1: \begin{abstract}

2: The autoregressive (AR) model is a widely used model to understand

3: time series data. Traditionally, the innovation noise of the AR is

4: modeled as Gaussian. However, many time series applications, for example,

5: financial time series data, are non-Gaussian, therefore, the AR model

6: with more general heavy-tailed innovations is preferred. Another issue

7: that frequently occurs in time series is missing values, due to system

8: data record failure or unexpected data loss. Although there are numerous

9: works about Gaussian AR time series with missing values, as far as

10: we know, there does not exist any work addressing the issue of missing

11: data for the heavy-tailed AR model. In this paper, we consider this

12: issue for the first time, and propose an efficient framework for parameter

13: estimation from incomplete heavy-tailed time series based on a stochastic

14: approximation expectation maximization (SAEM) coupled with a Markov

15: Chain Monte Carlo (MCMC) procedure. The proposed algorithm is computationally

16: cheap and easy to implement. The convergence of the proposed algorithm

17: to a stationary point of the observed data likelihood is rigorously

18: proved. Extensive simulations and real datasets analyses demonstrate

19: the efficacy of the proposed framework. 

20: \end{abstract}