Markov Switch: Fixing the Regime Position

[FelipeP]

Hi,
Is it possible to fix the regime position, or impose a restriction, in a 2-regime Markov Switch? Re-estimating the model doesn't guarantee the regimes 1 and 2 will have the same order.

I would like to test an out-of-sample trading strategy based on the regime probabilities of the model. However, when I write a program to estimate a model on each date it is not guaranteed my regime 1 probabilities will be from within the same regime 1 as before. Basically, I need the regime probabilities on each date of the rolling-window sample the Markov model is estimated.

If it is not possible, maybe a way to compare the regime probability with the previously estimated would work. If they deviate a lot (which should not be the case with 1 extra data point), save the regimes on the inverse order.

I have already searched the forum and also tried to hold the seed constant, but I have had no success, yet.


Thank you in advance for your inputs on this issue,
Felipe

[EViews Glenn]

There is no way since the model is completely unidentified with respect to position.

The closest you can do is to use the starting values to push the regimes into natural positions. In your case, with the additional single observation, it should be enough to use the previously estimated results as starting values. Alternately, you could estimate, examine the results, and then decide whether to reorder the coefficients and use those as starting values.

Note: it should be possible to set the seed for random starting values. From the command line use seed= and rnd=

http://www.eviews.com/help/helpintro.ht ... chreg.html

Or just set the values in the dialog options page.

[FelipeP]

Thank you for you quick and clear reply Glenn.
I've managed to identify the regime using an IF to compare coefficients. Since one regime has at least one clear coefficients that is higher than the other through specifications and sample used, this will work just fine for now.