Change‑point estimation for Weibull time series with copula‑based Markov models

Li-Hsien Sun, Zong-Yuan Huang, Yi-Ling Huang, Chi-Yang Chiu, Ning Ning · 2026

Read Change‑point estimation for Weibull time series with copula‑based Markov models by Li-Hsien Sun, Zong-Yuan Huang, Yi-Ling Huang, Chi-Yang Chiu, Ning Ning (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-29541.txt.

Introduction

Li-Hsien Sun and four colleagues build a tool to spot pattern shifts in a time series with linked, lopsided dependence. Their model joins Weibull margins for non-negative measures with Clayton and Joe copulas for lower-tail and upper-tail links. Change points and model terms come from a maximum likelihood fit, with bootstrap intervals to gauge doubt. Trials show the estimators hold up well even when the copula form is wrong. A run on the VIX gauge through the COVID-19 shock catches shifts in spread and links.

Catalog record

CREATORLi-Hsien Sun, Zong-Yuan Huang, Yi-Ling Huang, Chi-Yang Chiu, Ning Ning
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE137536
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.29541

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Change‑point estimation for Weibull time series with copula‑based Markov models Li-Hsien Sun, Zong-Yuan Huang, Yi-Ling Huang, Chi-Yang Chiu, Ning Ning Abstract We study offline change-point estimation for time series data exhibiting nonlinear serial dependence. To address this problem, we propose a copula-based Markov chain model with Weibull marginal distributions, which is suitable for modeling nonnegative data such as event times and volatility measures. Nonlinear dependence is incorporated through the Clayton and Joe copulas, allowing the model to capture asymmetric lower-tail and upper-tail dependence structures, respectively. We derive the corresponding likelihood function and estimate the change point and model parameters using maximum likelihood estimation implemented through the Newton–Raphson algorithm. Confidence intervals are constructed via a parametric bootstrap Monte Carlo procedure. Extensive numerical studies are conducted to evaluate the finite-sample performance and robustness of the proposed method under different dependence structures and copula misspecification scenarios. The results demonstrate that the proposed estimators perform well in terms of RMSE and relative error, particularly for the estimation of the change point. An empirical application to the VIX index during the COVID-19 pandemic further illustrates the practical usefulness of the proposed approach in detecting structural changes in both the marginal distributions and serial dependence structure. Keywords: Change-point estimation; Copula-based Markov chain; Monte Carlo simulation; Weibull distribution; Maximum likelihood estimation. 1 Introduction Detecting change-points in sequential data is an important problem in statistical applications because structural changes often reflect significant shifts in the underlying data-generating mechanism. Change-point analysis has been widely applied in many areas, including finance 13 , climatology 18 , reliability studies, and medical research 16 . In general, change-point methodology can be divided into two categories: offline estimation and online monitoring. Offline change-point analysis aims to estimate structural changes retrospectively using a complete dataset observed over a fixed time horizon, whereas online monitoring procedures are designed to detect changes sequentially in real time as new observations become available. Most classical change-point methods are developed under assumptions of independent observations or linear dependence structures, such as autoregressive models 24 . Under these settings, change-points are commonly estimated through likelihood-based methods, particularly maximum likelihood estimation (MLE). However, linear dependence assumptions are often inadequate for describing the nonlinear and asymmetric dependence structures frequently observed in practical time series data 15 ; 7 . To overcome this limitation, copula-based Markov chain models provide a flexible framework for modeling nonlinear serial dependence 2 ; 11 . By separating the marginal distributions from the dependence structure, copula models allow greater modeling flexibility and can capture asymmetric lower- and upper-tail dependence that conventional linear models fail to describe. Moreover, copula-based Markov models satisfy strict stationarity under mild regularity conditions and have been shown to perform well in nonlinear time series settings; see, for example, 1 and 6 . In this paper, we propose a copula-based Markov chain model with Weibull marginal distributions for offline change-point estimation in nonnegative time series exhibiting nonlinear dependence. The Weibull distribution is widely used in survival analysis, reliability modeling, and duration analysis because of its flexibility in accommodating different hazard rate behaviors. To capture asymmetric tail dependence structures, we incorporate the Clayton and Joe copulas into the Markov framework, where the Clayton copula is capable of modeling stronger lower-tail dependence and the Joe copula captures upper-tail dependence. This framework allows us to simultaneously model nonlinear serial dependence and structural changes in the marginal distributions and dependence structures. Related copula-based approaches have been considered in the literature. For example, Emura and Murotani 5 proposed estimation procedures for copula-based dependent truncation models, while Sun et al. 21 developed Bayesian inference methods for copula-based Markov time series models with Student’s t marginals. For statistical inference, we develop a likelihood-based estimation procedure using maximum likelihood estimation. Since the likelihood function is not differentiable with respect to the change point, we employ a profile likelihood approach combined with the Newton–Raphson algorithm to obtain the parameter estimates numerically 4 ; 22 . To ensure numerical stability under parameter constraints, a suitable re-parameterization technique is adopted. In addition, confidence intervals for both the model parameters and the change point are constructed using a parametric bootstrap Monte Carlo procedure. The finite-sample performance of the proposed method is investigated through extensive simulation studies under different sample sizes, dependence strengths, change-point locations, and copula misspecification scenarios. The simulation results demonstrate that the proposed estimators perform well in terms of RMSE and relative error (RE), particularly for the estimation of the change point. The results also show that the proposed procedure remains relatively robust under moderate copula misspecification, although the estimation of tail dependence parameters becomes more sensitive when the copula family is misspecified. To illustrate the practical usefulness of the proposed framework, we apply the model to the VIX index during the COVID-19 pandemic period. The VIX index, commonly referred to as the “fear index,” measures market expectations of future volatility over a 30-day horizon and is widely used as an indicator of financial market uncertainty. The empirical results identify structural changes associated with the onset of the COVID-19 pandemic and show that the Clayton copula-based Markov model provides the best fit according to the Akaike Information Criterion (AIC), suggesting the presence of stronger downside dependence during periods of market stress. The remainder of this paper is organized as follows. Section 2 introduces the proposed copula-based Markov chain model with Weibull marginal distributions and a single change point. Section 3 presents the likelihood-based estimation procedure, the Newton–Raphson algorithm, and the bootstrap Monte Carlo interval estimation method. Section 4 reports the simulation studies and investigates the effects of dependence strength and copula misspecification. Section 5 presents the empirical analysis of the VIX data during the COVID-19 pandemic. Finally, Section 6 concludes the paper and discusses possible directions for future research. 2 The Proposed Model We describe the copula-based Markov Weibull time series with one structural change. We first introduce the Weibull distribution and then the copula functions. Finally, we propose the conditional density for the likelihood inference based on copulas and marginals. 2.1 Weibull distribution Marginals Let X be a random variable driven by the Weibull distribution with the scale parameter \lambda>0 and the shape parameter k>0 , respectively. The cumulative distribution function (CDF) is F(x)=1-e^{-(\frac{x}{\lambda})^{k}},\quad x\geq 0, and the probability density function (PDF) is f(x)=\frac{k}{\lambda}\left(\frac{x}{\lambda}\right)^{k-1}e^{-(\frac{x}{\lambda})^{k}},\quad x\geq 0. The mean and variance are \mathbb{E}(X)=\lambda\Gamma\left(1+\frac{1}{k}\right) and \operatorname{Var}(X)=\lambda^{2}\left[\Gamma\left(1+\frac{2}{k}\right)-\Gamma\left(1+\frac{1}{k}\right)^{2}\right] , respectively. The shape parameter k has a more significant impact on the distribution because it directly determines the fundamental shape and characteristics of the distribution, including whether a peak exists, the direction of skewness, and the presence of long tails. When k>1 , the distribution becomes more symmetric and exhibits a distinct peak, meaning that the data is more concentrated, resembling a normal distribution. When k<1 , the distribution is right-skewed, with a pronounced long-tail effect. On the other hand, the scale parameter \lambda only affects the translation and stretching of the distribution. As \lambda increases, the distribution spreads out more, but its shape remains unchanged. When the scale parameter is fixed, the estimation of the shape parameter is relatively stable, and the estimation error remains small. This is because the scale parameter primarily affects the spread of the distribution without altering its overall shape characteristics. In contrast, when

Read the complete text