Abstract:We consider parameter estimation in finite hidden state space Markov models with time-dependent inhomogeneous noise, where the inhomogeneity vanishes sufficiently fast. Based on the concept of asymptotic mean stationary processes we prove that the maximum likelihood and a quasi-maximum likelihood estimator (QMLE) are strongly consistent. The computation of the QMLE ignores the inhomogeneity, hence, is much simpler and robust. The theory is motivated by an example from biophysics and applied to a Poisson- and linear Gaussian model.

What problem does this paper attempt to address?

This paper mainly discusses the maximum likelihood estimation problem in Hidden Markov Models (HMMs) with non-uniform noise. Traditional HMM assumes that the noise is homogeneous, but in practical applications, such as ion channel data in biophysical experiments, the noise may vary over time. The paper introduces Doubly Hidden Markov Models (DHMMs) to describe this phenomenon, where the observed sequence (Zn) gradually reduces the noise that changes over time. The main goal of the paper is to prove the consistency of maximum likelihood estimation (MLE) and quasi-maximum likelihood estimation (QMLE) in DHMMs. The authors propose a theoretical framework that utilizes the concept of asymptotic mean stationary processes and prove that under specific conditions, QMLE remains strongly consistent even in the presence of time-varying noise. This means that as the observed data increases, QMLE can approach the true model parameters more and more. Additionally, if additional conditions are satisfied, MLE is also consistent. Specifically, the paper analyzes the variation of ion channel currents in biophysical experiments, where the noise caused by voltage changes over time is non-uniform. By establishing a DHMM model, the paper demonstrates how to handle this non-uniform noise and proves that in certain cases, QMLE can be an effective method for estimating the true parameters. In summary, this paper addresses the statistical method problem of parameter estimation in Hidden Markov Models under non-uniform noise, providing a theoretical basis for analyzing such complex data.

Maximum likelihood estimation in hidden Markov models with inhomogeneous noise

Maximum Likelihood Estimation for Mixed Fractional Vasicek Processes

Electron spin resonance investigation of Mn^{2+} ions and their dynamics in manganese doped SrTiO_3

Maximum-penalized-likelihood estimation for independent and Markov-dependent mixture models.

Consistency of the maximum likelihood estimator in hidden Markov models with trends

Maximum Entropy Estimator for Hidden Markov Models: Reduction to Dimension 2

On Maximum a Posteriori Estimation of Hidden Markov Processes

Estimation for Piecewise Homogeneous Markov Jump Systems: A Dual Hidden Markov Model Approach

Near Universal Consistency of the Maximum Pseudolikelihood Estimator for Discrete Models

The Expectation-Maximization Algorithm for Continuous-time Hidden Markov Models

Maximum likelihood estimation of hidden Markov models for continuous longitudinal data with missing responses and dropout

Estimation in a class of nonlinear heteroscedastic time series models

Weak Consistency of Quasi-Maximum Likelihood Estimate in Generalized Linear Models

Quasi-Maximum Likelihood Estimation of long-memory linear processes

Beyond Time-Homogeneity for Continuous-Time Multistate Markov Models

Bayesian Quantile Nonhomogeneous Hidden Markov Models.

Asymptotic properties of the maximum likelihood estimator for Hidden Markov Models indexed by binary trees

Interacting Particle Langevin Algorithm for Maximum Marginal Likelihood Estimation

Maximum likelihood estimation in the non-ergodic fractional Vasicek model

Weak consistency and asymptotic normality of quasi-maximum likelihood estimates in generalized linear models

Nonparametric estimation of a mixing distribution for a family of linear stochastic dynamical systems