Change-Point Detection for Object-valued Time Series
Yi Zhang a {a}, Changbo Zhu and Xiaofeng Shao c {c} · 2026
Read Change-Point Detection for Object-valued Time Series by Yi Zhang a {a}, Changbo Zhu and Xiaofeng Shao c {c} (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-2606-00858.txt.
Introduction
Zhang, Zhu, and Shao hunt for breaks in streams of object-valued data in this 2026 paper. Object-valued data live in a metric space: networks, distributions, and shapes, not plain numbers. Past change-point tools handle lone data or spot shifts in mean or variance alone. Time-linked streams need more.
Their test rests on self-normalization. It looks for a shift in the whole marginal distribution of the series. The test needs almost no tuning, has a clean limit law, and uses pairwise distances alone. It also tolerates weak links across time. Paired with Wild Binary Segmentation, it can count and place many change points. They prove its traits under null and near-null cases. For the first time, they show segmentation of this kind is consistent for broad object-valued series. Wide tests and a real-data case back the claims, they report.
Catalog record
| CREATOR | Yi Zhang a {a}, Changbo Zhu and Xiaofeng Shao c {c} |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 135 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.00858
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Change-Point Detection for Object-valued Time Series Yi Zhang a {a}, Changbo Zhu and Xiaofeng Shao c {c} Abstract This article is concerned with change point detection for object-valued data that reside in a metric space, which has attracted some recent interests in statistics and econometrics literature. The existing methods either focus on independent data or can only detect change in the Fréchet mean or variance. In this paper, we propose a self-normalization (SN, hereafter) based statistic for detecting a shift in the marginal distribution of object-valued time series. Our test is universally applicable to a wide range of object-valued data, such as distributional and network data, and can accommodate weak serial dependence. In addition the proposed test statistic is almost tuning parameter free, has pivotal limiting null distribution and only uses the pairwise distances. When combined with the Wild Binary Segmentation algorithm (WBS, hereafter), our statistic can be used to estimate the number and locations of multiple change points. Asymptotic results for our SN based statistic are derived under both null and local alternatives in the single change point setting. For the first time, the WBS estimation consistency is shown for a broad class of object-valued time series and in a nonparametric setting, which requires new non-standard theoretical arguments. Extensive numerical experiments and real data analysis are conducted to illustrate the effectiveness and broad applicability of our proposed method. Keywords: Binary Segmentation; Functional data; Non-Euclidean Data; Sample Splitting; Self-normalization. 1 Introduction Statistical analysis of non-Euclidean data (object-valued data or random objects) that reside in a metric space is gradually emerging as an important branch of functional data analysis in econometrics and statistics, motivated by increasing encounter of such data in many modern applications. Examples include the analysis of sequences of age-at-death distributions over calendar years ( 38 ; 47 ) , covariance matrices in the analysis of diffusion tensors in medical imaging ( 19 ) , compositional brain volumes ( 62 ; 60 ) and graph Laplacians of networks ( 28 ; 53 ) . One of the main challenges in dealing with such data is that the usual vector/Hilbert space operations, such as projection and inner product may not be well defined and only the distance between two non-Euclidean data objects is available. Despite the challenge, the list of papers that propose new statistical techniques to analyze non-Euclidean data has been growing. See 21 ; 22 , 41 , 52 , 61 , 56 , 63 ; 64 , 32 ; 33 , among others. In this paper, we shall focus on offline change-point detection for object-valued time series. Such temporally dependent objects are prevalent in practice and well-known examples include yearly age-at-death distributions for countries in Europe ( 22 ) , annual observations of compositions of energy sources for power generation in the US ( 64 ) , and daily Pearson correlation matrices for five cryptocurrencies ( 33 ) . Also see Section 5 for additional data examples. The literature for offline change-point detection of object-valued data is quite recent. For example, 12 and 15 developed graph-based scan statistics for both the single change-point and the changed interval alternatives to detect the distributional change for independent object data; see 11 for a recent review on graph-based change point detection methods. Recently, 22 proposed a novel test for a change point in the Fréchet mean and variance ( 24 ) for a sequence of independent random objects and adopted binary segmentation for change point estimation. 20 proposed the concept of distance profiles, which fully characterize the distribution of random objects under conditions that basically require the metric space to be of strong negative type (see proposition 1 therein). Building on distance profile, 23 developed a test which can detect distributional change for serially independent random objects. Note that theoretical results developed for the above-mentioned methods are all for single change point detection. To accommodate serial dependence, 33 developed an self-normalization (SN, hereafter) based test by extending the SN idea ( 49 ; 48 ; 50 ) from vector time series to object-valued time series. However, the method in 33 is also limited to capture change point in Fréchet mean or variance only and their test may not be powerful when the change is in high-order characteristics of distributions; see numerical evidence in Section 4 . In contrast to the limited literature for change-point detection of object-valued data, the same problem for functional time series, which resides in a Hilbert space, has been extensively studied. For example, single and multiple change point detection probem for functional data has been studied in 57 ; 3 ; 51 ; 4 ; 34 ; 13 ; 42 ; 29 ; see Chapter 8 in 30 for a comprehensive review of the main developments in this area. It is worth noting that these methods can not be easily extended to the object-valued time series setting, due to the lack of algebraic operations for metric space valued data. The main goal of this paper is to fill in the gap and develop new testing and estimation procedures that target the shift in marginal distributions of object-valued time series with serial dependence. On the testing front, we propose a non-parametric method that can detect distributional changes for metric space valued time series based on sample splitting and self-normalization. To construct our test statistic, the object-valued data are first mapped into a Hilbert space, and then projected along the direction of change. After that, an SN-based test statistic is constructed using the projected one-dimensional sequence. In the single change point setting, the statistic we developed is nearly tuning parameter free (except for a cut-off parameter which restricts the true change point from being too close to the end-points) and has pivotal limiting null distribution. On the estimation front, we utilize the WBS algorithm ( 25 ) for multiple change point estimation and provide theoretical guarantee in terms of estimation consistency, which is a new contribution to the literature. As one of the state-of-the-art segmentation algorithm with promising practical performance, WBS has been combined with various single change point detection methods for Euclidean or metric space valued time series to enable multiple change point estimation (see e.g., 9 , 54 , 58 , 55 and 33 ) without theoretical justification. The WBS algorithm was originally developed to detect change points in the mean of univariate i.i.d data, and its performance on temporally dependent data could depend on the estimation of the long run variance (LRV), which was known to be a difficult problem due to the difficulty of choosing the bandwidth in the presence of change-points ( 16 ; 48 ; 43 ) . By using self-normalization, our method avoids direct estimation of LRV. At the same time, the adoption of a self-normalizer and the usage of projection in our test statistic significantly complicate our proof for estimation consistency, which requires substantially different theoretical arguments from those in 25 . To our knowledge, except for 35 where the WBS estimation consistency was proved for a parametric time series model, we are the first to show the WBS estimation consistency for a broad class of time series (which includes Euclidean valued time series as a special case) and in