Vector Auto-Regression, a multivariate time-series model / method. Under VAR, each univariate time-series is a linear combination of its own previous values and the previous values of the other series.
Vector Auto-Regression generalizes time-series methods by allowing multiple time series to be analyzed together. Using standard (i.e., Box-Jenkins) time-series methods, a series can be modeled such that a current value is, in part, a function of previous values (called lags). Vector auto-regression allows a series to also be partly a function of the lags of a second (or more) series, and the other series to be partly a function of the lags of the first series, simultaneously.
Important uses of VAR are forecasting and testing for Granger causality.