In this study, we aim to test how effective are multiecuacionales models like auto regressive vectors (VAR) and models of systems of equations, for forecast by short term of crimes, applied to the Santiago in the period between January 1st of 2001 and June 30th of 2004. We show that bests forecast models are different, depending on the sector that is under study, the errors of daily series near 27%, similar to studies for other countries like England and the United States.