Data and Code: "Risk sharing tests and covariate shocks"

Full risk-sharing implies that marginal utilities of expenditure (MUEs) have a simple factor structure; Pareto weights are divided by a common price. Thus full risk-sharing can be easily tested using panel data with two-way fixed effects. But MUEs must be inferred using data on expenditures. The standard approach assumes that all demands have unit price elasticities. This works well when shocks are idiosyncratic, affecting budgets without changing prices. But "covariate" shocks change prices, and risk-sharing tests which assume that no demands are inelastic will deliver apparently perverse results.
We obtain the larger class of nonhomothetic preferences that allows one to test risk-sharing using expenditure data. Demands are semi-parametric and nest the usual specification. We estimate demands and MUEs using Ugandan data, and show that our risk-sharing tests of covariate shocks deliver sensible results while the standard tests do not.
Households resident in Uganda, sampled by the Uganda National    Panel Survey under the World Bank LSMS program.