|
Digital Library of the
European Council for Modelling and Simulation |
Title: |
On
Biologically Inspired Predictions Of The Global Financial Crisis |
Authors: |
Peter Sarlin |
Published in: |
(2012).ECMS
2012 Proceedings edited by: K. G. Troitzsch, M. Moehring, U. Lotzmann. European
Council for Modeling and Simulation. doi:10.7148/2012 ISBN:
978-0-9564944-4-3 26th
European Conference on Modelling and Simulation, Shaping reality through simulation Koblenz,
Germany, May 29 – June 1 2012 |
Citation
format: |
Sarlin, P. (2012). On Biologically
Inspired Predictions Of The Global Financial Crisis. ECMS 2012 Proceedings
edited by: K. G. Troitzsch, M. Moehring,
U. Lotzmann (pp. 253-259).
European Council for Modeling and Simulation. doi:10.7148/2012-0253-0259 |
DOI: |
http://dx.doi.org/10.7148/2012-0253-0259 |
Abstract: |
This paper evaluates the
performance of biologically inspired early warning systems (EWS) for systemic
financial crises. We create three EWSs: a logit model, a standard back-propagation neural network
(NN) and a neuro-genetic (NG) model that uses a
genetic algorithm for choosing the optimal NN configuration. The performance of the NN-based models are compared with the
benchmark logit in terms of utility for
policymakers. For creating the NN-based EWSs, we
use two training schemes for parsimonious and generalized models and advocate
adopting to the EWS literature the scheme using validation sets for better
generalization of data-driven models. The performance evaluation shows that
NN-based models, in general, outperform the logit
model. The key finding is, however, that NG models not only provide largest
utility for policymakers as an EWS, but also in form of decreased expertise
and labor needed for, and uncertainty caused by, manual calibration of a NN. |
Full
text: |