MS-EGARCH query

[kishmat]

Hi,
I was wondering if there's any example or paper replication codes for MS-EGARCH model and 'MS-EGARCH involving determinant variable' in 'mean and variance equations' of the exogenous variable. The similar model is applied in Ólan T. Henry's paper, 'Regime switching in the relationship between equity returns and short-term interest rates in the UK' Journal of Banking and Finance. I have tried to modify the codes used in GrayGarch (MS-GARCH) available here. Yet,I haven't got any conclusive codes. Can I get some help regarding these codes, pls.

[TomDoan]

You'd have to show me what you tried to do. Note, however, that the Gray MS-GARCH approximation tends not to work as well as Dueker's version.

[kishmat]

Hi,
Here I have attached the file that i tried to construct MS EGARCH (following MS GARCH procedure in GrayGarch replication codes provided herein). Though I tried the my best, I am sure that there's much to correct. As there is no replication codes available for MS EGARCH, please help me complete the attached codes.

[TomDoan]

You might need to contact Prof. Henry to find out how he adapted Gray's algorithm to the E-GARCH. It looks like it might not go through quite as easily as for a standard GARCH for the same reason that EGARCH variances can't be forecast recursively.

[kishmat]

Thanks Tom.
In the mean time ,, I got confused that my estimated function converges in 1 Iteration.. Is this function a valid one. Can I interpret the results of this function?


MAXIMIZE - Estimation by BFGS
Convergence in 1 Iterations. Final criterion was 0.0000000 <= 0.0000100
Weekly Data From 2001:01:10 To 2014:06:11
Usable Observations 701
Function Value 1555.4932

Thank you!!

[TomDoan]

If you do too many simplex iterations (50's quite a few), then it may be converged before BFGS takes over.

Note that you have b01 in both of these, which I don't think is what you want:

compute hh1=exp(b01+b11*(abs(u(t-1)/h(t-1))-0.7977)+b21*log(h(t-1))+b31*u1(t-1)/h(t-1))
compute hh2=exp(b01+b12*(abs(u(t-1)/h(t-1))-0.7977)+b22*log(h(t-1))+b32*u2(t-1)/h(t-1))

Another question is whether you really intend to have the means switching between regimes. That's a major complication, plus a potentially major confusion in the regime switch. (Which is the dominant "switch"---the mean or the variance equation?)