In a VAR model, you can test for heteroskedasticity using the Breusch-Pagan test or White's test. These tests can be implemented in STATA using the command "bptest" or "whitetest", respectively.

If the tests indicate that heteroskedasticity is present, you can adjust the standard errors to make them robust using a heteroskedasticity-and-autocorrelation-consistent (HAC) estimator such as the Newey-West estimator. You can implement the Newey-West estimator in STATA using the "vce(robust)" option in your regression command.

It's worth noting that if you have serial autocorrelation in your residuals, a HAC estimator will not correct for this issue. To address both heteroskedasticity and serial autocorrelation, you may consider using a dynamic heteroskedasticity and autocorrelation consistent (HAC) VAR model, which allows for the estimation of autocorrelated and heteroskedastic errors in a VAR model.

It's always a good idea to also check the residuals plots (e.g., residuals versus fitted values, residuals autocorrelation plot) to ensure that the HAC estimator has successfully addressed the issues of heteroskedasticity and serial autocorrelation.

My answer is coming late but it might be of use to the wide academic community. You need to first estimate the VAR model with the suggested lag length from any of the selection criteria. Secondly, test the model for serial/autocorrelation. If the model fails the test, increase the lag length by one and repeat the entire process until the condition is resolved.