Robust estimation for discrete‐time state space models. (6th October 2020)

• Record Type: Journal Article
• Authors: Aeberhard, William H.; Cantoni, Eva; Field, Chris; Künsch, Hans R.; Mills Flemming, Joanna; Xu, Ximing
• Is Part Of: Scandinavian journal of statistics. Volume 48:Number 4(2021)
• Page Start: 1127
• Page End: 1147
• Publication Date: 2020-10-06
• Subjects: bounded influence function -- Fish stock assessment -- Laplace approximation -- random effects -- template model builder; Statistics -- Periodicals; 310
• DOI: 10.1111/sjos.12482
• Languages: English
• ISSNs: 0303-6898

Abstract

Abstract: State space models (SSMs) are now ubiquitous in many fields and increasingly complicated with observed and unobserved variables often interacting in nonlinear fashions. The crucial task of validating model assumptions thus becomes difficult, particularly since some assumptions are formulated about unobserved states and thus cannot be checked with data. Motivated by the complex SSMs used for the assessment of fish stocks, we introduce a robust estimation method for SSMs. We prove the Fisher consistency of our estimator and propose an implementation based on automatic differentiation and the Laplace approximation of integrals which yields fast computations. Simulation studies demonstrate that our robust procedure performs well both with and without deviations from model assumptions. Applying it to the stock assessment model for pollock in the North Sea highlights the ability of our procedure to identify years with atypical observations.