betaFeedback

WorksMathematics

Divide-and-Conquer is Tail- and Anomaly-Robust: Factor Model Estimation Without Moment Constraints

Abstract

We study robust estimation of elliptical factor models with abnormal observations and without requiring the existence of any positive moments. In particular, a divide-and-conquer (DC) estimator averages orthogonal projectors onto the full column spaces of small data blocks, and we rigorously establish its probability bound that allows weak and heterogeneous factors. The derived convergence rate is comparable to that of classical principal component estimators, up to the proportion of anomalies. Building on our DC estimator, we propose a split-sample tail- and anomaly-robust (STAR) estimator for the factor number using sample medians of successive projected coordinate ratios, which is shown consistent under mild conditions. Subsequent factor and common component estimation is also discussed, with the rates of convergence spelled out. Remarkably, given an upper bound on the true factor number, our suite of procedures are free from tuning, require no tail-dependent truncation threshold nor solving an optimization problem, admit parallel block computations, and hence has the same computational complexity as the classical principal component methods. Extensive experiments corroborate our theoretical findings, where DC and STAR are comparable and often outperform state-of-the-art competitors. We also illustrate our methods on a portfolio return dataset.

Introduction

High-dimensional factor models describe common variation in many observed variables through a small number of latent factors. They provide a framework for dimension reduction, forecasting, and the analysis of economic and financial panels. Principal component methods and their large-dimensional theory have played a central role in these developments [27, 3, 2, 5], where the primary focus lies in estimating the vector space spanned by the matrix of loadings, a.k.a. loading space. Another critical part of factor analysis is to determine the number of factors, and several approaches have been proposed which are based on information criteria or eigenvalue ratios [3, 1, 21]. While it is commonly assumed that the factors are pervasive across the cross-sections, there also exist studies for understanding how the strength of factors affects the rates of estimation [22, 4].

Two distinct stylized features of modern data complicate the use of conventional principal component methods. First, the data may have heavy tails as frequently observed in economics and finance [9], which invalidates the commonly made moment conditions underpinning covariance-based approaches. Second, some observations may be anomalous and arbitrarily deviate from the underlying model, due to the events such as financial crisis. When heavy tails and anomalies co-exist, extreme observations may be the realizations from the heavy-tailed distribution or anomalies and, regardless of their origin, negatively influence the accuracy of the existing principal component-based estimators [19] (He et al. 2022). In this paper, we develop methods for estimating the loading space and the number of factors under elliptical factor models while permitting the presence of heavy tails, anomalies, and weak factors.

Elliptical distributions are popularly considered for high-dimensional data analysis under heavy tails. For instance, Han and Liu [15] develop elliptical component analysis using multivariate Kendall’s tau, and Fan et al. [12] study robust covariance estimation through elliptical factor models. For factor modeling, methods based on spatial Kendall’s tau have been developed for estimating the loading space and the latent factor process (He et al., 2022) and the factor number (Yu et al., 2019). Other methods address heavy tails under different distributional assumptions, e.g., by adopting the Spearman correlation for factor number estimation under a scale-mixture structure [25], utilizing Huber loss [7] or data truncation [6]. These approaches offer different tradeoffs between distributional structure and moment requirements.

There also exists a closely related but distinct line of research which focuses on developing principal component methods that are robust against outliers. Such methods combine the projection pursuit with robust scatter estimation [19], and can accommodate sparse loadings [18], or cellwise and rowwise outliers [20]. It is worth pointing out that the procedures designed to control heavy-tailed fluctuations and those designed to be robust against the anomalies, address related but different objectives; see Raymaekers and Rousseau [26] for a comprehensive overview and further discussions.

In this paper, we propose the Divide-and-Conquer (DC) estimator for the loading space under elliptical factor models. Simply put, DC partitions the data into small blocks, computes the orthogonal projector onto the full column space of each block, and extracts the leading eigenspace of their average. When the factor number rr is known, we take the block size to be rr; if it is unknown, we set the block size to be a strict upper bound on rr. This sets the DC estimator apart from the existing tail-robust methods that often rely on selecting the tuning parameters that depend on the unknown tail index. The aggregation of local spectral information has an established role in distributed estimation [13, 14] (He et al. 2025) but in contrast to these earlier works, the DC estimator aggregates the blockwise orthogonal projectors of full column rank of the block size, even when the block size exceeds the number of factors. Our investigation reveals that this proposed approach allows for establishing a signal-to-noise eigengap under possibly weak and heterogeneous factor strengths, without making any moment constraint.

For factor number estimation, we propose the Split-sample Tail- and Anomaly-Robust (STAR) estimator. splitting the data into two, the STAR projects the second half of the data onto the DC estimator from the first half, takes the ratio of each pair of successive squared scores, and locates the maximizer of the sample median of these ratios over time. By its construction, the STAR estimator utilizes the ellipticity of the data and benefits from the eigengap preserved in the DC loading space estimator.

Theoretically, without imposing any finite-moment requirements while permitting the presence of anomalies, we establish a non-asymptotic probabilistic bound for the estimation error of the DC estimator. If factors are strong and the anomalies are absent, by setting the block size bb as the factor number (b=rb=r), the convergence rate of the DC estimator matches the rate attained by the classical principal component methods under the existence of the fourth moment in the data [3]. We also show the consistency of the STAR estimator with which the additional application of the DC estimator allows us to estimate the factors and common components, also without any moment constraints.

To summarize, Table 1 and Table 2 provide an overview of the methods most closely related to our study in comparison with DC and STAR estimators, respectively. For Barigozzi et al. [6], the displayed rate concerns the vector factor model under their general dependence conditions. Under their bounded deterministic factors and the remaining conditions of their loading estimation result, their rate would reduce to OP(1/T+1/p)O_{\mathrm{P}}(1/\sqrt{T}+1/p).

MethodMoment requirementModelConvergence rate
Bai [2]≥4\ge4General factor modelOP(1T+1p)O_P\left(\sqrt{\frac{1}{T}+\frac{1}{p}}\right)
Barigozzi et al. [6]2+2ϵ2+2\epsilon, ϵ>0\epsilon>0General factor modelOP({log⁡(Tp)T}ϵ/(1+ϵ)+1p)O_P\left(\left\{\frac{\log(Tp)}{T}\right\}^{\epsilon/(1+\epsilon)}+\frac{1}{p}\right)
Han and Liu [15]NoneElliptical factor modelOP(log⁡(p)T)O_P\left(\sqrt{\frac{\log(p)}{T}}\right)
He et al. (2022)NoneElliptical factor modelOP(1T+1p)O_P\left(\sqrt{\frac{1}{T}+\frac{1}{p}}\right)
Hubert et al. [19]UnspecifiedRobust PCAN/A
Hubert et al. [18]UnspecifiedRobust PCAN/A
Hubert et al. [20]UnspecifiedRobust PCAN/A
DCNoneElliptical factor modelOP(1T+1p)O_P\left(\sqrt{\frac{1}{T}+\frac{1}{p}}\right)

Table 1. Comparison of loading space and principal subspace estimators. The rates are derived for the distance between true loading matrix A\mathbf{A} and estimator A^\widehat{\mathbf{A}} measured by ∥A(ATA)−1AT−A^(A^TA^)−1A^T∥F\left\|\mathbf{A}(\mathbf{A}^{\mathsf{T}}\mathbf{A})^{-1}\mathbf{A}^{\mathsf{T}}-\widehat{\mathbf{A}}(\widehat{\mathbf{A}}^{\mathsf{T}}\widehat{\mathbf{A}})^{-1}\widehat{\mathbf{A}}^{\mathsf{T}}\right\|_{F}, under strong factors and in the absence of anomalies; the rate reported for DC is a special case of Theorem 1. Here, TT refers to the sample size and pp the dimension of the data.

MethodMoment requirementModel
Bai and Ng [3]≥4\ge4General factor model
Barigozzi et al. [6]2+2ϵ2+2\epsilon, ϵ>0\epsilon>0General factor model
Fan et al. [11]6+ϵ6+\epsilon, ϵ>0\epsilon>0Independent latent coordinates
Hubert et al. [19]UnspecifiedRobust PCA
Qiu et al. [25]NoneScale-mixture factor model
Yu et al. (2019)NoneElliptical factor model
STARNoneElliptical factor model

Table 2. Comparison of factor number estimators.

Notations. Unless specified otherwise, we denote scalars, vectors and matrices by plain letters, lower-case bold letters and capital bold letters, i.e., xx (or XX), x\mathbf{x} and X\mathbf{X}, respectively. We use xix_i and XijX_{ij} to denote the iith element of a vector x\mathbf{x} and the (i,j)(i,j)th element of a matrix X\mathbf{X}, respectively. We use a≍ba \asymp b (resp. a≍Pba \asymp_{\mathrm{P}} b) to denote a=O(b)a=O(b) and b=O(a)b=O(a) (resp. a=OP(b)a=O_{\mathrm{P}}(b) and b=OP(a)b=O_{\mathrm{P}}(a)). Hereafter, given a positive integer mm, define [m]:={1,2,…,m}[m]:=\{1,2,\ldots,m\}. The iith largest eigenvalue of a real symmetric matrix X\mathbf{X} is denoted by λi(X)\lambda_i(\mathbf{X}), and SVD⁡i(X)\operatorname{SVD}_i(\mathbf{X}) represents the leading ii eigenvectors of X\mathbf{X}, by which we always refer to the orthonormal eigenvectors of the matrix corresponding to the ii largest eigenvalues. For any symmetric matrices X1\mathbf{X}_1 and X2\mathbf{X}_2, the notation X1⪯X2\mathbf{X}_1 \preceq\mathbf{X}_2 means X2−X1\mathbf{X}_2-\mathbf{X}_1 is positive semi-definite.

Methodology

Model and problem formulation

Consider a sequence of pp-dimensional random vectors {xt0}t∈[T]\{x_t^0\}_{t\in[T]} admitting the factor model representation (e.g. Bai and Ng [3]; Bai [2]) such that

xt0=Aft+ϵt=:χt+ϵt,(1)x_t^0 = \mathbf{A}f_t+\epsilon_t=:\chi_t+\epsilon_t, \tag*{(1)}

where A∈Rp×r\mathbf{A} \in\mathbb{R}^{p \times r} is the factor loading matrix, ft∈Rr\mathbf{f}_t \in\mathbb{R}^r is the latent core factor, and ϵt∈Rp\boldsymbol{\epsilon}_t \in\mathbb{R}^p is the idiosyncratic noise. The factor-driven χt\boldsymbol{\chi}_t is referred to as the common component. The number of factors rr is fixed but unknown. Then, we observe {xt}t∈[T]\{\mathbf{x}_t\}_{t \in[T]}, where

xt={xt0,t∉T,zt,t∈T,(2)\mathbf{x}_t = \begin{cases} \mathbf{x}_t^0, & t \notin\mathcal{T},\\ \mathbf{z}_t, & t \in\mathcal{T}, \end{cases} \tag*{(2)}

for an unknown set T⊂[T]\mathcal{T} \subset[T] and arbitrary vectors zt∈Rp\mathbf{z}_t \in\mathbb{R}^p representing the anomalies. The index set of anomalies may be random or deterministic, see Assumption 4 below for the conditions on T\mathcal{T}. In this setup, it is of interest to estimate the loading space spanned by the columns of A\mathbf{A} and the factor number rr under (1), from the observations {xt}t∈[T]\{\mathbf{x}_t\}_{t \in[T]} which are potentially contaminated by anomalies. Further, even in the absence of anomalies, we permit {xt0}t∈[T]\{\mathbf{x}_t^0\}_{t \in[T]} to be heavy-tailed without any moment condition as specified below.

Following Fan et al. [12] and He et al. (2022), we formulate the model for xt0\mathbf{x}_t^0 through a joint elliptical distribution. We say a random vector y∈Rh\mathbf{y} \in\mathbb{R}^h follows an elliptical distribution, denoted by y∼EDh(μ,Σ0,ζ)\mathbf{y} \sim\mathrm{ED}_h(\boldsymbol{\mu}, \mathbf{\Sigma}_0, \zeta), if

y=dμ+ζA0u,(3)\mathbf{y} \overset{d}{=} \boldsymbol{\mu} + \zeta\mathbf{A}_0 \mathbf{u}, \tag*{(3)}

where μ∈Rh\boldsymbol{\mu} \in\mathbb{R}^h is deterministic, u∈Rq\mathbf{u} \in\mathbb{R}^q is a random vector uniformly distributed on the unit sphere Sq−1\mathbb{S}^{q-1} with q≤hq \le h, the radius ζ≥0\zeta\ge0 is a scalar random variable independent of u\mathbf{u}, A0∈Rh×q\mathbf{A}_0 \in\mathbb{R}^{h \times q} is a deterministic matrix with full column rank, and the matrix Σ0=A0A0T\mathbf{\Sigma}_0 = \mathbf{A}_0\mathbf{A}_0^{\mathsf{T}} is referred to as the scatter matrix of y\mathbf{y}. Such framework contains multivariate Gaussian distribution as a special case and, more importantly, includes a large class of heavy-tailed distributions such as the multivariate elliptical tt-distribution.

Note that the scatter matrix Σ0\mathbf{\Sigma}_0 is only defined up to a positive scalar factor. Namely, the distribution in (3) would remain unchanged if A0\mathbf{A}_0 and ζ\zeta are replaced by cA0c\mathbf{A}_0 and c−1ζc^{-1}\zeta, respectively. In view of this, in formulating the elliptical distribution of xt0\mathbf{x}_t^0, we adopt a representation where the block of the joint scatter matrix corresponding to the factor is Ir\mathbf{I}_r.

Assumption 1. For each t∈[T]t \in[T], we assume that ft\mathbf{f}_t and ϵt\boldsymbol{\epsilon}_t in (1) satisfy

\begin{pmatrix} \mathbf{f}_t\\ \boldsymbol{\epsilon}_t \end{pmatrix} = \zeta_t \begin{pmatrix} \mathbf{I}_r & \mathbf{0}\\ \mathbf{0} & \mathbf{A}_{\epsilon} \end{pmatrix} \frac{\mathbf{u}_t}{\lVert\mathbf{u}_t \rVert}, \tag*{(4)}\end{pmatrix}

where ζt\zeta_t and ut\mathbf{u}_t are independent copies of a scalar random variable ζ\zeta and a random vector u∼N(0,Ir+p)\mathbf{u} \sim\mathcal{N}(\mathbf{0}, \mathbf{I}_{r+p}), respectively. Also, ζ\zeta is independent of u\mathbf{u}, and is finite and strictly positive almost surely.

Assumption 1 implies that (ft⊤,ϵt⊤)⊤∼ED⁡r+p(0,diag⁡(Ir,Σϵ),ζ)(\mathbf{f}_t^\top,\boldsymbol{\epsilon}_t^\top)^\top\sim\operatorname{ED}_{r+p}(0,\operatorname{diag}(\mathbf{I}_r,\boldsymbol{\Sigma}_{\epsilon}),\zeta), where Σϵ:=AϵAϵ⊤∈Rp×p\boldsymbol{\Sigma}_{\epsilon}:=\mathbf{A}_{\epsilon}\mathbf{A}_{\epsilon}^{\top}\in\mathbb{R}^{p\times p}. Since linear transformations of elliptically distributed random vectors remain elliptical, the model in (1) further implies that xt0\mathbf{x}_t^0 is elliptically distributed itself. Specifically, let us define M:=[A,Aϵ]∈Rp×(r+p)\mathbf{M}:=[\mathbf{A},\mathbf{A}_{\epsilon}]\in\mathbb{R}^{p\times(r+p)} and vt:=ut/∥ut∥\mathbf{v}_t:=\mathbf{u}_t/\lVert\mathbf{u}_t\rVert for t∈[T]t\in[T]. Then xt0\mathbf{x}_t^0 is elliptically distributed and admits the representation

xt0=ζtMvt.(5)\mathbf{x}_t^0=\zeta_t\mathbf{M}\mathbf{v}_t. \tag*{(5)}

with its scatter matrix

References

  1. [1]Ahn, S. C. and Horenstein, A. R. (2013). Eigenvalue ratio test for the number of factors. Econometrica, 81(3):1203–1227.DOI
  2. [2]Bai, J. (2003). Inferential theory for factor models of large dimensions. Econometrica, 71(1):135–171.DOI
  3. [3]Bai, J. and Ng, S. (2002). Determining the number of factors in approximate factor models. Econometrica, 70(1):191–221.DOI
  4. [4]Bai, J. and Ng, S. (2023). Approximate factor models with weaker loadings. Journal of Econometrics, 235(2):1893–1916.arxiv.org/abs/2109.03773
  5. [5]Barigozzi, M. (2026). Principal component analysis for high-dimensional approximate factor models in time series: Assumptions, asymptotic theory, and identification. Journal of Time Series Analysis (in press).DOI
  6. [6]Barigozzi, M., Cho, H., and Maeng, H. (2026). Tail-robust factor modelling of vector and tensor time series in high dimensions. Biometrika, 113(2):asaf093.arxiv.org/abs/2407.09390
  7. [7]Barigozzi, M., He, Y., Li, L., and Trapani, L. (2023). Robust tensor factor analysis. arXiv preprint arXiv:2303.18163.arxiv.org/abs/2303.18163
  8. [8]Cen, Z. and Lam, C. (2025). Tensor time series imputation through tensor factor modelling. Journal of Econometrics, 249:105974.
  9. [9]Cont, R. (2021). Empirical properties of asset returns: stylized facts and statistical issues. Quantitative Finance, 1(2):223–236.
  10. [10]Dürre, A., Tyler, D. E., and Vogel, D. (2016). On the eigenvalues of the spatial sign covariance matrix in more than two dimensions. Statistics & Probability Letters, 111:80–85.
  11. [11]Fan, J., Guo, J., and Zheng, S. (2022). Estimating number of factors by adjusted eigenvalues thresholding. Journal of the American Statistical Association, 117(538):852–861.
  12. [12]Fan, J., Liu, H., and Wang, W. (2018). Large covariance estimation through elliptical factor models. The Annals of Statistics, 46(4):1383 – 1414.arxiv.org/abs/1507.08377
  13. [13]Fan, J., Wang, D., Wang, K., and Zhu, Z. (2019). Distributed estimation of principal eigenspaces. The Annals of Statistics, 47(6):3009 – 3031.arxiv.org/abs/1702.06488
  14. [14]Gao, Z. and Tsay, R. S. (2023). Divide-and-conquer: A distributed hierarchical factor approach to modeling large-scale time series data. Journal of the American Statistical Association, 118(544):2698–2711.arxiv.org/abs/2103.14626
  15. [15]Han, F. and Liu, H. (2018). ECA: High-Dimensional Elliptical Component Analysis in Non-Gaussian Distributions. Journal of the American Statistical Association, 113(521):252–268.
  16. [18]Hubert, M., Reynkens, T., Schmitt, E., and Verdonck, T. (2016). Sparse PCA for High-Dimensional Data With Outliers. Technometrics, 58(4):424–434.
  17. [19]Hubert, M., Rousseeuw, P. J., and Branden, K. V. (2005). ROBPCA: A New Approach to Robust Principal Component Analysis. Technometrics, 47(1):64–79.
  18. [20]Hubert, M., Rousseeuw, P. J., and den Bossche, W. V. (2019). MacroPCA: An All-in-One PCA Method Allowing for Missing Values as Well as Cellwise and Rowwise Outliers. Technometrics, 61(4):459–473.DOI
  19. [21]Lam, C. and Yao, Q. (2012). Factor modeling for high-dimensional time series: Inference for the number of factors. The Annals of Statistics, 40(2):694–726.DOI
  20. [22]Lam, C., Yao, Q., and Bathia, N. (2011). Estimation of latent factors for high-dimensional time series. Biometrika, 98(4):901–918.DOI
  21. [24]Marsaglia, G. (2006). Ratios of normal variables. Journal of Statistical Software, 16(4):1–10.DOI
  22. [25]Qiu, J., Li, Z., and Yao, J. (2025). Robust Estimation for Number of Factors in High Dimensional Factor Modeling via Spearman Correlation Matrix. Journal of the American Statistical Association, 120(550):1139–1151.arxiv.org/abs/2309.00870
  23. [26]Raymaekers, J. and Rousseeuw, P. J. (2026). Challenges of cellwise outliers. Econometrics and Statistics, 38:6–25.
  24. [27]Stock, J. H. and Watson, M. W. (2002). Forecasting using principal components from a large number of predictors. Journal of the American Statistical Association, 97(460):1167–1179.DOI
  25. [28]Taskinen, S., Koch, I., and Oja, H. (2012). Robustifying principal component analysis with spatial sign vectors. Statistics & Probability Letters, 82(4):765–774.DOI
  26. [29]Tropp, J. A. (2012). User-friendly tail bounds for sums of random matrices. Foundations of Computational Mathematics, 12:389–434.
  27. [30]Vershynin, R. (2018). High-dimensional probability: An introduction with applications in data science, volume 47. Cambridge university press.

Paper details

vlnc:W_fnk3bqzq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents