Central banks usually use non-structural and semi-structural models to forecast inflation. The Disaggregated Prediction System is a set of non-structural Sparse VAR models designed to forecast CPI and GDP growth. Even though the models are parsimonious and hence give rise to precise forecasts (Barrera, 2005), the parameter estimates are sensitive to the presence of outliers. The paper lays out a robust statistical multi-equation procedure for Sparse VAR models and quantifies the precision gaisn with a sample that includes a sequence of outliers. The results point out that robust Sparse VAR models improve the forecast of CPI in the medium term.