{"id":"8347b570-4e75-474d-ab66-e37cc666c13a","arxiv_id":"2606.24554","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"Introduces Partial Wavelet Canonical Coherence as the first frequency-domain partial canonical correlation method for multivariate nonstationary time series using wavelets and principal component reduction.","lead":"The paper introduces Partial Wavelet Canonical Coherence, a new wavelet-based approach to measure direct associations between two multivariate nonstationary time series after adjusting for high-dimensional confounders. A generalist reader might examine it for tools that handle time-varying relationships in complex datasets such as financial markets or biological signals.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"Reader correctly flags the locally stationary wavelet assumption as the point where the entire target quantity and estimation procedure are defined. No additional load-bearing gap is visible from the supplied material.","tokens_in":1637,"tokens_out":218,"duration_ms":11437,"concrete_test":"Re-derive the population partial canonical coherence quantity from the local wavelet spectral matrix definition in the absence of the locally stationary wavelet assumption; if the quantity remains well-defined and the estimator remains consistent, the framework is not load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central construction defines the target partial canonical coherence and its estimator under the multivariate locally stationary wavelet model. This framework supplies the local spectral matrices whose eigenstructure yields the coherence after PCA-based confounder adjustment. No internal inconsistency appears in the abstract-level description of the construction, and the simulations are reported to recover direct association after confounding removal. The \"first method\" claim is a standard novelty statement whose verification lies outside the paper's technical argument.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","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.","tokens_in":1709,"tokens_out":501,"duration_ms":15439,"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":[{"comment":"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.","section":"Simulations section"},{"comment":"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.","section":"Methodology / Model assumptions"}],"minor_comments":[{"comment":"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.","section":"Introduction"},{"comment":"Notation for the local spectral matrices and the PCA reduction step should be introduced with explicit definitions and dimensions to aid readability.","section":"Estimation procedure"}],"recommendation":"major_revision","confidential_remarks":"The simulation validation gap noted above is the primary concern; if the full manuscript contains only qualitative simulation statements, this would warrant requesting quantitative results before acceptance. The scope fits stat.ME well."},"author_rebuttal":{"model":"grok-4.3","summary":"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.","responses":[{"response":"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_made":"yes","referee_comment":"[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."},{"response":"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_made":"yes","referee_comment":"[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."}],"tokens_in":1293,"tokens_out":488,"duration_ms":14692,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"This paper introduces Partial Wavelet Canonical Coherence to measure direct canonical associations between two sets of nonstationary time series after adjusting for high-dimensional confounders. It claims to be the first frequency-domain partial canonical correlation formulation for time series.\n\nThe new element is the combination of wavelet localization to get local spectral matrices under the multivariate locally stationary wavelet model, followed by PCA reduction for stable high-dimensional adjustment. This produces a scale-specific, time-varying partial coherence. The ETF example shows it recovering time- and scale-dependent direct links once market effects are removed, which matches a recurring need in applied work.\n\nThe construction looks internally consistent with no detectable circularity. The weakest assumption is that the series fit the locally stationary wavelet framework, which is required to define the target quantity but is a standard modeling choice rather than a hidden flaw.\n\nThe main soft spot is the simulation evidence. The abstract states that the method removes spurious marginal associations and recovers direct ones, including in higher dimensions, yet gives no quantitative metrics, error bars, or description of the data-generating processes. That leaves the performance claim hard to evaluate from the given information.\n\nThis is for researchers working with frequency-domain questions in nonstationary multivariate series who need to partial out confounders. A reader focused on time-frequency methods or high-dimensional time series would get something from it. It deserves peer review because the problem is practical and the approach is new, even if the simulations will require closer checking.","headline":"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.","tokens_in":2173,"tokens_out":370,"would_cite":false,"duration_ms":15558,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A new wavelet-based partial canonical coherence isolates direct associations between nonstationary multivariate time series after adjustment for high-dimensional confounders.","keywords":["partial canonical coherence","wavelet analysis","nonstationary time series","high-dimensional confounders","multivariate locally stationary wavelet","time-varying association","frequency-domain analysis","canonical correlation"],"falsifier":"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.","tokens_in":2536,"feed_emoji":"","tokens_out":866,"duration_ms":11067,"temperature":0.7,"pith_summary":"The paper develops Partial Wavelet Canonical Coherence to measure the direct canonical association between two multivariate nonstationary time series after removing the effects of possibly many confounding series. It provides the first frequency-domain version of partial canonical correlation analysis for time series by working in the wavelet domain, which produces a measure that changes over time and across scales. This is useful because real signals such as asset returns or physiological recordings are often nonstationary and contain indirect links created by shared external drivers. The method is defined under the multivariate locally stationary wavelet model, estimated from local wavelet spectral matrices, and stabilized in high dimensions by principal-component reduction on the confounder block. Simulations demonstrate that the estimator removes spurious marginal associations and recovers the true direct relationships, while an analysis of exchange-traded funds shows time-varying and scale-dependent direct links once market-wide effects are removed.","feed_headline":"Wavelet method isolates direct links after high-dimensional adjustment","feed_subtitle":"Partial canonical coherence defined on local wavelet spectra removes spurious associations and recovers time-scale specific direct relations","key_machinery":"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.","core_discovery":"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","pith_inferences":["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."],"forward_implications":["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."],"fun_headline_variants":["Wavelet partial coherence measures direct multivariate time series associations","Adjusts for high-dimensional confounders using local wavelet spectral matrices","Time-scale specific direct canonical links after partial wavelet adjustment","Frequency-domain formulation of partial canonical analysis for time series"],"cache_read_input_tokens":64,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Wavelet partial coherence measures direct multivariate time series associations","Adjusts for high-dimensional confounders using local wavelet spectral matrices","Time-scale specific direct canonical links after partial wavelet adjustment","Frequency-domain formulation of partial canonical analysis for time series"]},"model":"grok-4.3","cost_usd":0.006469,"raw_usage":{"total_tokens":3010,"prompt_tokens":630,"num_sources_used":0,"completion_tokens":62,"cost_in_usd_ticks":64687000,"prompt_tokens_details":{"text_tokens":630,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2318,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":630,"tokens_out":62,"duration_ms":15254,"temperature":1.0,"reasoning_tokens":2318,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T23:05:22.129111+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"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.","supporting_citations":[],"review_version":1}