This paper develops a heterogeneous-agent model to assess the impacts of removing lucerne growing subsidies, increasing livestock numbers and including district-level equilibrium conditions on optimal farm plans in the Qingyang district of Gansu Province, China. The model is a five-year dynamic linear program that solves across 96 farm households whilst incorporating district-level constraints. The approach used allows us to observe seasonal variations in incomes, infer the distribution of a policy shock among households and highlight trade patterns at the district level. The results suggest that without lucerne growing subsidies the total area of lucerne grown by all modelled households falls by 18%. Increasing livestock numbers by 25% reduces net household incomes by 17% as changes to labour allocations reduce off-farm employment opportunities. When external trade in forages is included in the model, total livestock numbers held by all 96 households rise from 502 to 838, this highlights the benefits of integrated feed markets. Shadow prices for crop production rise when livestock numbers increase, implying that benefits exist to improving crop yields.