Demand System for SNAP Participants and Ex-Ante Simulation of Food Restrictions

I estimate an Exact Affine Stone Index (EASI) demand system for seven food and beverage categories using monthly purchase data from a sample of Supplemental Nutrition Assistance Program (SNAP) entrants. Entrants are identified by combining a household consumer panel of food purchases with a two-wave SNAP participation survey administered to the same households. The paper characterizes how household budget shares respond to the full vector of prices, real income, and program participation within a utility-theoretic framework. Household-level Fisher Ideal price indices are constructed from retail scanner data following Zhen et al. (2014), and the price index of the composite numéraire good is built from Bureau of Economic Analysis Regional Price Parities and Bureau of Labor Statistics monthly Consumer Price Indices in the manner of Zhen et al. (2024). I estimate the incomplete two-way approximate EASI of Zhen et al. (2014) (Model 1, prices and total expenditure assumed exogenous) by the Amemiya generalized least squares (AGLS) estimator of Amemiya (1979) and Newey (1987), which combines a single-equation censored Tobit first step with a minimum-distance second step that imposes Slutsky symmetry, homogeneity, and adding-up. The SNAP indicator enters the system through three channels: as an additive intercept shifter, as an interaction with all log prices, and as an interaction with real income, mirroring the treatment of the obesity indicator in Zhen et al. (2024). Standard errors for the elasticities are computed by parametric bootstrap (Krinsky-Robb simulation). This version presents the data construction, model, identification, and estimation framework. The full set of structural coefficient estimates, price and expenditure elasticities, and robustness checks are withheld from this conference version pending review by the data provider, and will be added to the paper upon approval.