Known methods for predicting one variable are generalized to the case of predicting a vector of endogenous variables in the multidimensional regression model. By using the information of all equations we get better predictions (in the sense of a fixed optimalization criterion) than by predicting the vector components one by one.