pith. sign in

arxiv: 1907.09522 · v1 · pith:DT4J76HUnew · submitted 2019-07-22 · 📊 stat.ME

Factor Analysis for High-Dimensional Time Series with Change Point

Xialu Liu , Ting Zhang This is my paper

Pith reviewed 2026-05-24 17:52 UTC · model grok-4.3

classification 📊 stat.ME
keywords factor analysischange pointhigh-dimensional time seriesself-normalizationlatent factor modelsstructural breakscross-sectional dependence
0
0 comments X

The pith

Consistent estimators exist for factor loading spaces before and after a single change point in high-dimensional time series models, even when noise has strong cross-sectional dependence.

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

The paper develops methods for latent factor models of high-dimensional time series that allow a structural break in the factor loadings. It constructs consistent estimators for the loading spaces on each side of the break and also estimates the break location itself. The approach works when the noise terms exhibit strong cross-sectional dependence, a setting not covered by earlier results. Self-normalization is applied to the change-point test statistic so that it becomes pivotal without needing to know the strength of that dependence. Readers in statistics and econometrics would care because many real high-dimensional series, such as asset returns or macroeconomic indicators, display both breaks and cross-sectional correlations.

Core claim

In a change-point latent factor model for high-dimensional time series, consistent estimators can be obtained for the factor loading spaces before and after the change point, and the change-point location itself can be estimated; these results continue to hold when the noise process has strong cross-sectional dependence because self-normalization is used to make the change-point test statistic pivotal.

What carries the argument

Self-normalization of the change-point test statistic, which removes the need to know the precise strength of cross-sectional dependence in the noise and thereby produces a pivotal limiting distribution.

If this is right

  • The estimators of the pre-change and post-change loading spaces are consistent.
  • The single change-point location can be consistently estimated.
  • The change-point test remains valid and pivotal for any unknown degree of cross-sectional noise dependence.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same self-normalization device might be reusable for multiple change points if the single-break theory extends.
  • Financial or macroeconomic panels with common shocks could now be analyzed without first assuming weak cross-sectional correlation.
  • Rates of convergence could be derived explicitly as functions of the dependence strength to guide sample-size requirements.

Load-bearing premise

The data follow a latent factor model with exactly one change point in the loading space, and the noise terms obey moment and weak-dependence conditions that make the self-normalized statistic converge to a known limit.

What would settle it

A Monte Carlo experiment or real-data example in which strong cross-sectional dependence is present yet the self-normalized test statistic fails to converge to its claimed pivotal distribution would refute the consistency and pivotalization claims.

Figures

Figures reproduced from arXiv: 1907.09522 by Ting Zhang, Xialu Liu.

Figure 1
Figure 1. Figure 1: Histograms of estimated threshold value under different settings when [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
read the original abstract

We consider change-point latent factor models for high-dimensional time series, where a structural break may exist in the underlying factor structure. In particular, we propose consistent estimators for factor loading spaces before and after the change point, and the problem of estimating the change-point location is also considered. Compared with existing results on change-point factor analysis of high-dimensional time series, a distinguished feature of the current paper is that our results allow strong cross-sectional dependence in the noise process. To accommodate the unknown degree of cross-sectional dependence strength, we propose to use self-normalization to pivotalize the change-point test statistic. Numerical experiments including a Monte Carlo simulation study and a real data application are presented to illustrate the proposed methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript develops change-point analysis for high-dimensional latent factor models of time series. It claims consistent estimators for the pre- and post-break factor loading spaces, a procedure for estimating the unknown change-point location, and a self-normalized CUSUM-type test statistic that remains pivotal without estimating the strength of cross-sectional dependence in the noise. The results are stated to hold under strong cross-sectional dependence in the idiosyncratic errors, with supporting Monte Carlo experiments and a real-data illustration.

Significance. If the self-normalized statistic converges to a pivotal limit (e.g., Brownian bridge) uniformly in the cross-section when the noise covariance eigenvalues may grow with dimension p, the work would usefully extend existing change-point factor methods that typically impose weaker dependence. The explicit allowance for strong cross-sectional correlation and avoidance of long-run variance estimation could improve robustness in applications such as macroeconomics or finance.

major comments (2)
  1. [Theoretical results on the test statistic (around the self-normalization step)] The pivotal property of the self-normalized test statistic is load-bearing for both detection and change-point estimation. The manuscript must verify that the normalization constant converges uniformly (in probability) to a non-random positive limit when the maximum eigenvalue of the noise covariance is allowed to grow with p; standard self-normalization arguments require this uniformity, yet the stated moment/mixing conditions on the noise process do not obviously guarantee it when cross-sectional dependence strength is arbitrary.
  2. [Change-point estimation procedure and its consistency proof] Consistency of the change-point location estimator relies on the test statistic behaving like a Brownian bridge under the null and diverging under the alternative. If the self-normalization fails to be uniform, the argmax of the normalized CUSUM may not be consistent; the paper should supply a explicit rate condition on the growth of the noise covariance eigenvalues relative to p and T that restores the required uniform convergence.
minor comments (2)
  1. [Model setup] Notation for the factor loading matrices before and after the break should be introduced once and used consistently; the current alternation between pre/post subscripts and time-indexed versions is occasionally ambiguous.
  2. [Simulation study] The Monte Carlo section would benefit from reporting the empirical size of the self-normalized test under the strongest dependence configurations considered, to illustrate that the pivotal limit is attained in finite samples.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help clarify the theoretical foundations of our self-normalized procedures. We address each major comment below.

read point-by-point responses

  1. Referee: [Theoretical results on the test statistic (around the self-normalization step)] The pivotal property of the self-normalized test statistic is load-bearing for both detection and change-point estimation. The manuscript must verify that the normalization constant converges uniformly (in probability) to a non-random positive limit when the maximum eigenvalue of the noise covariance is allowed to grow with p; standard self-normalization arguments require this uniformity, yet the stated moment/mixing conditions on the noise process do not obviously guarantee it when cross-sectional dependence strength is arbitrary.

    Authors: We appreciate the referee drawing attention to the uniformity requirement. Under the moment and strong-mixing conditions stated in Assumption 2.3, the quadratic-variation process used for self-normalization converges uniformly in probability to a positive non-random limit because the mixing coefficients control the dependence uniformly across the cross-section, even when the largest eigenvalue of the noise covariance grows with p (provided the growth is compatible with the moment bounds). To make this explicit and address the concern, we will insert a supporting lemma establishing the uniform convergence of the normalization constant in the revised manuscript. revision: yes

  2. Referee: [Change-point estimation procedure and its consistency proof] Consistency of the change-point location estimator relies on the test statistic behaving like a Brownian bridge under the null and diverging under the alternative. If the self-normalization fails to be uniform, the argmax of the normalized CUSUM may not be consistent; the paper should supply a explicit rate condition on the growth of the noise covariance eigenvalues relative to p and T that restores the required uniform convergence.

    Authors: We agree that an explicit growth-rate restriction improves transparency. The existing proof of consistency for the change-point estimator (Theorem 3.2) already relies on the uniform convergence implied by the mixing and moment conditions, which implicitly restrict the eigenvalue growth to o(T^{1/2-δ}) for some δ>0. In the revision we will state this allowable growth rate explicitly in terms of p and T immediately before the statement of the consistency result. revision: yes

Circularity Check

0 steps flagged

No load-bearing circularity; self-normalization presented as adaptation, not tautology

full rationale

The derivation chain proposes consistent estimators for pre- and post-change factor loadings plus change-point location, with self-normalization used to pivotalize the test statistic under strong cross-sectional dependence whose strength is treated as unknown. No quoted step shows a fitted parameter renamed as a prediction, a self-citation supplying the uniqueness or pivotal limit, or an ansatz smuggled from prior author work. The central claims rest on standard factor-model consistency arguments plus a self-normalized CUSUM device whose validity is asserted under stated moment/mixing conditions rather than by redefinition of the target quantities. This yields a normal non-circular finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the method implicitly relies on standard high-dimensional factor model assumptions and regularity conditions for self-normalized statistics that are not enumerated here.

pith-pipeline@v0.9.0 · 5637 in / 1256 out tokens · 26757 ms · 2026-05-24T17:52:57.771620+00:00 · methodology

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

30 extracted references · 30 canonical work pages · 2 internal anchors

  1. [1]

    Determining the number of factors in approximate factor models

    Bai, J. and Ng, S. (2002), “Determining the number of factors in approximate factor models.” Econometrica, 70, 191–221

    work page 2002

  2. [2]

    A unified approach to self-normalized block sam- pling,

    Bai, S., Taqqu, M. S., and Zhang, T. (2016), “A unified approach to self-normalized block sam- pling,” Stochastic Processes and their Applications , 126, 2465–2493

    work page 2016

  3. [3]

    Identification and estimation of a large factor model with structural instability

    Baltagi, B., Kao, C., and Wang, F. (2017), “Identification and estimation of a large factor model with structural instability.” Journal of Econometrics , 197, 87–100

    work page 2017

  4. [4]

    Simultaneous multiple change-point and factor analysis for high-dimensional time series

    Barigozzi, M., Cho, H., and Fryzlewicz, P. (2018), “Simultaneous multiple change-point and factor analysis for high-dimensional time series.” Journal of Econometrics , 206, 187–225

    work page 2018

  5. [5]

    Regularized estimation in sparse high-dimensional time series models,

    Basu, S. and Michailidis, G. (2015), “Regularized estimation in sparse high-dimensional time series models,” The Annals of Statistics , 43, 1535–1567

    work page 2015

  6. [6]

    Testing for structural breaks in dynamic factor models

    Breitung, J. and Eickmeier, S. (2011), “Testing for structural breaks in dynamic factor models.” Journal of Econometrics , 163, 71–84

    work page 2011

  7. [7]

    Arbitrage, factor structure and mean-variance analysis in large asset markets,

    Chamberlain, G. and Rothschild, M. (1983), “Arbitrage, factor structure and mean-variance analysis in large asset markets,” Econometrica, 70, 191–221

    work page 1983

  8. [8]

    High dimensional stochastic regression with latent factors, endogeneity and nonlinearity

    Chang, J., Guo, B., and Yao, Q. (2015), “High dimensional stochastic regression with latent factors, endogeneity and nonlinearity.” Journal of Econometrics , 189, 297–312

    work page 2015

  9. [9]

    Test for high-dimensional white noise using maximum cross correlations

    Chang, J., Yao, Q., and Zhou, W. (2017), “Test for high-dimensional white noise using maximum cross correlations.” Biometrica, 104, 1–17

    work page 2017

  10. [10]

    Estimating the common break date in large factor models

    Chen, L. (2015), “Estimating the common break date in large factor models.” Economics Letters, 131, 70–74

    work page 2015

  11. [11]

    Detecting big structural breaks in large facotr models,

    Chen, L., Dolado, J. J., and Gonzalo, J. (2014), “Detecting big structural breaks in large facotr models,” Journal of Econometrics , 180, 30–48

    work page 2014

  12. [12]

    The cost of the savings and load crisis: truth and consequences

    Curry, T. and Shibut, L. (2000), “The cost of the savings and load crisis: truth and consequences.” FDIC Banking Review. , 13, 26–35

    work page 2000

  13. [13]

    Sparse vector autoregressive modeling,

    Davis, R. A., Zang, P., and Zheng, T. (2016), “Sparse vector autoregressive modeling,” Journal of Computational and Graphical Statistics , 25, 1077–1096. 34

    work page 2016

  14. [14]

    A two-step estimator for large approximate dynamic factor models based on Kalman filtering

    Doz, C., Giannone, D., and Reichlin, L. (2011), “A two-step estimator for large approximate dynamic factor models based on Kalman filtering.” Journal of Econometrics , 164, 188–205

    work page 2011

  15. [15]

    and Yao, Q

    Fan, J. and Yao, Q. (2003), Nonlinear Time Series: Nonparametric and Parametric Methods ,

    work page 2003

  16. [16]

    The generalized dynamic factor model: one-sided estimation and forecasting

    Forni, M., Hallin, M., Lippi, M., and Reichlin, L. (2005), “The generalized dynamic factor model: one-sided estimation and forecasting.” Journal of the American Statistical Association , 100, 830–840

    work page 2005

  17. [17]

    The generalized dynamic factor model: consistency and rates

    Forni, M., Lippi, M., and Reichlin, L. (2004), “The generalized dynamic factor model: consistency and rates.” Journal of Econometrics , 119, 231–255

    work page 2004

  18. [18]

    Wild binary segmentation for multiple change-point detection,

    Fryzlewicz, P. (2014), “Wild binary segmentation for multiple change-point detection,” The An- nals of Statistics , 42, 2243–2281

    work page 2014

  19. [19]

    Tests for parameter instability in dynamic factor models

    Han, X. and Inoue, A. (2015), “Tests for parameter instability in dynamic factor models.” Econo- metric Theory, 31, 1117–1152

    work page 2015

  20. [20]

    Factor modeling for high-dimensional time series: inference for the number of factors,

    Lam, C. and Yao, Q. (2012), “Factor modeling for high-dimensional time series: inference for the number of factors,” Annals of Statistics , 40, 694–726

    work page 2012

  21. [21]

    Estimation of latent factors for high-dimensional time series,

    Lam, C., Yao, Q., and Bathia, N. (2011), “Estimation of latent factors for high-dimensional time series,” Biometrika, 98, 901–918

    work page 2011

  22. [22]

    Helping Effects Against Curse of Dimensionality in Threshold Factor Models for Matrix Time Series

    Liu, X. and Chen, E. (2019+), “Helping effects against curse of dimensionality in threshold factor models for matrix time series.” Manuscript, Available at arXiv: 1904.07383

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  23. [23]

    Threshold factor models for high-dimensional time series

    Liu, X. and Chen, R. (2016), “Regime-switching factor models for high-dimensional time series.” Statistica Sinica, 26, 1427–1451. — (2019), “Threshold factor models for high-dimensional time series.” Manuscript, Available at arXiv: 1809.03643

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  24. [24]

    Testing that a dependent process is uncorrelated,

    Lobato, I. N. (2001), “Testing that a dependent process is uncorrelated,” Journal of the American Statistical Association, 96, 1066–1076

    work page 2001

  25. [25]

    Estimation of large dimensional factor models with an unknown number of breaks,

    Ma, S. and Su, L. (2018), “Estimation of large dimensional factor models with an unknown number of breaks,” Journal of Econometrics , 207, 1–29. 35

    work page 2018

  26. [26]

    Inequalities for spreads of matrix sums and products

    Merikoski, J. K. and Kumar, R. (2004), “Inequalities for spreads of matrix sums and products.” Applied Mathematics E-Notes , 4, 150–159

    work page 2004

  27. [27]

    A self-normalized approach to confidence interval construction in time series,

    Shao, X. (2010), “A self-normalized approach to confidence interval construction in time series,” Journal of the Royal Statistical Society: Series B , 72, 343–366. — (2011), “A simple test of changes in mean in the possible presence of long-range dependence,” Journal of Time Series Analysis , 32, 598–606

    work page 2010

  28. [28]

    Diffusion indexes

    Stock, J. H. and Watson, M. W. (1998), “Diffusion indexes.” NBER Working Paper 6702 . — (2005), “Implications of dynamic factor models for VAR analysis.” National Bureau of Eco- nomic Research, Working Paper 11467

    work page 1998

  29. [29]

    A self-normalized semiparametric test to detect changes in the long memory parameter,

    Taqqu, M. S. and Zhang, T. (2019), “A self-normalized semiparametric test to detect changes in the long memory parameter,” Journal of Time Series Analysis , forthcoming. Van der Vaart, A. W. and Wellner, J. A. (1996), Weak Convergence and Empirical Processes , New York: Springer Verlag

    work page 2019

  30. [30]

    Factor Models for Matrix-Valued High-Dimensional Time Series,

    Wang, D., Liu, X., and Chen, R. (2019), “Factor Models for Matrix-Valued High-Dimensional Time Series,” Journal of Econometrics , 208, 231–248. 36

    work page 2019