Abstract: The continuous-time time-homogeneous Markov chain was used to construct a prediction model about the probability of insurance lapse or surrender to calculate the probability of being in various states at any time, and a parameter estimation method was given. In reality, the state of the insurance policy would have discrete events at a specific time, so the multi-stage Markov chain model was used to characterize this feature. That was, at a specific time when a discrete event occurs, a jump matrix was defined to describe the state transition at the specific time. The model was applied to the study of the life insurance lapse or surrender probability of insurance companies, and the model parameters were estimated and calibrated by actual data. Finally, the calibrated Markov chain model was used to predict the life insurance lapse or surrender probability and a good prediction result was obtained.

Key words: Markov chains, intensity matrix, transition probability, life insurance lapse or surrender probability prediction