Volatility Estimation of Hidden Markov Processes and Adaptive Filtration

TL;DR

This paper develops kernel-based estimators for volatility in partially observed Gaussian systems, enabling adaptive filtering and parameter estimation through a one-step MLE process.

Contribution

It introduces a novel nonparametric estimator for volatility and constructs an adaptive Kalman-Bucy filter using a Fisher-score based one-step MLE.

Findings

Effective kernel estimators for quadratic variation

Construction of adaptive Kalman-Bucy filter

Implementation of one-step MLE process

Abstract

The partially observed linear Gaussian system of stochastic differential equations with low noise in observations is considered. A kernel-type estimators are used for estimation of the quadratic variation of the derivative of the limit of the observed process. Then this estimator is used for nonparametric estimation of the integral of the square of volatility of unobservable component. This estimator is also used for construction of substitution estimators in the case where the drift in observable component and the volatility of the state component depend on some unknown parameter. Then this substitution estimator and Fisher-score device allows us to introduce the One-step MLE-process and adaptive Kalman-Bucy filter.