The RATS Software Forum

SWARCH.RPF is an example of a Markov switching ARCH model. MS-ARCH was proposed by Hamilton and Susmel(1994), and independently, Cai(1994) as an alternative to a (standard) GARCH model. In these models, instead of the lagged variance term providing the strong connection for volatility from one period to the next, a Markov model governs switches between several variance regimes. Markov Switching GARCH models are much more technically demanding and results with them tend to be suspect.

Detailed Description

Hi Tom

Would you kindly show me how to modify so that the program can also do both the mean equation and arch parameter switches?
Similarly on hs_swarch_gl22.rpf for Hamilton and Susmel(1994), "Autoregressive Conditional Heteroskedasticity and Changes in Regime"

I am planning to add regressors to the mean equation. I understand that it could be over-parameterized, but I will only try it for two states.
Thanks!

Ratskdup wrote:Hi Tom

Would you kindly show me how to modify so that the program can also do both the mean equation and arch parameter switches?
Similarly on hs_swarch_gl22.rpf for Hamilton and Susmel(1994), "Autoregressive Conditional Heteroskedasticity and Changes in Regime"

I am planning to add regressors to the mean equation. I understand that it could be over-parameterized, but I will only try it for two states.
Thanks!

Extra regressors isn't hard if they are fixed. It is much harder if they switch. For instance, if the mean switches, the likelihood for the Cai model will then depend upon lagged regimes (since the lagged u's will be regime-dependent) which makes it more similar to the Hamilton-Susmel setup. The Hamilton-Susmel model isn't altered quite as much, but needs the UU's and U's to be computed and saved for each regime.

Have you found a published paper that has done that? I would think it's unlikely that it would end up being a useful extension.

TomDoan wrote:

Ratskdup wrote:Hi Tom

Would you kindly show me how to modify so that the program can also do both the mean equation and arch parameter switches?
Similarly on hs_swarch_gl22.rpf for Hamilton and Susmel(1994), "Autoregressive Conditional Heteroskedasticity and Changes in Regime"

I am planning to add regressors to the mean equation. I understand that it could be over-parameterized, but I will only try it for two states.
Thanks!

Extra regressors isn't hard if they are fixed. It is much harder if they switch. For instance, if the mean switches, the likelihood for the Cai model will then depend upon lagged regimes (since the lagged u's will be regime-dependent) which makes it more similar to the Hamilton-Susmel setup. The Hamilton-Susmel model isn't altered quite as much, but needs the UU's and U's to be computed and saved for each regime.

Have you found a published paper that has done that? I would think it's unlikely that it would end up being a useful extension.

In Maheu J, McCurdy TH. 2000. Identifying bull and bear markets in stock returns. Journal of Business and Economics Statistics 18:100-112, one of their models, DDMS-ARCH model that allows for both mean-switching and MSARCH effect