Vector Autoregression with Python Statsmodels

Question

I am trying to implement multidimensional Granger causality in python. For that matter I am using Vector Autoregression from Statsmodels, but when I try to get coeffcients out of it, it returns me an empty matrix. Can somebody tell me what is wrong exactly?

import numpy as np
from statsmodels.tsa.vector_ar import var_model
def multi_dim_granger(X_ts,Y_ts,order=5,test='F-test'):
    """Multivariate Granger cusality.
    input:
        X_ts: the first vector time series. 
              TxK matrix with T being the time instance and K is the dimension  
        Y_ts: the second vector time series. 
              TxK matrix with T being the time instance and K is the dimension  

        order: the maximum number of lags for fitting a VAR process
        test: the statistical test to check for the residual covariance matrix
    """
    ts=np.hstack((X,Y))
    print ts.shape
    VAR_model=var_model.VAR(ts)
    ts=VAR_model.fit(ic='aic',maxlags=order)
    return ts.coefs
X=np.random.randn(1000,2)
Y=(np.arange(4000)*np.random.randn(4000)).reshape((1000,4))
multi_dim_granger(X,Y)

Answer 1

You can use the test_causality method of the VARResults instance to test Granger causality. See the documentation here and the example here.