To my knowledge, it is only possible to 2-step estimation when you utilize the R packages. firstly estimate the VAR model, then use the residuals of the VAR in estimating the BEKK model. I believe the correct method is to run the two models as one system. in the system and the parameters must be estimated through optimization. In addition, the information matrix must be directly estimated and the significance of the estimated coefficients must be tested directly. This work is tedious and time-consuming