12.07.2015 Views

"Frontmatter". In: Analysis of Financial Time Series

"Frontmatter". In: Analysis of Financial Time Series

"Frontmatter". In: Analysis of Financial Time Series

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

424 MCMC METHODSht0 20 40 601970 1980 1990 2000yeargarch0 20 40 601970 1980 1990 2000Figure 10.5. <strong>Time</strong> plots <strong>of</strong> fitted volatilities for monthly log returns <strong>of</strong> S&P 500 index from1962 to 1999. The upper panel shows the posterior means <strong>of</strong> a Gibbs sampler with 5000iterations. The lower panel shows the results <strong>of</strong> a GARCH(1, 1) model.yearThe posterior mean <strong>of</strong> α 1 is 0.685, which is smaller than that obtained by Jacquier,Polson, and Rossi (1994) who used daily returns <strong>of</strong> the S&P 500 index. But it confirmsthe strong serial dependence in the volatility series. Finally, we have used differentinitial values and 3100 iterations for other Gibbs sampler, the posterior means<strong>of</strong> the parameters change slightly, but the series <strong>of</strong> posterior means <strong>of</strong> h t are stable.10.7.2 Multivariate Stochastic Volatility Models<strong>In</strong> this subsection, we study multivariate stochastic volatility models using theCholesky decomposition <strong>of</strong> Chapter 9. We focus on the bivariate case, but the methodsdiscussed also apply to the higher dimensional case. Based on the Choleskydecomposition, the innovation a t <strong>of</strong> a return series r t is transformed into b t such thatb 1t = a 1t , b 2t = a 2t − q 21,t b 1t ,where b 2t and q 21,t can be interpreted as the residual and least squares estimate <strong>of</strong>the linear regressiona 2t = q 21,t a 1t + b 2t .

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!