Fiche du document
1 mars 2017
- handle: 10670/1.zm8thl
- hal: hal-03458875
- SCIENCESPO: 2441/20hflp7eqn97boh50no50tv67n
Ce document est lié à :
info:eu-repo/semantics/altIdentifier/hdl/2441/20hflp7eqn97boh50no50tv67n
info:eu-repo/semantics/OpenAccess
Mots-clés
Agent-based model Calibration Machine learning Surrogate Meta-model C - Mathematical and Quantitative Methods/C.C1 - Econometric and Statistical Methods and Methodology: General/C.C1.C15 - Statistical Simulation Methods: General C - Mathematical and Quantitative Methods/C.C5 - Econometric Modeling/C.C5.C52 - Model Evaluation, Validation, and Selection C - Mathematical and Quantitative Methods/C.C6 - Mathematical Methods • Programming Models • Mathematical and Simulation Modeling/C.C6.C63 - Computational Techniques • Simulation ModelingCiter ce document
Francesco Lamperti et al., « Agent-Based Model Calibration using Machine Learning Surrogates », Archive ouverte de Sciences Po (SPIRE), ID : 10670/1.zm8thl
Métriques
Partage / Export
Résumé
Taking agent-based models (ABM) closer to the data is an open challenge. This paper explicitly tacklesparameter space exploration and calibration of ABMs combining supervised machine-learning and intelligentsampling to build a surrogate meta-model. The proposed approach provides a fast and accurateapproximation of model behaviour, dramatically reducing computation time. In that, our machine-learningsurrogate facilitates large scale explorations of the parameter-space, while providing a powerful filter to gaininsights into the complex functioning of agent-based models. The algorithm introduced in this paper mergesmodel simulation and output analysis into a surrogate meta-model, which substantially ease ABM calibration.We successfully apply our approach to the Brock and Hommes (1998) asset pricing model and to the “Island”endogenous growth model (Fagiolo and Dosi, 2003). Performance is evaluated against a relatively large outof-sampleset of parameter combinations, while employing different user-defined statistical tests for outputanalysis. The results demonstrate the capacity of machine learning surrogates to facilitate fast and preciseexploration of agent-based models’ behaviour over their often rugged parameter spaces.