Computational Statistics

, Volume 14, Issue 3, pp 375–395

Adaptive proposal distribution for random walk Metropolis algorithm

  • Heikki Haario
  • Eero Saksman
  • Johanna Tamminen

DOI: 10.1007/s001800050022

Cite this article as:
Haario, H., Saksman, E. & Tamminen, J. Computational Statistics (1999) 14: 375. doi:10.1007/s001800050022

Summary

The choice of a suitable MCMC method and further the choice of a proposal distribution is known to be crucial for the convergence of the Markov chain. However, in many cases the choice of an effective proposal distribution is difficult. As a remedy we suggest a method called Adaptive Proposal (AP). Although the stationary distribution of the AP algorithm is slightly biased, it appears to provide an efficient tool for, e.g., reasonably low dimensional problems, as typically encountered in non-linear regression problems in natural sciences. As a realistic example we include a successful application of the AP algorithm in parameter estimation for the satellite instrument ‘GOMOS’. In this paper we also present systematic performance criteria for comparing Adaptive Proposal algorithm with more traditional Metropolis algorithms.

Key words: MCMC, Adaptive MCMC, Metropolis-Hastings algorithm, convergence. 

Copyright information

© Physica-Verlag, Heidelberg 1999

Authors and Affiliations

  • Heikki Haario
    • 1
  • Eero Saksman
    • 1
  • Johanna Tamminen
    • 2
  1. 1.Department of Mathematics, P.O.Box 4 (Yliopistonkatu 5), FIN-00014 University of Helsinki, FinlandFI
  2. 2.Finnish Meteorological Insitute, Geophysical Research Division, P.O.Box 503, FIN-00101 Helsinki, FinlandFI

Personalised recommendations