Economic effects of agricultural policies estimated with deterministic models may be biased because certainty equivalent requirements are not satisfied in the agricultural sector. This article compares numerically estimated expected values of key aggregate economic variables for two levels of target prices (i.e. supported price level) from deterministic and stochastic formulations of AGSIM, which is a large-scale econometric simulation model of crop and livestock production in the United States. For nominal target prices set by the 1990 Food, Agriculture, Conservation and Trade Bill (FACT), the deterministic formulation underestimated the level of deficiency payments (the difference between target price and market price) by a cumulative total of $800 million (6·2%) over the 5 year life of the Bill compared to estimates from the stochastic model. For target prices 10% below the FACT levels, the deterministic model underestimated deficiency payments by $1220 million (36·8%). Much of the economic impact estimation bias can be attributed to price received by participating producers under FACT having a probability distribution truncated at the target price, which is a strong departure from the certainty equivalent requirements needed to have mean outcomes from stochastic models equal outcomes from their deterministic counterpart. Estimated changes in deficiency payments from lowering target prices by 10% were overestimated by $420 million (4·3%) with the deterministic model, which suggests that deterministic models are less biased for estimating the change in deficiency payments than in estimating the level of payments. The bias associated with mean values of other variables, such as crop prices and farm income, was much less in percentage terms than bias in estimates of deficiency payments.