Forecasting out-of-sample discrete_markov_chain.ipynb

No, I don’t think so.

But be careful, because the “statespace” packages you usually see (ours, and the one in statsmodels.tsa) is talking about linear gaussian state space models, which a lot of models don’t fall into.

HMMs are not linear, so you can’t represent those in the LGSS framework. Certain Holt-Winter exponential smoothing also can’t be expressed as LGSS. Non-gaussian models also don’t fall under the LGSS umbrella, so GARCH models are also out.

But in general, “state space” is so broad that yeah, pretty much everything falls under it’s umbrella.

1 Like