Sorry to say, the standard errors reported for VAR coefficients got seriously out of whack as a result of some changes made to gretl's internal SVD code in late December/early January (and not sufficiently tested at the time). This problem was not present in the 2019d release, and it's now fixed in git and snapshots. We strongly recommend anyone using a recent snapshot, or building gretl from the git sources, and interested in estimating VARs to update. Allin Cottrell