Hands-On Machine Learning for Algorithmic Trading

Metropolis-Hastings sampling

The Metropolis-Hastings algorithm randomly proposes new locations based on its current state to effectively explore the sample space and reduce the correlation of samples relative to Gibbs sampling. To ensure that it samples from the posterior, it evaluates the proposal using the product of prior and likelihood, which is proportional to the posterior. It accepts with a probability that depends on the result, which is relative to the corresponding value for the current sample.

A key benefit of the proposal evaluation method is that it works with a proportional evaluation rather than an exact evaluation of the posterior. However, it can take a long time to converge because the random movements that are not related ...