Bandwidth selection with a frequency-domain version of the AIC
Bandwidth selection with a frequency-domain version of the AIC
Authors: Erhard Reschenhofer Date: 2026-06-01 Paper ID: openalex:2606.02295
Summary
This paper addresses the lack of standardized, objective methods for automatic bandwidth selection in nonparametric spectral density estimation, which often relies on subjective visual inspection. The author proposes a frequency-domain adaptation of the AIC, providing a computationally simple and reliable alternative for determining smoothing parameters. Through experiments on both synthetic and real-world time series, the study demonstrates that this automated approach yields results competitive with traditional parametric AR model-based estimation.
Key Contributions
- Introduces a frequency-domain version of the Akaike Information Criterion (AIC) for automatic bandwidth selection in nonparametric spectral density estimation.
- Demonstrates that the proposed method achieves estimation performance comparable to standard parametric AR-based approaches.
- Provides empirical validation using both synthetic and real-world time series with varying spectral complexity to support objective parameter selection over visual methods.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2606.02295
- paper_id
- 2606.02295
- paper_source
- openalex
- domain
- time-series
- tags
-
- architectures
-
- datasets
-
- concept_slugs
-
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-06-04T06:34:29Z