T2: Tutorial on Model Selection

Partners involved
Univ. of Cyprus (Local Organizer), Univ. “La Sapienza” Roma, Univ. of Klagenfurt.

Motivation and Objectives

Model selection is a central issue in statistical and econometrical modelling. The tutorial will describe classical methods and recent advances in statistical model selection. Computational aspects will be emphasized. A brief introduction to conventional approaches to model selection and their limitations will be provided. The tutorial will focus on recent advances in tackling large-scale and real-world model selection problems. This will involve the use of optimization techniques and heuristics for computing the best subset models for linear econometric models.


Two days with seven lectures of 90 minutes and several short (60 minutes) sessions in the computer laboratory.

  1. Statistical foundation for variable selection: Model selection strategies for simple regression models; Bayesian model selection.
  2. Foundations of macroeconometric modelling: Structural models, SVAR models, rational expectations models, time series models: ARMA, ARIMA, VAR models.
  3. Principles of model analysis: Model estimation philosophies, ex-post and ex-ante (stochastic) simulation, dynamic multipliers, model evaluation and comparison.
  4. Computational and numerical efficient methods for estimating and updating (non)linear models, e.g., matrix factorization and combinatorial optimization.
  5. Regression trees and graphs for computing the best subset regression models: Branch-and-bound strategy and specific software developed for this purpose.
  6. Heuristic strategies for large-scale model selection problems: Heuristic branch-and-bound strategies; genetic algorithms and neural networks.
  7. Design of parallel algorithms for model selection.