Massively Parallel Computing in Economics and Finance

Friday May 11, 2012

Challenging statements appear recently in the literature on advances in computational procedures like: ‘Tapping the super computer under your desk’ and ‘It is trivial to parallelize a value function iteration using Graphical Processing Units’. These statements refer to the fact that massively parallel computing is becoming an easy and revolutionary tool for speeding up computations in a tremendous way. Applications are spreading rapidly in many fields but occur so far in few areas in economics and finance. Important examples of applications are massively parallel sequential Monte Carlo for Bayesian Inference, in particular particle filtering and/or independence sampling like importance sampling. Other topics are solving dynamic equilibrium models and analyzing agent based models.

However, implementation of GPU-parallelized estimation of econometric models is not so trivial due to requirements of partial independence of the parallel operations, and to data level parallelism that differs from the more common parallel computation model in which parallel processing elements perform separate instruction on separate data. Thus, traditional algorithms must be analytically transformed to allow data level parallelism and implemented in custom GPU kernel code in order to achieve computational speed.

The purpose of this workshop is to discuss the experience and possibilities of massive parallel computing in economics and finance.

Speakers and programme

  • John Geweke, University of Technology Sydney and Erasmus University
  • Neil Shephard, Oxford University
  • Arnaud Doucet, Oxford University
  • Ron Gallant, Duke university
  • Christophe Andrieu, University of Bristol
  • Nicholas Chopin, ENSAE Paris
  • Drew Creal, University of Chicago Booth School of Business

Programme (PDF-format)

Titel and abstract

  • John Geweke
    Adaptive sequential posterior simulators for massively parallel computing environments (abstract, presentation, slides)
  • Nalan Basturk and Tommi Tervonen (joint work with Rui J. Almeida, Lennart Hoogerheide and Herman K. van Dijk (EUR/VU))
    Parallelization experience with three canonical econometric models (slides)
  • Neil Shephard (with Arnaud Doucet)
    Even Bayesians can be wrong robust inference on parameters via particle filters (abstract)
  • Arnoud Doucet
    Hierarchical particle methods for inference in intractable state-space models (abstract, slides)
  • Ron Gallant
    Parallelization strategies: Hardware and software (two decades of personal experience)(slides)
  • Christophe Andrieu
    Particle MCMC, pseudo marginal approaches: An overview and some properties (slides)
  • Nicholas Chopin
    The SMC2: an efficient algorithm for sequential analysis of state-space models (abstract, slides)
  • Drew Creal
    Exact likelihood inference for C.I.R. stochastic volatility models (slides)

Local organizers

  • Nalan Basturk
  • Siem Jan Koopman
  • Lennart Hoogerheide
  • Richard Paap
  • Herman van Dijk (Chair)
  • Ursula David (Assistant Office Manager)

Practical information

For more information please contact ms. Ursula David or Nalan Basturk