Question about VARMA forecast

Hi ozcanalp,

Sorry for the late reply.

The model that is passed to VARMAForecastOneStep should be a zero mean model. You can get the demeaned model easily by (following your code): varmaModel.getDemeanedModel(). Therefore, changing “instance = new VARMAForecastOneStep(X_T, varmaModel);” to “instance = new VARMAForecastOneStep(X_T, varmaModel.getDemeanedModel());”

should solve your problem. And finally add the mean back to get the forecast of the original VARMA model.

Please let me know if there is anything else I can help.

Best wishes, Ding Hao

• This reply was modified 2 years, 5 months ago by hao.ding.

Thank you very much Ding Hao,

I have one more question about “public VARFitting(MultivariateIntTimeTimeSeries mts,int p)”. How can I estimate p value by using akaike information criterion. For example if I used ARIMA model below code is ok for me. But how can I estimate p value for VARMA model by using akaike information criterion?

Best wishes, ozcanalp

double[] inputNumbers = new double[] { 2, 3, 4 };
int P_VALUE = 2;
int D_VALUE = 1;
int Q_VALUE = 3;

int p = P_VALUE;
int d = D_VALUE;
int q = Q_VALUE;

ARIMAModel arima = null;

double maxAic = 0;
double actualAic = 0;
int maxp = 10;
int maxd = 10;
int maxq = 10;

ConditionalSumOfSquares csos = new ConditionalSumOfSquares(inputTimeSeries, p, d, q);

for(int ii=0;ii<=3;ii++){
for(int jj=0;jj<=2;jj++){
for(int kk=0;kk<=3;kk++){

p = ii;
d = jj;
q = kk;

csos = new ConditionalSumOfSquares(inputTimeSeries, p, d, q);

actualAic = csos.AIC();

if(actualAic > maxAic){
maxAic = actualAic;

maxp = p;
maxd = d;
maxq = q;

csos = new ConditionalSumOfSquares(inputTimeSeries, p, d, q);
}
}
}
}
Console.Write(“\n”);
Console.Write(“maxp: ” + maxp + “,” + maxd + “,” + maxq);
Console.Write(“\n”);

arima = csos.getModel();
IntTimeTimeSeries xt = new SimpleTimeSeries(inputTimeSeries);

ARIMAForecast instance = new ARIMAForecast(xt, arima);

ARIMAForecast.Forecast frc = instance.next();
double next = frc.xHat();
double err = frc.var();

Hi ozcanalp,

for the AIC of VARMA model, please see p388 of the paper here.

Basically speaking AIC can be computed as the norm of covariance matrix of VARMA residuals plus a penalty of order p, number of variables n, and length T.

• This reply was modified 2 years, 5 months ago by hao.ding.

Thank you very much Ding Hao,

I want to perform pairwise Granger causality wald tests after VAR but I can’t find any function for The Granger causality test F statistics in the suanshu package.How can I determine the first series does not case the second series or the second series does not cause the first series by using Granger causality test. Can you help me ?

Best wishes,

ozcanalp

Hi ozcanalp,

For the Granger causality test, here is an interesting article. It describes procedure step by step. Hope this helps.

Thank you very much Ding Hao,

When I’ve fitted a model, i got an {“no solution to this system of linear equations”} exception on this line “VARFitting fitting = new VARFitting(X_T, 2);”. How can I fit the model? My code and exception are given below. any help or advise will be appreciated.

code:

X_T = new MultivariateSimpleTimeSeries( new double[][]{ new double[]{0, 0}, new double[]{0, 0.008}, new double[]{0, 0.269}, new double[]{0, 0.008}, new double[]{0, 0.009}, new double[]{0, 0.002}, new double[]{0, 0.005}, new double[]{0, 0.009}, new double[]{0.005, 0.008}, new double[]{0.006, 0.003}

});

try
{
VARFitting fitting = new VARFitting(X_T, 2);

VARMAModel varmaModel = fitting.getVARMA();
VARMAForecastOneStep instance = null;

instance = new VARMAForecastOneStep(X_T, varmaModel.getDemeanedModel());
int T = X_T.size();
Vector xTHat = instance.xHat(T + 1);
Matrix V = instance.covariance(T);
Console.Write(“Predictor at time ” + (T+1) + “: ” + xTHat + “, covariance of errors: ” + V);
Console.Write(“\n”);
}catch(Exception e){

Console.WriteLine(“\nStackTrace —\n{0}”, e.StackTrace);

}

Exception :

at com.numericalmethod.suanshu.algebra.linear.matrix.doubles.linearsystem.BackwardSubstitution.new(UpperTriangularMatrix , Vector , Double ) at com.numericalmethod.suanshu.algebra.linear.matrix.doubles.linearsystem.BackwardSubstitution.solve(UpperTriangularMatrix U, Vector b, Double epsilon) at com.numericalmethod.suanshu.algebra.linear.matrix.doubles.linearsystem.OLSSolverByQR.solve(LSProblem problem) at com.numericalmethod.suanshu.stats.regression.linear.ols.OLSRegression..ctor(LMProblem problem, Double epsilon) at com.numericalmethod.suanshu.stats.regression.linear.ols.OLSRegression..ctor(LMProblem problem) at com.numericalmethod.suanshu.stats.timeseries.linear.multivariate.stationaryprocess.arma.VARFitting.new(MultivariateIntTimeTimeSeries , Int32 ) at com.numericalmethod.suanshu.stats.timeseries.linear.multivariate.stationaryprocess.arma.VARFitting.true(MultivariateIntTimeTimeSeries , Int32 ) at com.numericalmethod.suanshu.stats.timeseries.linear.multivariate.stationaryprocess.arma.VARFitting..ctor(MultivariateIntTimeTimeSeries mts, Int32 p) at MultivariateTimeSeries.MultiLinearTimeSeries.run() in \Visual Studio 2010\Projects\MultivariateTimeSeries\MultivariateTimeSeries\MultiLinearTimeSeries.cs:line 323

Your data is ill-defined. For example, you have two time series, one of them is:0,0,0,0,0,0,0,0,0.005,0.006. It seems very odd to me. And it is very likely that it will cause numerical problems in computations.