Simulating Uncorrelated Samples Using Markov Chains Monte Carlo Through Gpu for Stochastic Volatility Models Estimation

Stochastic volatility models are very flexible models able to characterize financial volatility evolution. This article explores computational capabilities based on Graphical Processing Units to simulate many Monte Carlo Markov chains in estimating stochastic volatility model parameters through Bayesian methods. We simulate a massive number of parallel chains to approximate the posterior distributions. Sophisticated algorithms were developed to mitigate high autocorrelation levels in simulated Markov chains. However, some of the complex algorithms developed are error-prone for code implementation. Simple algorithms coupled with the increased computational power associated with parallel computing can produce more reliable estimates obtained through the final observations of the parallel chains, inducing uncorrelated samples with increased estimation efficiency