Two methods for building vector autoregression forecasting models are proposed. The first allows exclusion of intermediate lags; the second considers •the effects of jointly entering lags from different series into an equation. Live hog market models are developed and out-of-sample forecasting results suggest both methods have merit.