Ramadiah, Amanah; (2020) The Quest for Accurate Systemic Stress Tests. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Main.pdf Download (23MB) | Preview

Abstract

After the global financial crisis of 2007-09, network approaches have become more prominent for analysing systemic risk in financial networks. Accordingly, various systemic stress testing models have been introduced in the literature. However, relatively little work has been devoted to study the accuracy of these models to predict systemic risk. This is especially important, given that stress tests are subjected to a range of modelling choices. In this thesis, I address this gap by studying the impact of modelling choices on the outcome of stress tests. I mainly focus on indirect contagion channel due to common assets holdings (overlapping portfolios) between financial institutions. In particular, this thesis is concerned with three aspects of stress tests related to indirect contagion: (i) the network of interactions, (ii) the dynamics of contagion, and (iii) the perimeter, or the types of institutions that are included in the stress test. With regard to the first aspect, the network of financial linkages between financial institutions is often lacking, and one has to resort to network reconstruction methods to infer the network from partial information. In this thesis, I conduct a horse race between different network reconstruction methods in terms of their ability to reproduce the topological features and the levels of systemic risk of the actual bank-firm credit networks in Japan. I find that there is no single "best" network reconstruction method - it depends on the assumed criterion of interest, but the reconstruction method which preserves the actual degree distribution overall consistently performs best. Furthermore, I find that the observed credit network displays relatively high levels of systemic risk compared with most reconstruction methods. Concerning the second aspect, the propagation of shocks between financial institutions is usually modelled by means of effective dynamics, which are only approximations of the true dynamics. In this thesis, I introduce a generalised stress testing model that captures a wide range of behavioural assumptions with regard to banks' liquidation dynamics under stress. The literature has proposed two alternative classes of liquidation dynamics in this regard, all of which are covered by my model: threshold dynamics (banks liquidate their investment portfolios only after they have defaulted), and leverage targeting dynamics (banks constantly rebalance their portfolios to maintain a target leverage ratio). I test the capability of the generalised model to predict actual bank defaults (and non-defaults) in the United States for the years 2008-10. I find that the model performs better than alternative benchmarks that do not account for the network of common asset holdings, irrespective of the assumed liquidation dynamics. I also show how the best performing liquidation dynamics depend on the combination of the initial shock level and the market impact parameter, on the cross-sectional variation in the market impact parameter, and on the number of asset liquidation rounds. Finally, with respect to the third aspect, it is mandatory to define the types of financial institutions that the stress test covers. In this thesis, I study the impact of common asset holdings across different financial sectors on financial stability. In particular, I model systemic risk arising from indirect contagion between UK banks, UK open-ended investment funds and UK insurance companies. I show that performing a stress simulation that accounts for common asset holdings across multiple sectors results in systemic losses that are 47% larger than those obtained by summing the losses of sector-specific stress simulations. In other words, ignoring common asset holdings between different financial sectors may result in an underestimation of systemic risk.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item