Title: Achieving accuracy and correctness in parametric frequentist inference.

Prof. Alastair Young, Imperial College, UK. Email: Contact

Summary: Likelihood-based procedures commonly used for parametric inference are reviewed and analysed in terms of accuracy and conditionality. Two main approaches are considered: the parametric bootstrap and procedures based on refined analytic approximation of the distributional properties of likelihood quantities. Particular focus will be on elucidation of:

• the extent to which parametric bootstrap automatically respects ancillary information;
• quantification of the effects of high-dimensional nuisance parameters on inference and implications in commonly used parametric models;
• effective choice of inference procedure.

Title: Simulation Based Bayesian Econometric Inference for Forecasting and Decision Analysis: Introduction and Recent Developments

Prof. Herman K. van Dijk, dr. Lennart Hoogerheide, and dr. Nalan Basturk (Econometric Institute Erasmus University Rotterdam, Econometrics Department Vrije Universiteit Amsterdam and Tinbergen Institute), The Netherlands. Email: Contact

Summary: This one afternoon set of lectures assumes basic background in simulation based Bayesian econometric inference. The focus is on more advanced, recently developed simulation methods and filtering methods that may be useful for the analysis of flexible dynamic time series models like GARCH processes, time varying parameter models, dynamic mixture models and further for Bayesian model averaging involving marginal and predictive likelihoods. Applications are in fields of economics, neuro-imaging and DNA analysis.

The focus is on the following three topics:

• A concise introduction to Bayesian econometric inference and discussion of the usefulness of basic Monte Carlo simulation methods like Gibbs, Metropolis Hastings and Importance sampling in this context.
• Key issues of the 21-st Century: The knowledge economy with income-education effects, patterns in the brain; risk of rare events and the study of DNA sequences using a class of importance sampling EM algorithms in order to construct accurate finite mixture candidate densities for effective MCMC and Importance Sampling.
• Parallel Sequential Monte Carlo for Efficient Density Combinations: the Deco MatLab Toolbox with Macroeconomic and Financial Economic Applications.