Nearest comoment estimation with unobserved factors

Kris Boudt, Dries Cornilly*, Tim Verdonck

*Corresponding author for this work

Research output: Contribution to JournalArticleAcademicpeer-review

285 Downloads (Pure)

Abstract

We propose a minimum distance estimator for the higher-order comoments of a multivariate distribution exhibiting a lower dimensional latent factor structure. We derive the influence function of the proposed estimator and prove its consistency and asymptotic normality. The simulation study confirms the large gains in accuracy compared to the traditional sample comoments. The empirical usefulness of the novel framework is shown in applications to portfolio allocation under non-Gaussian objective functions and to the extraction of factor loadings in a dataset with mental ability scores.

Original languageEnglish
Pages (from-to)381-397
JournalJournal of Econometrics
Volume217
Issue number2
Early online date7 Jan 2020
DOIs
Publication statusPublished - Aug 2020

Funding

We thank the Editor (Jeroen Rombouts), two anonymous referees and seminar participants at CREST, ETH Zürich and Vrije Universiteit Amsterdam for their valuable comments. We also benefited from fruitful discussions with participants at the CMStatistics, JSM and R/Finance conferences. We gratefully acknowledge support from the Research Foundation — Flanders ( FWO (Belgium) research grant G 023815 N and PhD fellowship 1114119 N ) and the Internal Funds KU Leuven, Belgium (project C 16 ∕ 15 ∕ 068 ). The computational resources and services used in this work were provided by the VSC (Flemish Supercomputer Center), funded by the FWO (Belgium) and the Flemish Government — Department EWI, Belgium . Appendix A

FundersFunder number
Flemish Government — Department EWI
Vlaams Supercomputer Centrum
Fonds Wetenschappelijk Onderzoek1114119N, G023815N
KU LeuvenC16∕15∕068

    Keywords

    • Higher-order multivariate moments
    • Latent factor model
    • Minimum distance estimation
    • Risk assessment
    • Structural equation modelling

    Fingerprint

    Dive into the research topics of 'Nearest comoment estimation with unobserved factors'. Together they form a unique fingerprint.

    Cite this