About the residuals of the variance quations

[xyzh]

Dear Sir,
I use the following codes to estimate a VAR-GARCH-DCC model, how can i do the specification test such as serial correlation or Heteroscedasticity of residual from the vaiance equations?

Thanks alot .

Code: Select all

    *
    * Multivariate GARCH with two-step DCC estimator
    *
    open data fulldata.xls
    data(format=xls,org=columns) / r1 r2 r2
    *
    system(model=var1)
    variables r1 r2 r3
    lags 1
    det constant
    end(system)
    *
    garch(p=1,q=1,mv=dcc,model=var1,variance=varma,method=bhhh)

[TomDoan]

You have to save the residuals (rvector option) and the covariance matrices (hmatrices option). Good examples of this are in the Tsay textbook examples, in particular, tsayp449.prg, tsayp452.prg, tsayp460.prg. This is from tsayp460.prg. The fact that it's mv=diag rather than dcc doesn't affect the post-processing. It computes the standardized residuals, then tests for serial correlation (multivariate) and for serial correlation in the squares.

Code: Select all

garch(p=1,q=1,mv=diag,model=bimean,hmatrices=hh,rvector=rr)
set stdhk = rr(t)(1)/sqrt(hh(t)(1,1))
set stdja = rr(t)(2)/sqrt(hh(t)(2,2))
@mvqstat(lags=4)
# stdhk stdja
@mvqstat(lags=8)
# stdhk stdja
set stdhksq = stdhk**2
set stdjasq = stdja**2
@mvqstat(lags=4)
# stdhksq stdjasq
@mvqstat(lags=8)
# stdhksq stdjasq

[xyzh]

I found it out just after my post, still thank you for the immediate reply.

Zhang