1
$\begingroup$

Is there a model/technique that is able to estimate transition matrix (which would be consistent, i.e. sums of their rows would be always 1) conditional on some continuous variable X?

Let's say I have a system that can acquire one of 3 possible states (A,B,C). I have some observations and I could apply standard Markov chains. This way I can get something we could call "unconditional" transition matrix.

However, the transition probabilities (probably) depend on a continuous variable $X\in(0,1)$. The higher X is, the higher is the chance of A going to B and B going to C (C is absorption state).

$\endgroup$
1
$\begingroup$

One way to represent these is a Markov switching model. This page has some useful links.

These model the probability of being in a particular state on various distributions, e.g. on a continuous variable X∈(0,1). R has a nice package named MSwM for this.

There is also the concept of Hidden Markov Models where the conditional transition probabilities are quantified. However, as the name suggests, these only apply where you have 'latent' (i.e. hidden) states - which does not seem to be your case.

$\endgroup$
0
$\begingroup$

This is a time-inhomogeneous Markov Jump Chain

Transition matrix,$$P_{ij}(t)=\pmatrix{1-p(t)&p(t)&0\\0&1-p(t)&p(t)\\0&0&1}$$

As $X \in (0,1)$ , so for every $X(t)$ you'll get $p(t)$ so that you can make up.

$\endgroup$
  • 1
    $\begingroup$ This answer has many problems. The question asks how to estimate this matrix (presumably using data). Your formulation doesn't meet the constraints on the transition probabilities, either. And your "$X$" does not appear to mean the same thing as the "$X$" in the question. $\endgroup$ – whuber Apr 29 '15 at 16:23
  • $\begingroup$ @whuber As he said X belongs (0,1) , so for every X(t) he'll get p(t) so that he can make up. $\endgroup$ – Hemant Rupani Apr 29 '15 at 19:01
  • 1
    $\begingroup$ That's not anywhere close to what your notation says! $\endgroup$ – whuber Apr 29 '15 at 19:03
  • $\begingroup$ @whuber thanks! But I am wonder what my notations said? $\endgroup$ – Hemant Rupani Apr 29 '15 at 19:09
  • $\begingroup$ "$p(t)\sim X(t)$", at least in a statistical context, would suggest that $X(t)$ is a family of distributions indexed by $t$, $p(t)$ is a family of random variables indexed by $t$, and that each $p(t)$ follows the distribution $X(t)$. $\endgroup$ – whuber Apr 29 '15 at 19:12

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.