Advanced

Learning hidden Markov models with persistent states by penalizing jumps

Nystrup, Peter LU ; Lindström, Erik LU and Madsen, Henrik (2020) In Expert Systems with Applications 150.
Abstract

Hidden Markov models are applied in many expert and intelligent systems to detect an underlying sequence of persistent states. When the model is misspecified or misestimated, however, it often leads to unrealistically rapid switching dynamics. To address this issue, we propose a novel estimation approach based on clustering temporal features while penalizing jumps. We compare the approach to spectral clustering and the standard approach of maximizing the likelihood function in an extensive simulation study and an application to financial data. The advantages of the proposed jump estimator include that it learns the hidden state sequence and model parameters simultaneously and faster while providing control over the transition rate, it... (More)

Hidden Markov models are applied in many expert and intelligent systems to detect an underlying sequence of persistent states. When the model is misspecified or misestimated, however, it often leads to unrealistically rapid switching dynamics. To address this issue, we propose a novel estimation approach based on clustering temporal features while penalizing jumps. We compare the approach to spectral clustering and the standard approach of maximizing the likelihood function in an extensive simulation study and an application to financial data. The advantages of the proposed jump estimator include that it learns the hidden state sequence and model parameters simultaneously and faster while providing control over the transition rate, it is less sensitive to initialization, it performs better when the number of states increases, and it is robust to misspecified conditional distributions. The value of estimating the true persistence of the state process is illustrated through a simple trading strategy where improved estimates result in much lower transaction costs. Robustness is particularly critical when the model is part of a system used in production. Therefore, our proposed estimator significantly improves the potential for using hidden Markov models in practical applications.

(Less)
Please use this url to cite or link to this publication:
author
; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Clustering, Dynamic programming, Regime switching, Regularization, Time series analysis, Unsupervised learning
in
Expert Systems with Applications
volume
150
article number
113307
publisher
Elsevier
external identifiers
  • scopus:85079852674
ISSN
0957-4174
DOI
10.1016/j.eswa.2020.113307
language
English
LU publication?
yes
id
5165e9dd-24e6-4893-bdb8-ea79f5cba8b0
date added to LUP
2020-03-09 13:27:00
date last changed
2020-12-29 03:22:08
@article{5165e9dd-24e6-4893-bdb8-ea79f5cba8b0,
  abstract     = {<p>Hidden Markov models are applied in many expert and intelligent systems to detect an underlying sequence of persistent states. When the model is misspecified or misestimated, however, it often leads to unrealistically rapid switching dynamics. To address this issue, we propose a novel estimation approach based on clustering temporal features while penalizing jumps. We compare the approach to spectral clustering and the standard approach of maximizing the likelihood function in an extensive simulation study and an application to financial data. The advantages of the proposed jump estimator include that it learns the hidden state sequence and model parameters simultaneously and faster while providing control over the transition rate, it is less sensitive to initialization, it performs better when the number of states increases, and it is robust to misspecified conditional distributions. The value of estimating the true persistence of the state process is illustrated through a simple trading strategy where improved estimates result in much lower transaction costs. Robustness is particularly critical when the model is part of a system used in production. Therefore, our proposed estimator significantly improves the potential for using hidden Markov models in practical applications.</p>},
  author       = {Nystrup, Peter and Lindström, Erik and Madsen, Henrik},
  issn         = {0957-4174},
  language     = {eng},
  publisher    = {Elsevier},
  series       = {Expert Systems with Applications},
  title        = {Learning hidden Markov models with persistent states by penalizing jumps},
  url          = {http://dx.doi.org/10.1016/j.eswa.2020.113307},
  doi          = {10.1016/j.eswa.2020.113307},
  volume       = {150},
  year         = {2020},
}