Using data from the Health and Retirement Survey and the Assets and Health Dynamics of the Oldest Old survey, we estimate the stochastic process that determines both the distribution and dynamics of health care costs. We find that the data-generating process for log health costs is well represented as the sum of a white noise process and a highly persistent AR(1) process. We also find that the innovations to this process can be modelled with a normal distribution that has been adjusted to capture the risk of catastrophic health care costs. Simulating this model, we find that in any given year 0.1% of households receive a health cost shock with a present value of at least $125,000.
