10.4225/03/5947315c0327f
KIN PONG LEUNG
KIN PONG
LEUNG
Data-driven Particle Filters for Particle Markov Chain Monte Carlo
Monash University
2017
Sequential Monte Carlo
Bayesian inference
Unscented transforms
Unbiased likelihood estimation
State space models
Econometric and Statistical Methods
Financial Econometrics
2017-06-19 02:05:13
Thesis
https://bridges.monash.edu/articles/thesis/Data-driven_Particle_Filters_for_Particle_Markov_Chain_Monte_Carlo/5110780
This thesis proposes two computationally efficient algorithms suitable for analysing complex time series models under the Bayesian statistical paradigm. Certain critical theoretical properties of the algorithms are established, with their superior empirical performance demonstrated via a suite of controlled simulation experiments. A comprehensive review of the literature is provided, and the performance of many existing methods compared with that of the proposed algorithms. The thesis explores three particular classes of time series models, however the approach is general and finds application across the economic and other social science disciplines, as well as disciplines from the physical sciences.