Markov Chain Monte Carlo (MCMC) techniques, in the context of Bayesian
inference, constitute a practical and effective tool to produce samples from an
arbitrary distribution. These algorithms are applied to calculate parameter
values of predictive models of the phenomenon of varying volatility in data
time series. For this purpose, 3 such research models of time-varying
volatility are simulated in STAN a probabilistic programming language for
statistical inference. The accuracy of these modelsí predictive function is
confirmed by applying in data time series with known prior values. Moreover,
Stan modelsí performance is illustrated by the real stock prices of two shares
in the stock market of New York. Finally, an Information Criterion of the
results is applied to each model as well, to evaluate their predictive ability,
comparing and selecting the most effective one.
Keywords: MCMC, Time-series, Time-varying volatility models, STAN.