A1585
Title: Efficiency improvement of Bayesian estimation by applying ASIS and its applicability
Authors: Makoto Nakakita - RIKEN (Japan) [presenting]
Teruo Nakatsuma - Keio University (Japan)
Abstract: Researchers have developed more complex models for more realistic data analysis. In general, model complexity tends to increase computational burdens in terms of both computing time and memory/storage usage. As for Bayesian statistics, in particular, the model complexity makes statistical inference with the posterior distribution almost intractable and impractical. To tackle this problem, numerous computational methods have been developed since the late 20th century. In this context, the ancillary-sufficiency interweaving strategy (ASIS) was proposed in a past study. ASIS is an algorithm to improve the efficiency of the Markov Chain Monte Carlo (MCMC) method. It is a very powerful tool for improving statistical analysis's computational speed and accuracy. ASIS is applied to Bayesian estimation using artificial data with pre-known true values, and it is shown that there are no problems with convergence. In addition, posterior distributions of the parameters estimated by "Centred Parametrisation" by another study on which ASIS is based and by plain-vanilla MCMC are compared with those of convergence destination, convergence speed, and inefficiency factors. Finally, by applying the method to time series data and panel data analysis using real data, the efficiency of statistical numerical analysis is demonstrated.
