Using SOMs to analyse financial/banking crises Systemic financial crises are known to have real, adverse impacts on the socio-economic well-being of not only the originating country’s citizens, but also those whose financial systems are affected through cross-border contagion. In particular, systemic banking crises arefound to increase the probability of both currency and sovereign crises.It has been shown that with timeous intervention by national governments, the severity and duration of this impact can be lessened. However, effective policy intervention requires the existence of a reliable early-warning system (EWS).
In general, the variables chosen to develop early-warning systems for financial crises are basedon economic theory and known transmission mechanisms. However, given the complex, non-linear nature of these systemic events, and the importance of early policy intervention, it is not well-established whether (1) this approach captures the full set of variables that could beused to improve the accuracy, precision, and recall of an EWS, and (2) this set of variables,or perhaps a subset thereof, are consistent across the considered countries and time periods.
The goal of this project is to produce a rigorous data analysis report, using newer machine-learning techniques, to identify the variables that appear to signal imminent financial crises,or the nature of ongoing crises, and how these variables may change across countries and/or time. This is in support of a larger research effort to establish the statistical superiority (or lack thereof) of a variety of methodological approaches to building an EWS, ranging from extant statistical approaches to newer machine-learning approaches.
Specifically, the scope of this project is two-fold, utilizing self-organising maps (SOMs) to:
- Identify per country, which variables appear to indicate periods of imminent and on-going financial crisis. (This analysis will employ a supervised SOM.)
- Identify across time, which of the countries are most similar, and analyse the cross-border transmission of financial crises (that is, whether they cluster together on themap when a crisis occurs in any one of them). (This analysis will use an unsupervised SOM.)
In addition, this will produce a unique visual representation of financial crises in two-dimensional space that might assist in furthering the understanding of intra- and cross-border transmission mechanisms.
The project proposal contains more information on the scope of this project and details the methodology used.