A new estimator is proposed for linear triangular systems, where identification results from the model errors following a bivariate and diagonal GARCH(1,1) process with potentially time-varying error covariances. This estimator applies when traditional instruments are unavailable. I demonstrate...