time series after adjustment for high-dimensional confounders.
desk verdict The paper introduces wavelet-based partial canonical coherence for nonstationary multivariate series with high-dim confounder adjustment, but the simulation claims rest on unshown details. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
Partial wavelet canonical coherence, computed from the local wavelet spectral matrices of the combined target and confounder series by removing the linear projection onto the confounders at each time and scale.
What would settle it
A simulation in which the true direct partial canonical coherence is known to be zero after exact linear adjustment for the confounders, yet the estimator still reports nonzero values at the relevant scales and times, would show that the adjustment step fails to isolate direct association.
Extended reading notes
Core claim
The central claim is that partial canonical correlation analysis can be formulated in the frequency domain for time series through the partial wavelet canonical coherence, which is obtained by applying the inverse of the local wavelet spectral matrix of the confounder series to remove its linear effects from the cross-spectrum between the two target series. Estimation uses local wavelet spectral matrices constructed under the multivariate locally stationary wavelet framework, with principal-component reduction applied to the confounder spectral matrix to ensure numerical stability when the number of confounders is large. The resulting quantity is time-varying and scale-specific, allowing dir
Load-bearing premise
The series must follow the multivariate locally stationary wavelet model so that the target partial canonical coherence is well-defined and the local spectral matrix estimators are consistent.
Editorial extensions
If this is right
- The estimator removes spurious marginal associations induced by confounding and recovers the true direct associations in simulations, including when the confounder dimension is high.
- The measure is time-varying and scale-specific, so direct association can be tracked separately at different resolutions even when the series are nonstationary.
- Principal-component reduction on the confounder spectral matrix makes the procedure stable for high-dimensional confounders without requiring the number of confounders to be smaller than the sample size.
- Application to U.S. exchange-traded funds data shows substantial time-varying and scale-dependent direct canonical association once external market effects are removed.
Reading between the lines
- The same local-spectral adjustment idea could be applied to other frequency-domain association measures such as partial coherence or partial mutual information if analogous wavelet representations are available.
- In settings where the wavelet model is only approximate, the method might still serve as a practical exploratory tool whose output can be cross-checked against time-domain partial correlation estimates.
- The separation of direct from indirect effects suggests the procedure could be used to build networks in which only direct links are retained, provided the confounders are observed.
- Testing the estimator on data generated from known vector autoregressive processes with time-varying coefficients would provide an independent check outside the wavelet framework.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops Partial Wavelet Canonical Coherence (PWCC), a frequency-domain formulation of partial canonical correlation analysis for multivariate nonstationary time series. It defines the target quantity under the multivariate locally stationary wavelet model, estimates it via local wavelet spectral matrices with PCA-based adjustment for high-dimensional confounders, and reports that simulations recover direct associations while an application to U.S. ETFs shows time-varying scale-dependent links after market adjustment. The abstract positions this as the first such method.
Significance. If the central construction and empirical claims hold, the work would supply a novel scale-specific, time-varying tool for direct association measurement in nonstationary multivariate series with confounders. The wavelet framework combined with PCA for stable high-dimensional adjustment is a clear technical contribution, and the method's applicability to finance data is demonstrated. However, the absence of quantitative simulation metrics limits assessment of practical performance.
major comments (2)
- [Simulations section] The abstract asserts that 'simulations show that the method removes spurious marginal association induced by confounding and accurately recovers direct association, including in higher-dimensional settings,' yet supplies no quantitative results, error bars, or description of the data-generating processes. This is load-bearing for the empirical validation claim and requires detailed reporting (with metrics and DGPs) in the simulation section.
- [Methodology / Model assumptions] The target partial canonical coherence and its estimator are formulated under the multivariate locally stationary wavelet framework; the paper should include explicit checks or sensitivity analyses for departures from this assumption, as it underpins both the quantity and the local spectral matrix estimation procedure.
minor comments (2)
- [Introduction] The novelty claim ('first method') would benefit from a more explicit literature comparison in the introduction, citing any prior frequency-domain or wavelet-based partial CCA approaches for time series.
- [Estimation procedure] Notation for the local spectral matrices and the PCA reduction step should be introduced with explicit definitions and dimensions to aid readability.
Simulated Author's Rebuttal
We thank the referee for the thoughtful review and constructive suggestions. We address each major comment below. Where the comments identify gaps in the current manuscript, we agree to revise accordingly and will provide the requested details and analyses in the next version.
read point-by-point responses
-
Referee: [Simulations section] The abstract asserts that 'simulations show that the method removes spurious marginal association induced by confounding and accurately recovers direct association, including in higher-dimensional settings,' yet supplies no quantitative results, error bars, or description of the data-generating processes. This is load-bearing for the empirical validation claim and requires detailed reporting (with metrics and DGPs) in the simulation section.
Authors: We agree that the simulation section requires more detailed quantitative support to substantiate the abstract claims. The current manuscript describes the simulation setup at a high level but does not include explicit data-generating processes, numerical metrics (e.g., bias, MSE, or recovery rates), or error bars. In the revised version we will expand the Simulations section to include: (i) full specification of the DGPs used (including the multivariate locally stationary wavelet processes and confounding structures), (ii) quantitative performance metrics with standard errors across Monte Carlo replications, and (iii) results for both low- and high-dimensional settings. This will directly address the empirical validation concern. revision: yes
-
Referee: [Methodology / Model assumptions] The target partial canonical coherence and its estimator are formulated under the multivariate locally stationary wavelet framework; the paper should include explicit checks or sensitivity analyses for departures from this assumption, as it underpins both the quantity and the local spectral matrix estimation procedure.
Authors: The referee correctly notes that the theoretical development and estimation rely on the multivariate locally stationary wavelet model. While this is a standard and well-justified framework for nonstationary time series, the manuscript does not currently contain sensitivity analyses for departures from local stationarity. We will add a dedicated subsection (or appendix) that reports simulation results under controlled violations of the assumption, such as the introduction of slow global trends or localized non-stationarities, and will discuss the robustness of the PWCC estimator in those cases. This addition will strengthen the methodological justification without altering the core contribution. revision: yes
Circularity Check
No significant circularity identified
full rationale
The paper formulates the partial canonical coherence quantity under the multivariate locally stationary wavelet framework, estimates it via local wavelet spectral matrices, and applies PCA for high-dimensional confounder adjustment. No equations or derivations are supplied in the abstract or described construction that reduce the target to fitted inputs, self-definitions, or self-citation chains by construction. The framework supplies the local spectral matrices whose eigenstructure yields the coherence after adjustment, with no internal reduction to the inputs. Simulations and the ETF analysis are presented as external validation steps. This is a standard methodological construction without detectable circular steps.
Assumptions & free parameters
free parameters (1)
- number of retained principal components
assumptions (1)
- domain assumption Observed series obey the multivariate locally stationary wavelet process model
Cite this review
Pith. "Pith review of Partial Wavelet Canonical Coherence for Nonstationary Signals with High Dimensional Confounders." pith.science (2026). https://pith.science/paper/5RWCRIQG
@misc{pith2026260624554,
author = {Pith},
title = {Pith review of: Partial Wavelet Canonical Coherence for Nonstationary Signals with High Dimensional Confounders},
year = {2026},
howpublished = {\url{https://pith.science/paper/5RWCRIQG}},
note = {Machine review of arXiv:2606.24554}
}
read the original abstract
We develop Partial Wavelet Canonical Coherence for measuring the direct canonical association between two multivariate nonstationary time series after adjustment for possibly high-dimensional confounders. To the best of our knowledge, this is the first method that establishes a frequency-domain formulation of the partial canonical correlation analysis for time series. Through a wavelet approach, the proposed method yields a scale-specific, time-varying measure of association capable to work with potential data nonstationarities. We formulate the target quantity under the multivariate locally stationary wavelet framework, develop principled estimation through local wavelet spectral matrices, and incorporate principal-component reduction for stable adjustment in high-dimensions. Simulations show that the method removes spurious marginal association induced by confounding and accurately recovers direct association, including in higher-dimensional settings. Analysis of U.S. exchange-traded funds reveals substantial time-varying and scale-dependent direct canonical association after adjustment for external market effects.
Reference graph
Works this paper leans on
-
[1]
Biometrika , volume =
Relations Between Two Sets of Variates , author =. Biometrika , volume =. 1936 , publisher =
1936
-
[2]
Trabajos de Estadística y de Investigación Operativa , volume =
Partial canonical correlations , author =. Trabajos de Estadística y de Investigación Operativa , volume =
-
[3]
1999 , Address =
Convergence of Probability Measures , Author =. 1999 , Address =
1999
-
[4]
Box, G. E. P. and Tiao, G. C. , title =. Biometrika , volume =. 1977 , month =
1977
-
[5]
Annals of Applied Statistics , volume=
Multiscale spectral modelling for nonstationary time series within an ordered multiple-trial experiment , author=. Annals of Applied Statistics , volume=. 2022 , publisher=
2022
-
[6]
Journal of the American Statistical Association , volume =
Geweke, John , title =. Journal of the American Statistical Association , volume =. 1982 , month =
1982
-
[7]
Canonical Partial Autocorrelation Function of a Multivariate Time Series , volume =
Serge Dégerine , journal =. Canonical Partial Autocorrelation Function of a Multivariate Time Series , volume =
-
[8]
Brillinger, D. R. , title =. 2001 , publisher =
2001
Show all 24 references
-
[9]
The Annals of Statistics , volume=
A likelihood approximation for locally stationary processes , author=. The Annals of Statistics , volume=. 2000 , publisher=
2000
-
[10]
2014 , pages=
Park, Timothy and Eckley, Idris and Ombao, Hernando , title=. 2014 , pages=
2014
-
[11]
Schulte , title =
Justin A. Schulte , title =. Nonlinear Processes in Geophysics , volume =
-
[12]
Cohen, E A. K. and Walden, A. T. , journal=. Wavelet Coherence for Certain Nonstationary Bivariate Processes , year=
-
[13]
39th Conference on Neural Information Processing Systems (NeurIPS 2025) , year=
Wavelet Canonical Coherence for Nonstationary Signals , author=. 39th Conference on Neural Information Processing Systems (NeurIPS 2025) , year=
2025
-
[14]
Biometrika , volume =
Xiucai Ding and Zhou Zhou , title =. Biometrika , volume =
-
[15]
and Nunes, Matthew A
Knight, Marina I. and Nunes, Matthew A. and Hargreaves, Jessica K. , title=. 2024 , pages=
2024
-
[16]
The Generalized Dynamic-Factor Model: Identification and Estimation , volume =
Mario Forni and Marc Hallin and Marco Lippi and Lucrezia Reichlin , journal =. The Generalized Dynamic-Factor Model: Identification and Estimation , volume =
-
[17]
The Annals of Statistics , volume =
Clifford Lam and Qiwei Yao , title =. The Annals of Statistics , volume =
-
[18]
Biometrika , volume =
Palasciano, H A and Knight, M I and Nason, G P , title =. Biometrika , volume =. 2025 , month =
2025
-
[19]
, title =
Peña, Daniel and Yohai, Victor J. , title =. Journal of the American Statistical Association , volume =
-
[20]
and Sachs, Rainer and Kroisandt, Gerald , title=
Nason, Guy P. and Sachs, Rainer and Kroisandt, Gerald , title=. Journal of the Royal Statistical Society Series B , volume=. 2000 , publisher=
2000
-
[21]
Journal of the Royal Statistical Society Series B , volume=
A test for second-order stationarity and approximate confidence intervals for localized autocovariances for locally stationary time series , author=. Journal of the Royal Statistical Society Series B , volume=. 2013 , publisher=
2013
-
[22]
arXiv preprint arXiv:2305.10878 , year =
Wu, Haibo and Knight, Marina I and Ombao, Hernando , title =. arXiv preprint arXiv:2305.10878 , year =
-
[23]
1992 , publisher=
Daubechies, Ingrid , title=. 1992 , publisher=
1992
-
[24]
2002 , publisher =
Principal Component Analysis , author =. 2002 , publisher =
2002
Reviewed June 25, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.