Realistic Specifications and Model Predictability: Testing the Performance of a Stochastic CGE Model with Regionally Correlated Yield Variability in the Wheat and Rice Sectors

Although the computable general equilibrium (CGE) model has gained immense popularity, the trustworthiness of CGE results are sometimes questioned. A number of modelers have attempted to make the models more “realistic” by using various methods, yet the effectiveness of such modeling efforts has rarely been checked. Over the past two decades, stochastic CGE models were developed with, however, random shocks being generated following the independent and identically distributed (i.i.d.) normal distributions. In other words, correlations of agricultural productivity shocks between regions were ignored in spite of that such correlations are statistically observed in the real world. This article identifies the replicability of standard CGE models with regard to producer price volatilities of wheat and rice with regionally correlated random productivity shocks. We find that incorporating regional correlations improves predictability for wheat, while doing so for rice does not remarkably indicate amelioration, due to the limited tradability on the international rice market.