Alternative method for discretizing general state Markov processes


The QuantEcon libraries already have the tauchen and rouwenhorst methods for discretizing AR(1) processes, reducing them to finite Markov chains. Here’s a more general approach, that could be used for discretizing other state processes.

I think it would be a good addition to both libraries (Python lib and Julia lib)

Estimate MarkovChain from data using ML

This looks very cool and potentially helpful for some research problems I’m about to tackle.

Here’s a link to their Matlab code:

(though it is unlicensed so perhaps best not to look at it unless we get permission)


I’ve discussed this with Alexis Toda and he’s given permission to port his Matlab code to Julia and Python, and to disseminate it under an open source licence of our choosing.

If anyone wants to work on the conversion please let me know (or submit a PR to corresponding GitHub repo).


marginally more up to date papers and appendix from the journal website (paper is forthcoming):


If no one has started writing the code for this method, can I send a PR regarding this method in a couple of days?

Although I wrote a code focusing on evenly-spaced grid case, it has been pending because I haven’t attained robustness.

I think it’s a good time to send it after fixing the issue since @cc7768 sent a PR for estimation of transition matrix.


That would be great! You can compare your output against the Matlab code listed above.