Abstract
To discuss the reform of the medical insurance system, it is important to estimate lifetime medical costs. However, when using receipt data, there are three problems to estimate lifetime medical costs. First, since there is no data covering from birth to death, we use longitudinal data with a few years and simulate lifetime medical costs using some random variables. Second, many studies are not robust and parametric. They use type 2 Tobit model and bivariate normal random variables to simulate lifetime medical costs. Third, to avoid the parametric problem, Eichner et al. (1996, 2002) apply semiparametric random variables based on the resemble sample. However, this method is statistically inconsistent. To simulate lifetime medical costs, this paper proposes a new method. We generalize the type 2 Tobit model using Hermite polynomials. In simulation, we propose semiparametric random variables which are generated by Metropolis-Hastings algorithm.
