Functional disability in the Health and Retirement Study: a semi-Markov multi-state analysis

We propose a direct non-parametric estimation framework for multi-state disability models that does not rely on the Markov assumption. In this context, "direct" denotes a method that does not necessitate the initial estimation of transition intensities. Previous research has demonstrated evidence of non-Markovian properties in disability dynamics, and a number of studies have employed semi-Markov models for Long-Term-Care products. We test the non-Markov assumption on an up-to-date American dataset that includes more than 200,000 records, finding insights on the non-Markovian character of disability. We then focus on transition probabilities, showing the impact of our assumption on these crucial metrics for actuaries. In addition, we investigate the influence of health conditions on all the transitions analyzed.