MichaelWeylandt
summary
Asserts hidden Markov models are practical in Stan by implementing the calculable log-likelihood via the forward algorithm and sampling the posterior with MCMC, not relying on Kalman tricks. Demonstrates a switching GARCH model for stock returns where the “bad” state shows roughly threefold volatility and the state probabilities are softer than committee recession dates. Diagnoses sampling trouble from multimodality and weak identifiability and recommends strong priors or semi‑supervised hints.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:07Hidden Markov models in Stan
- 3:21Understanding hidden Markov model structure
- 5:10Bayesian inference challenges
- 8:02Forward algorithm
- 10:06Finance two-state HMM
- 12:59Bayesian vs non-Bayesian models
- 13:42Closing remarks
- 14:17Q&A: Normality assumptions
- 15:22Q&A: Exact inference
- 16:26Q&A: Multi-modality issues
related talks
28mStan: why does it exist? when is it useful? why do I use it?
01h 06mNonparametric Bayesian Methods: Models, Algorithms, and Applications I
44mIdentification and Estimation of Dynamic Structural Models with Unobserved Choices
34mFast, beautiful and easy Bayesian modelling: Can you have it all?
01h 27mBayesian Time Series
01h 04mNonparametric Bayesian Methods: Models, Algorithms, and Applications II
30mProbabilistic modelling with Scala
23mDynamic inference in probabilistic graphical models