A Nonlinear Dynamics Approach to Evaluating the 'Realism' of Food Systems Models

How can theoretical market models—which necessarily abstract from reality—satisfy demands for realism when used to support high-stakes food policy? Past work concludes that modelers can be reasonably required to demonstrate the ‘degree of correspondence’ between a model and reality, but leaves open the question of how to demonstrate correspondence. We suggest that correspondence be demonstrated by requiring modelers to produce persuasive empirical evidence of real-world market dynamics that their models skillfully reproduce. Real-world market dynamics are masked in volatile observed prices. Agricultural economists conventionally attribute price volatility to exogenous random shocks that can be modeled with linear stochastic approaches, but there is increasing recognition that price volatility also may be generated endogenously by nonlinear market dynamics. Selecting between these competing explanations for market instability matters in food policy because they present policymakers very different surrogate realities with divergent policy implications. We propose pre-modeling application of Nonlinear Time Series analysis to distinguish between linear and nonlinear dynamic structure in observed price data, and provide a framework guiding its sound application. Price data testing positive for nonlinear dynamic structure provides evidence that observed market volatility may be explained with parsimonious nonlinear specifications. Alternatively, price data testing negative for nonlinear dynamics provides evidence that linear stochastic approaches may better model observed volatility.