Bayesian Estimation Of Non-Gaussian Stochastic Volatility Models

After decades of rapid economic development in our country, at the same time of rapid economic growth, is showing some problems, such as significant volatility in financial markets. In the financial markets, asset pricing, risk measurement and portfolio management is an important part of its composition, depends on the volatility of asset. Volatility in assets were previously assumed to be a simple constant is not zero, it is clear that the volatility will do so simple assumption is unreasonable. Garch model assumptions variance with the previous disturbances and the early conditional variance has a specific function, but the SV model proposed conditional variance can introduce random disturbance, which is more in line with the dynamic characteristics of financial time series.In the practical application of SV model, the process is the most complex part of the model parameter estimation. Since assumptions with respect to the Gaussian distribution, financial time series generally exhibit significant characteristic peak thick tail, and volatility and convergence of data sequences are generally correlated, so the basic SV model of the RMB exchange rate of market estimated results unsatisfactory. Taking into account the exchange rate of foreign currencies against the RMB continuous, time-varying, especially its “bad shrink, enlarge good” asymmetric characteristics, binding model fitting test results, we take a different stochastic volatility models based on different data characteristics.We take Bayesian estimation and Markov Chain Monte Carlo simulation method for the basic SV model and SVt model Bayesian analysis, according to the posterior distribution of the density of the computed model parameters to calculate its Gibbs sampling and MCMC. By operating results of Winbugs and Eviews softwares to calculate the estimated value of the two models contained in various parameters. Finally, after analysis comparing the estimates obtained, finding it suitable to describe the case of the US dollar against the RMB exchange rate fluctuations model, and thus made different from the non-Gaussian stochastic volatility model of the present study ordinary Gaussian distribution. When the observed value of financial time series has obvious fat tail and sudden changes, this model is more applicable than normal Gaussian stochastic volatility models.