Robust estimation is considered in the context of a factor model for tensor-valued sequences. The consistency of the estimated common factors and loadings space is studied when using estimators based on minimising quadratic loss functions. Building on the observation that such loss functions are adequate only if sufficiently many moments exist, the results are extended to the case of heavy-tailed distributions by considering estimators based on minimising the Huber loss function, which uses an L1-norm weight on outliers. Such a class of estimators is shown to be robust to the presence of heavy tails, even when only the (2 + ε)-th moments of the data exist. A modified version of the eigenvalue-ratio principle is also proposed to estimate the dimensions of the core tensor, and the consistency of the resultant estimators is shown without any condition on the relative rates of divergence of the sample size and dimensions. Extensive numerical studies are conducted to show the advantages of the proposed methods over the state- of-the-art ones, especially under the heavy-tailed cases. An import/export dataset of a variety of commodities across multiple countries is analysed to show the practical usefulness of the proposed robust estimation procedure. Software implementing the proposed methods is available; identifying details are omitted for anonymous review.
Barigozzi, M., He, Y., Li, L., Trapani, L. (2026). Robust tensor factor analysis. ECONOMETRICS AND STATISTICS, 1, 1-22 [10.1016/j.ecosta.2026.10.002].
Robust tensor factor analysis
Barigozzi, Matteo;
2026
Abstract
Robust estimation is considered in the context of a factor model for tensor-valued sequences. The consistency of the estimated common factors and loadings space is studied when using estimators based on minimising quadratic loss functions. Building on the observation that such loss functions are adequate only if sufficiently many moments exist, the results are extended to the case of heavy-tailed distributions by considering estimators based on minimising the Huber loss function, which uses an L1-norm weight on outliers. Such a class of estimators is shown to be robust to the presence of heavy tails, even when only the (2 + ε)-th moments of the data exist. A modified version of the eigenvalue-ratio principle is also proposed to estimate the dimensions of the core tensor, and the consistency of the resultant estimators is shown without any condition on the relative rates of divergence of the sample size and dimensions. Extensive numerical studies are conducted to show the advantages of the proposed methods over the state- of-the-art ones, especially under the heavy-tailed cases. An import/export dataset of a variety of commodities across multiple countries is analysed to show the practical usefulness of the proposed robust estimation procedure. Software implementing the proposed methods is available; identifying details are omitted for anonymous review.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



