New techniques for the analysis of stochastic volatility models in whi
ch the logarithm of conditional variance follows an autoregressive mod
el are developed. A cyclic Metropolis algorithm is used to construct a
Markov-chain simulation tool. Simulations from this Markov chain conv
erge in distribution to draws from the posterior distribution enabling
exact finite-sample inference. The exact solution to the filtering/sm
oothing problem of inferring about the unobserved variance states is a
by-product of our Markov-chain method. In addition, multistep-ahead p
redictive densities can be constructed that reflect both inherent mode
l variability and parameter uncertainty. We illustrate our method by a
nalyzing both daily and weekly data on stock returns and exchange rate
s. Sampling experiments are conducted to compare the performance of Ba
yes estimators to method of moments and quasi-maximum likelihood estim
ators proposed in the literature. In both parameters estimation and fi
ltering, the Bayes estimators outperform these other approaches.