Finding the Best Stress Test Model: The Case of Non-linear Fire Sales

We present a model to assess the resilience of financial networks under contagion due to overlapping portfolios. Our model consists of a free parameter \gamma which allows to generate different behavior of banks in term of the amount of asset being sold during fire sales. We apply the model to the US Commercial Banks Balance Sheet Data for the year 2007 and compare the model prediction of defaulted banks with the actual defaulted banks list for the year 2008 - 2011 from the Federal Deposit Insurance Corporation. By iterating over different possible values of \gamma, we are able to find the banks behavior for which the model performs optimal. In general, we find that the model has predictive power by performing better than a random classifier regardless the \gamma value. However, we find that \gamma significantly contributes to the model performance in particular regimes of initial shock and market impact. Finally, we show that considering multiple rounds of deleveraging leads to higher true positive rates, although it does not mean that the model performance gets better.

Our main contributions are two-folds. First, we introduce a stress test model that allows to generate a wide range of banks behavior in term of the volume of assets being sold during fire sales. Our model thus enables us to compare the performance of different models using empirical data which that lead to our second contribution.

Amanah Ramadiah, Daniel Fricke and Fabio Caccioli
Tuesday, September 25, 2018 - 12:15 to 12:30


The official Hotel of the Conference is
Makedonia Palace.

Conference Organiser: NBEvents

The official travel agency of the Conference is: Air Maritime

Photo of Thessaloniki seafront courtesy of Juli Bellou
fb flickr flickr