The RATS Software Forum

Hi,

I´m estimating the dynamic evolution of stock market integration between China, Japan and South Korea from 2002 to 2012. I have been using mostly eviews and stata during the past years, but since I do not have them now, I have gone learning by doing with the Rats.

I have estimaded dynamic conditional correlation for the whole period, but I ended up with a problem of basic kind.

I need to check if the trilateral summits (external shock) held between China, Japan and South Korea between 2008 and 2012 had any impact on dynamic conditional correlation. So it will play role as a dummy (DM) variable in following linear equation:

ρijt = ω+ aDM1t+ ∑ β + η

According to Otto et al. (2001) the conditional correlation of stock market returns must lie between -1 and 1, it cannot be normally distributers, so transformation of dependent variable ρij,t has to be done:

Wij,t = ln(1+ρij,t/1-ρij,t)

The problem is that I just can not find the correct codes for all this this (regression & transformation). I googled and found user´s guide 7, but did not find the right procedure for this.

If there´s any one (Dr. Thomas Doan, etc.) who could help me with this, I would be really grateful!

BR,

Simo Koskelo


(Otto, G., Voss, G., Willard, L.(2001). Understanding OECD Output Correlations. Research Discussion Paper 05. Reserve Bank of Australia.)

I'm assuming that ρijt is, at this point, "data". The inverse logistic transformation that you're doing can be done with something like

set wij = log((1+rhoij)/(1-rhoij))

where rhoij is the dynamic correlation series (whatever you actually call it) and wij is the resulting transformed series. (You need to be very careful about the parentheses since / is done before + or -).

The dummy can be created with something like

set summit = t>=2008:1.and.t<=2012:12

if you have monthly data and you want the dummy to cover all of 2008 through 2012. If you have daily or weekly data, it would be

set summit = t>=2008:1:1.and.t<=2012:12:31

in either case with adjustments if you want a slightly different period.

The regression would then be done with

linreg wij
# constant summit other-explanatory-variables