Likelihood-Based Estimates of Household Consumption Insurance

We propose a panel unobserved components model of household income and consumption. The model allows both idiosyncratic income and consumption to have permanent and transitory components, with possible cointegration and spillovers between components. We use likelihood-based methods for inference and find that (quasi) maximum likelihood estimation for a simple version of the model provides more precise and robust estimates of household consumption insurance against permanent income shocks than generalized method of moments for a widely-used annual panel dataset. Monte Carlo analysis demonstrates that (quasi) maximum likelihood estimation produces more accurate inferences than generalized method of moments in finite samples. Bayesian estimation allows straightforward comparison of different model specifications and favors the simple version of the model that has no cointegration or persistence in the transitory components. In subgroup analysis, we find that consumption insurance is higher for older and more educated households.