Pith. sign in

REVIEW 2 major objections 1 minor 3 references

A frequency-domain version of the AIC automatically selects bandwidth for nonparametric spectral estimation with results comparable to parametric methods.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-28 11:51 UTC pith:GPTQDSZW

load-bearing objection The paper applies a basic frequency-domain AIC to bandwidth selection and claims comparability to AR-AIC on real and synthetic series, but the support rests on unspecified metrics and limited cases. the 2 major comments →

arxiv 2606.02295 v1 pith:GPTQDSZW submitted 2026-06-01 stat.AP

Bandwidth selection with a frequency-domain version of the AIC

Erhard Reschenhofer This is my paper
classification stat.AP
keywords bandwidth selectionAICspectral density estimationnonparametric estimationfrequency domaintime series analysisautomatic selectionparametric comparison
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper sets out to show that the simplest frequency-domain AIC can determine an appropriate bandwidth for nonparametric spectral density estimates and produce results comparable to the standard parametric approach of fitting AR models with AIC order selection. Evidence comes from applications to both real-world time series and synthetic series whose spectral densities vary in complexity. A sympathetic reader would care because nonparametric estimation is flexible yet often relies on subjective visual choices for the smoothing parameter, and an objective automatic alternative could reduce that dependence without requiring prior knowledge of the unknown spectrum. If the claim holds, practitioners could obtain reliable nonparametric estimates through a straightforward criterion rather than case-by-case tuning.

Core claim

The central claim is that applying the simplest and most straightforward frequency-domain version of the AIC to select the bandwidth for nonparametric spectral density estimation yields results comparable to those obtained from parametric estimation using AR models and AIC order selection, as demonstrated on real-world time series and on synthetic series with spectral densities of varying complexity.

What carries the argument

The frequency-domain AIC criterion used to choose the smoothing bandwidth in nonparametric spectral estimation.

Load-bearing premise

That comparability of results on the selected real-world and synthetic series is sufficient to recommend the method for general use without concrete prior information about the unknown spectral density.

What would settle it

A time series, either real or synthetic, on which the nonparametric estimate obtained with the frequency-domain AIC bandwidth shows substantially larger error than the parametric AR estimate would falsify the comparability claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a simple frequency-domain version of the AIC criterion for automatic bandwidth selection in nonparametric spectral density estimation. It claims this yields results comparable to the standard parametric AR-AIC approach, with supporting evidence drawn from applications to real-world time series and synthetic series whose spectral densities have varying complexity.

Significance. If the comparability claim can be substantiated with explicit quantitative metrics, the method would offer a straightforward objective alternative to subjective visual bandwidth choice and could encourage wider adoption of automatic nonparametric procedures. The approach is conceptually direct, but the current evidence base does not yet establish general reliability.

major comments (2)
  1. [Abstract] Abstract: the claim that the frequency-domain AIC 'enables results that are comparable' to the parametric approach supplies no quantitative performance metric (e.g., ISE, MISE), error bars, or explicit description of how comparability was measured across the real and synthetic examples.
  2. [Synthetic experiments] Synthetic experiments: the description of 'synthetic time series with spectral densities of varying complexity' does not indicate whether the chosen spectra include cases in which low-order AR fits are known to be poor while nonparametric smoothing remains appropriate; without such coverage the evidence supports only case-by-case similarity rather than a general recommendation.
minor comments (1)
  1. [Methods] Provide the precise algebraic form of the frequency-domain AIC (including the definition of the periodogram and any smoothing weights) in the methods section so that the criterion can be reproduced exactly.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address each major comment below and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the frequency-domain AIC 'enables results that are comparable' to the parametric approach supplies no quantitative performance metric (e.g., ISE, MISE), error bars, or explicit description of how comparability was measured across the real and synthetic examples.

    Authors: We agree that the abstract would be improved by a concise statement of the quantitative criteria used to assess comparability. In the revised version we will add a sentence specifying that comparability was evaluated via integrated squared error on the synthetic examples and via a combination of ISE and visual inspection on the real series, with the numerical values reported in the main text. revision: yes

  2. Referee: [Synthetic experiments] Synthetic experiments: the description of 'synthetic time series with spectral densities of varying complexity' does not indicate whether the chosen spectra include cases in which low-order AR fits are known to be poor while nonparametric smoothing remains appropriate; without such coverage the evidence supports only case-by-case similarity rather than a general recommendation.

    Authors: The synthetic spectra were deliberately chosen to include both low-order AR processes and spectra with multiple narrow peaks and troughs that are known to be poorly approximated by low-order AR models. To make this explicit we will expand the methods section to list the exact spectral densities employed, cite the literature showing where low-order AR fits fail, and note which examples fall into that category. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical validation of frequency-domain AIC bandwidth selector

full rationale

The paper presents an empirical comparison showing that a frequency-domain AIC for nonparametric bandwidth selection yields results comparable to parametric AR-AIC on selected real and synthetic series. No derivation chain is claimed; the central claim rests on direct performance matches rather than any self-definitional equation, fitted parameter renamed as prediction, or load-bearing self-citation. The method is introduced as a straightforward adaptation of AIC without reduction to its own inputs by construction. This is the expected non-finding for an applied methodological paper whose evidence is external to its own fitted values.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the central claim rests on an unstated assumption that the tested series adequately represent general spectral densities.

pith-pipeline@v0.9.1-grok · 5694 in / 1022 out tokens · 16126 ms · 2026-06-28T11:51:51.019432+00:00 · methodology

0 comments
read the original abstract

When it comes to estimating an unknown spectral density as simply and reliably as possible, parametric spectral density estimation using AR models and order selection via AIC is the method of choice. In contrast, no standard method has yet emerged for automatic nonparametric spectral density estimation, and there seems to be little willingness to weigh the advantages and disadvantages of different risk functions and the various methods for estimating them on a case-by-case basis, particularly because it is unclear whether the effort is even worthwhile without concrete prior information about the unknown spectral density. As a result, subjective visual methods are still widely used in practice to determine the appropriate smoothing parameter for a nonparametric estimation. This article aims to encourage the increased use of objective automatic methods by presenting evidence that using what is arguably the simplest and most straightforward frequency-domain version of the AIC for the automatic determination of an appropriate bandwidth enables results that are comparable to those obtained using the standard parametric approach. This evidence is based on both real-world time series and synthetic time series with spectral densities of varying complexity.

Figures

Figures reproduced from arXiv: 2606.02295 by Erhard Reschenhofer.

Figure 1
Figure 1. Figure 1: (a) Monthly GMST from January 1850 to December 2025 (HadCRUT5) [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Quarterly GDPC1 from 1947.I to 2026.I (Bureau of Economic Analysis) [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Periodograms (gray) of (a) monthly GMST, (b) detrended GMST, (c) monthly [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Periodograms (gray) of (a) quarterly GDPC1, (b) monthly UNRATE, (c) daily [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Periodograms (gray) of trimmed (a) quarterly GDPC1, (b) monthly UNRATE, [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The log periodogram of the trimmed daily EURUSD=X at the first twenty [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Spectral densities (black) of data generating MA processes (22) with parameter [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    (1969) Fitting autoregressive models for prediction

    Akaike, H. (1969) Fitting autoregressive models for prediction. Annals of the Institute of Statistical Mathematics, 21, 243-247. https://doi.org/10.1007/BF02532251 Akaike, H. (1973) Information theory and an extension of the maximum likelihood principle. In: Petrov, B.N., Csaki, F. (eds.), Second International Symposium on Information Theory, Akademia Kia...

    work page doi:10.1007/bf02532251 1969

  2. [2]

    (1978) Information criteria for discriminating among alternative regression models

    https://doi.org/10.2307/1267069 Sawa, T. (1978) Information criteria for discriminating among alternative regression models. Econometrica, 46, 6, 1273-1291. https://doi.org/10.2307/1913828 Schwarz, G. (1978) Estimating the dimension of a model. The Annals of Statistics, 6, 461-

    work page doi:10.2307/1267069 1978

  3. [3]

    (1980) Asymptotically efficient selection of the order of the model for estimating parameters of a linear process

    https://www.jstor.org/stable/2958889 14 Shibata, R. (1980) Asymptotically efficient selection of the order of the model for estimating parameters of a linear process. Annals of Statistics, 8, 147-164. https://www.jstor.org/stable/2240749 Shibata, R. (1981) Optimal selection of regression variables. Biometrika, 68, 1, 45-54. https://doi.org/10.2307/2335804...

    work page doi:10.2307/2335804 1980