The analysis of uncertainty in large-scale agricultural economic models has gained attention from a policymakers and researchers’ viewpoint. The different methodologies available vary depending on data availability and the nature of the variables subject to analysis, which in turn influences the outcomes. When evaluating the results of previously applied partial stochastic methodologies to partial equilibrium models, underperformance and generation of biases have been observed. This paper evaluates different stochastic methods for introducing yield and macroeconomic uncertainty in a large-scale agricultural economic model and proposes a new methodology for partial uncertainty analysis consisting of a combination of parametric and non-parametric estimators chosen to minimize the statistical prediction error and distributional assumptions. Results suggest that the best methodologies are those relaxing distributional assumptions and allowing for a better representation of historical variability. For uncertainty extraction, the cubic polynomial (for yields) and the multivariate vector auto-regression (for macroeconomic variables) methods perform best. For uncertainty simulation, tests favor semi-parametric methods against parametric approaches. These methods are applied to the ex-ante analysis of global agricultural commodity markets and supplement the traditional deterministic analysis with a statistical representation of stochastic uncertainty.