Gradient-based filtering under misspecification: Stability and error bounds

Simon Donker van Heel, Rutger-Jan Lange, Bram van Os, and Dick van Dijk · 2025

Read Gradient-based filtering under misspecification: Stability and error bounds by Simon Donker van Heel, Rutger-Jan Lange, Bram van Os, and Dick van Dijk (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-05021.txt.

Introduction

Simon Donker van Heel and co-authors prove stability results for gradient-based filters in a paper from 2025. Such filters track moving parameters, like a shifting risk level, from noisy data. Each step nudges the estimate along the slope of a chosen objective. Score-driven filters, built on the log density, are the best-known kind.

The catch is misspecification: the model is never quite the true data process. The authors derive conditions that keep the filter path stable anyway. Two forms, explicit and implicit, each get their own guarantees. Stability holds regardless of the true process, a strong form of safety. Extra moment conditions then yield error bounds. Forecasters in finance and engineering gain firmer ground.

Catalog record

CREATORSimon Donker van Heel, Rutger-Jan Lange, Bram van Os, and Dick van Dijk
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE189 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-05021

Browse the full library · About the Tower of Babel