Discretization of bivariate processes

I have a question for you. Imagine that you have two exogenous and stochastic variables, say TFP and real exchange rate. You estimate a VAR(1) process for them and want to discretize them in order to add them in a model. How would you do that? One obvious idea is to stack the two exogenous variable into one and use the logic of Tauchen-86 to build the transition matrix between states. Do you know of anybody having done that? Other alternatives?