Testing The Model

Once the model was developed and the algorithm implemented, our next goal was to find the best way to test the effectiveness of our model. Because our aim was obviously to model a real financial system, we decided that inputting data into our algorithm from a real financial network was the logical way to understand the effectiveness of our model design. The financial data from the European Banking Authority’s 2011 EU stress test is available on an online database, and provides more than enough data to input into our model. We decided to, specifically, use the balance sheet data (risk weighted assets, capital composition) from Greece’s financial network due to its relatively small size. The result of this real-data simulation is explained below.

The results of our test simulation accurately reflected the designed result of our model. Specifically, the modeled banks began to liquidate assets once dropping below the minimum capital ratio threshold, and the consequential fire sale led to a rapid decline of value of the assets being sold. This, in turn, caused a feedback effect of asset devaluation and further liquidation among all banks in the network. The result of the simulation can be seen in the figure shown below. 

The figure above is the output returned from our algorithm implemented in Python. The orange line that decays in the middle of the plot represents the capital ratio of Bank 1. Toward the end of the given time period (near 1), one can see the line plateau at about 0.8. At this plateau, the banks capital ratio has dropped below the minimum threshold. Consequentially, all banks begin to liquidate assets, as their selling ratios increase from zero. This increase in selling ratios is reflected in the lines that move horizontally across the bottom of the figure until a slight increase near the end.