Next ->: Transforming Independent Variables
A call to setCVIndEqs sets up a list of independent variables to be included
in a conditional variance equation for a particular run.
library fanpac;
session amzn 'inCV model';
setDataSet stocks;
setSeries AMZN;
setIndVars AMZNvol;
setIndEqs none;
setCVIndEqs AMZNvol;
estimate run2 garchv(1,1);
causes FANPAC to include amzn1 as an independent variable
in the conditional variance equation for run2. The call to
setIndEqs with argument "none" removes AMZNvol from the
mean equation.
For multivariate models, setCVIndEqs sets lists of independent variables
for each conditional variance.
The call to setCVIndEqs causes a FANPAC global
_fan_CVIndEquations to be created. This global
is a matrix with rows equal to the number of conditional variances
and covariances in the model and columns
equal to the number of independent variables declared in the call
to setIndVars. The elements are zeros and ones, where ones indicate
a coefficient to be estimated.
To declare different sets of independent variables for different
dependent variables in the multivariate model, the dependent variable
name is included as the first argument in the call to setCVIndEqs.
For example,
library fanpac;
session mult 'multivariate example';
setDataSet stocks;
setSeries msft cpwr;
setIndVars msftVol AMZNvol;
setIndEqs none;
setCVIndEqs amzn AMZNvol;
setCVIndEqs msft MSFTvol;
estimate run1 dvgarchv(1,1);
For this model there will be two conditional variances and one conditional
covariance. Thus _fan_CVIndEquations will have three rows. There are
two independent variables, and one of them, AMZNvol, is specified
above to be in the AMZN conditional variance equation but not in
the MSFT conditional variance equation. Correspondingly, MSFTvol
is in the MSFT conditional variance equation but not the AMZN
conditional variance equation. However, any independent variable specified
to be in a conditional variance equation is also specified automatically
to be in any conditional covariance equations that it is associated
with. Thus in the above case, there will be coefficients computed for
both AMZNvol and MSFTvol in the conditional covariance equation.
The matrix in _fan_CVIndEquations will look like this
print _fan_CVIndEquations;
1 0
1 1
0 1
This will not be an issue in constant correlation models, however. Thus
for
library fanpac;
session mult 'multivariate example';
setDataSet stocks;
setSeries msft cpwr;
setIndVars msftVol AMZNvol;
setIndEqs none;
setCVIndEqs amzn AMZNvol;
setCVIndEqs msft MSFTvol;
estimate run1 cdvgarchv(1,1);
we have
print _fan_CVIndEquations;
1 0
0 1