Skip to content
All library documents

Using Hidden Markov Models to Group Time-Series Regimes

Article Quant Q&A · Author: Danish A. Alvi
Read the original

Summary

The question asks how to reduce a continuous time series to a smaller set of recurring pattern states, then model transitions among those states as a Markov chain. This is a regime-detection problem: observations are treated as outcomes associated with underlying states, and the state sequence is used to describe changing behavior over time.

The response points to a hidden Markov model as a relevant method. In this setup, regimes are latent rather than directly observed, and the model represents probabilistic transitions between them alongside the observations generated in each regime. The document provides no worked example, estimation procedure, or trading results, and it does not discuss selecting the number of regimes, choosing observation features, or validating stability. It therefore identifies a useful modeling direction but leaves the practical design and evaluation to the researcher.

Key ideas

  • A continuous series can be represented through a smaller collection of recurring pattern states.
  • A Markov chain can describe the probabilities of moving between states.
  • A hidden Markov model is suggested for detecting regimes that are not directly observed.
  • The brief response gives no guidance on regime count, estimation, or out-of-sample validation.

Tags

Machine learningStatistics
Full text
# How can I 'quantize' a time-series in 'groups' exhibiting similar patterns?


# How can I 'quantize' a time-series in 'groups' exhibiting similar patterns?












In Signal processing, there is a topic of 'Quantization' (the process of mapping input values from a large set to output values in a (countable) smaller set ('states') ). I would like to construct a Markov Chain by relating the states these different 'states' interact with each other and the probability of these states coming about.

## Answer by Lipton (score 1)

https://quant.stackexchange.com/a/37964

Regime detection with hidden Markov model: http://scikit-learn.sourceforge.net/stable/modules/hmm.html

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.