The RATS Software Forum

bekkdcc wrote:Dear Tom,

Thanks for your help. It seems perfect now, but I have some question for the last time

1. I corrected the set dlogp = log(ith) but what is the abbreviation BTW ?

"By the way".

bekkdcc wrote: 2. There is a code to narrow down the sample range and analyze it, but I couldn't remember what was the code beginning with .....:

calendar(m) 2010
data(org=obs,format=xls) 2010:01 2018:12 usd ith
*
*
............ 2011:04 2015:09

The problem is the estimation range, not the data range.

bekkdcc wrote: 4. phi seems statistically insignificant, is that mean SW model not appropriate or something else?

You don't want to see that.

bekkdcc wrote: 3. You can't start your estimation with entry 2, as you're losing entries due to the lags. Where is the entry 2 and how can change it?

Thanks in advance

dlm(method=bfgs,sw=sw,sv=varx2,y=ysq,type=filter,c=1.0, $
sx0=sw/(1-phi^2),x0=gammax,a=phi,z=gammax*(1-phi)) 2 * states

The start range on the DLM (both of them) can't be 2 (colored) as you've lost data points due to lags in the mean model. It looks like it should be 5.

As with a GARCH model, the first assumption with an SV model is that the series being modeled (your residuals) are serially uncorrelated. Are they? It doesn't look like your model would be able to do that.

Dear Tom,

I have some question from RATS Handbook for ARCH/GARCH and volatility models page 264 table 12.3.

I can not be sure which variable means what in the equation that I am sending you as an attached doc. format.

I will be so plaeased if you checked and corrected it if there is an error.

Thanks very much,

h(t) is the variance of the y process---if you want to use an alternative way to write that, it would be sigma^2_t. It's not different from the variance, just a different notation for it. (There's a reason h is used in the volatility literature---it's cleaner notation than writing sigma^2 for the variance).

GAMMAX is the (gamma*), not (gamma*)(1-phi). The latter is the gamma, which (as it says) can be hard to estimate when phi is near one. gamma* is not the log of the (unconditional) mean of h, it's the mean of the log of h which aren't the same things.

SW is the variance of w.