[Paper Review] Quantile Spectral Analysis for Locally Stationary Time Series

Stefan Birr, Stanislav Volgushev+3 more|arXiv (Cornell University)|Apr 17, 2014

Statistical Methods and Inference53 references4 citations

TL;DR

This paper introduces a quantile spectral analysis framework for locally stationary time series by extending copula-based spectral concepts to time-varying dependence structures. It proposes a local lag-window estimator for time-varying copula spectra and establishes a central limit theorem, enabling detection of non-Gaussian, nonstationary serial dependence across quantiles and time—revealing tail dynamics undetected by classical covariance-based methods.

ABSTRACT

Classical spectral methods are subject to two fundamental limitations: they only can account for covariance-related serial dependencies, and they require second-order stationarity. Much attention has been devoted lately to quantile-based spectral methods that go beyond covariance-based serial dependence features. At the same time, covariance-based methods relaxing stationarity into much weaker {\it local stationarity} conditions have been developed for a variety of time-series models. Here, we are combining those two approaches by proposing quantile-based spectral methods for locally stationary processes. We therefore introduce a time-varying version of the copula spectra that have been recently proposed in the literature, along with a suitable local lag-window estimator. We propose a new definition of local {\it strict} stationarity that allows us to handle completely general non-linear processes without any moment assumptions, thus accommodating our quantile-based concepts and methods. We establish a central limit theorem for the new estimators, and illustrate the power of the proposed methodology by means of a simulation study. Moreover, in two empirical studies (namely of the Standard \& Poor's 500 series and a temperature dataset recorded in Hohenpeissenberg) we demonstrate that the new approach detects important variations in serial dependence structures both across time and across quantiles. Such variations remain completely undetected, and are actually undetectable, via classical covariance-based spectral methods.

Motivation & Objective

• To extend quantile-based spectral analysis beyond strictly stationary processes to handle locally stationary time series.
• To develop a time-varying copula spectral concept that captures serial dependence across quantiles without moment assumptions.
• To propose a local lag-window estimator for time-varying copula spectra with theoretical justification.
• To establish a central limit theorem for the proposed estimators under weak regularity conditions.
• To demonstrate the method's ability to detect non-Gaussian, nonstationary dependence structures in real financial and environmental data.

Proposed method

• Proposes a new definition of local strict stationarity that allows non-linear, non-Gaussian processes without moment restrictions.
• Introduces a time-varying copula spectrum indexed by quantile levels $(\tau_1, \tau_2)$ and time $t$, capturing conditional dependence across quantiles.
• Develops a local lag-window estimator using kernel-weighted averaging of rank-based periodograms to estimate time-varying copula spectra.
• Applies a local spectral smoothing technique with bandwidth $B_n$ and kernel $K_n(k)$ to control bias and variance.
• Uses rank-based periodograms derived from empirical copula processes to avoid moment assumptions and enable robustness to heavy tails.
• Establishes asymptotic normality of the estimator via a central limit theorem under weak regularity conditions on the spectral density and kernel function.

Experimental results

Research questions

• RQ1 Can quantile-based spectral methods detect time-varying serial dependence in non-Gaussian, nonstationary time series where classical methods fail?
• RQ2 How can copula spectral concepts be extended to locally stationary processes without requiring second-order stationarity or finite moments?
• RQ3 What is the asymptotic behavior of the proposed local lag-window estimator for time-varying copula spectra?
• RQ4 To what extent can the method detect tail dependence dynamics in financial and environmental time series?
• RQ5 How does the proposed method compare to classical periodograms in detecting serial dependence across different quantiles?

Key findings

• The proposed method detects strong tail dependence in the S&P500 index series—evident as peaks at the origin in rank-based copula periodograms for $\tau = 0.1$ and $\tau = 0.9$, which are absent in the median ($\tau = 0.5$) and undetected by classical methods.
• In a temperature dataset from Hohenpeissenberg, the method reveals time-varying serial dependence structures across quantiles, indicating nonstationary tail behavior.
• The central limit theorem for the local estimator is established under weak regularity conditions, including uniform convergence and integrability of spectral derivatives.
• The bias of the estimator is shown to be of order $O(B_n^{-r})$, with optimal bandwidth choice balancing bias and variance.
• The method successfully identifies non-Gaussian, nonstationary dependence in both simulated and real-world data, outperforming classical covariance-based spectral analysis.
• Theoretical results confirm that the estimator is consistent and asymptotically normal, even under non-linear, non-elliptical, and heavy-tailed dependence structures.