I’m working with historical credit performance data and would like to build a transition matrix to predict defaults and delinquencies. I can model the transition between states (ie current - delinquent) based on some covariates using a logit or other technique. Can I assemble a transition matrix simply from the predicted state-to-state probabilities or is it more complicated?

  • $\begingroup$ And why aren’t standard methods sufficient? E.g. stat.cmu.edu/~cshalizi/462/lectures/06/markov-mle.pdf $\endgroup$ Jun 27, 2019 at 16:44
  • 1
    $\begingroup$ openscholarship.wustl.edu/cgi/… Describes building a conditional Markov chain with multinomial logistic regression to set the transition probabilities. I’m thinking of doing the same thing, but modeling state transitions with machine learning techniques - im hoping to understand what can go wrong with this approach. Thanks for the feedback! $\endgroup$ Jun 27, 2019 at 18:02


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.