Forecasting ability of theory-constrained two-stage demand systems

Demand models are commonly used to forecast effects of policy changes and two-stage demand systems are frequently used when disaggregated food items are involved. A two-stage system implies interconnections between the stages. These interconnections can be modelled to make unconditional forecasts, or the second stage can be modelled separately to make conditional forecasts. We compare conditional and unconditional elasticity-based and direct statistical forecasts. For our data, conditional forecasts are superior to unconditional forecasts and forecasts derived from elasticities are superior to direct statistical forecasts. Imposition of the homogeneity and symmetry restrictions of consumer theory does not improve the forecasts.