Markov-Switching Regression and error message

[Roksana]

Hello

I use daily stock return data and I got the below error message when I run regression model with Markov-switching.

I am very keen to know if I could find a solution to this problem.

The error is:
Dependent Variable:
Method: Switching Regression (Markov Switching)
Date: 03/16/15 Time: 12:40
Sample (adjusted): 1/01/1990 7/29/2013
Included observations: 6151 after adjustments
Number of states: 2
Initial probabilities obtained from ergodic solution
Ordinary standard errors & covariance using numeric Hessian
Random search: 250 starting values with 50 iterations using 1 standard
deviation (rng=kn, seed=1476496574)
Convergence achieved after 1 iteration
WARNING: Singular covariance - coefficients are not unique

Variable Coefficient Std. Error z-Statistic Prob.

Regime 1

C -0.000264 NA NA NA
RMRF 0.940327 NA NA NA
EVENTDAYS -0.002640 NA NA NA
LOG(SIGMA) -4.105172 NA NA NA

Regime 2

C 6.77E-05 NA NA NA
RMRF 0.649317 NA NA NA
EVENTDAYS 0.000365 NA NA NA
LOG(SIGMA) -4.950007 NA NA NA

Transition Matrix Parameters

P11-C 2.630129 NA NA NA
P21-C -3.030979 NA NA NA

Mean dependent var 1.48E-05 S.D. dependent var 0.025065
S.E. of regression 0.012123 Sum squared resid 0.902852
Durbin-Watson stat 1.708586 Log likelihood 19099.61
Akaike info criterion -6.206996 Schwarz criterion -6.196063
Hannan-Quinn criter. -6.203204


Regards

[trubador]

You can:
1) Clear the "Seed" and rerun the model (Options\Starting values)
2) Try different initial regime probabilities (Options\Switching options)
3) Try the alternative "Information matrix" (Options\Coefficent covariance matrix)
4) Check "Regime specific error variances"
5) Try higher number of regimes

If all these fail, then you should revise your model specification or consider using another method other than a MS model.