On model risk and interconnectedness in banks

Sammanfattning: Nowadays banks are more reliant on the use of models and many of these models depend on each other. In this thesis techniques from graph theory are used to depict and study the network caused by these dependencies. This is done by creating a portfolio that corresponds to a simplified version of a bank and then selecting models appropriate to evaluate the portfolio. The importance of each model in the network is then measured using the centrality measures; degree centrality, Katz centrality and Page rank. The model risk associated with specific models can differ depending on the interested party’s views. These views can be reflected in the Katz and Page rank measurements by slightly modifying them. In this thesis three perspectives representing possible views of interested parties are investigated. The perspectives were focused on the amount valued by each valuation model, the sensitivity and the complexity of the models. The results indicate that the connections between the models affect the centrality measures to a greater extent than the risk introduced dependent on the different perspectives. Moreover the results indicate that the centrality measures used are more appropriate to identify potential victims of contagion rather than sources of contagion.