This paper applies a combined methodology of a recently developed directed acyclic graph (DAG) analysis with Johansen and Juselius' methods of the cointegrated vector autoregression (VAR) model to a monthly U.S. system of markets for soybeans, soy meal, and soy oil. Primarily a methods paper, Johansen and Juselius' procedures are applied, with a special focus on statistically addressing information inherent in well-known sources of non-normal data behavior to illustrate the effectiveness of modeling the system as a cointegrated multi-market system. Perhaps for the first time, methods of the cointegrated VAR model are combined with DAG analysis to account for contemporaneously correlated residuals, and are applied to this U.S. soy-based system. Analysis of the error correction or cointegration space illuminates the empirical nature of policy-relevant market elasticities, price transmission parameters, and effects of important policy and institutional changes/events on U.S. soy-related markets at long-run horizons beyond a single crop cycle. A statistically strong U.S. demand for soybeans emerged as the primary cointegrating relation in the error-correction space. Analysis of the DAG-adjusted cointegrated VAR model's forecast error variance decomposition illuminates how the soy-related variables and the three U.S. soy product markets dynamically interact at alternative time horizons extending up to two-years.