Abstract
We examine the sizes and powers of three tests of convergence of Markov Chain Monte Carlo draws: the Kolmogorov–Smirnov test, fluctuation test, and Geweke's test. We show that the sizes and powers are sensitive to the existence of autocorrelation in the draws. We propose a filtered test that is corrected for autocorrelation. We present a numerical illustration using the Federal funds rate.
Acknowledgment
The authors thank an anonymous referee whose comments greatly help improve the article.
Notes
1A proof for the asymptotic distribution of the FT is given in Ploberger et al. (Citation1989).
Notes: (1) Sample size is 3,000 (N = 3,000).
(2) Number of replications is 3,000.
Notes: (1) Sample size is 3,000 (N = 3,000).
(2) Number of replications is 3,000.
AR(1) correction is made by Eq. (Equation12)
.
2We may check stationarity of draws by a Bayesian procedure given in Goldman et al. (Citation2001).