par Roukny, Tarik
Président du jury Stützle, Thomas
Promoteur Bersini, Hugues
Co-Promoteur Pirotte, Hugues
Publication Non publié, 2016-01-11
Thèse de doctorat
Résumé : The recent financial crisis has brought to the fore the need to better understand systemic risk, that is, the risk of collapse of a large part of the financial system and its potential effects on the real economy. In this thesis, we argue that a proper assessment of systemic risk must include an analysis of the network of interdependencies that exists between the different financial institutions. In fact, today's level of financial interconnectedness between and among markets has proven to have ambiguous effects. On the one hand, a highly connected system allows to diversify risk at the micro level. On the other hand, too much interdependencies provide various paths for contagion to take place and propagate at the macro level. In what follows, we analyze financial markets as networks of interactions and dependencies between financial agents. Through this lens, we investigate three major aspects: (i) how the structure of financial networks can amplify or mitigate the propagation of financial distress, (ii) what are the implications for macro-prudential regulation and (iii) which patterns of interactions characterize real financial networks.We start out by delivering a stability analysis of a network model of interbank contagion that accounts for panics and bank runs. We identify the effects of market architecture, banks' capital ratios, market liquidity and shocks. Our results show that no single network architecture is always superior to others. In particular, highly concentrated networks can both be the most robust and the most fragile depending on other market characteristics, mainly, liquidity.We then move on to tackle issues related to the building of regulatory frameworks that adequately account for the effects of financial interdependencies. We propose a new methodology to compute individual and systemic probabilities of default and show that certain network characteristics give rise to uncertainty. More precisely, we find that network cycles are responsible for the emergence of multiple equilibria even in the presence of complete knowledge. In turn, multiple equilibria give rise to uncertainty for the regulator in the determination of default probabilities. We also quantify the effects of network structures, leverage, volatility and correlations.Having introduced a way to overcome multiplicity, we deliver a method that quantifies the price of complexity in financial markets based on the above mentioned model. This method consists of determining the scope of possible levels of systemic risk that can be obtained when some parameters are subject to small deviations from their true value. The results show a price to the interconnected nature of credit markets even when the equilibrium is unique: small errors can lead to large mistakes in measuring the likelihood of systemic default. Extending the model to account for derivative contracts, we show that error effects increase dramatically as more types of contracts are present in the system. While there is an intuition for such phenomenon, our framework formalizes the idea and quantifies its determinants.In the last part of this thesis, we contribute to the quantitative analysis of real financial networks. We start with a temporal network analysis of one of the major national interbank markets, that is, the German interbank market. We report on the structural evolution of two of the most important over-the-counter markets for liquidity: the interbank market for credit and for derivatives. We find that the majority of interactions is concentrated onto a set of few market participants. There also exists an important correlation between the borrowing and lending activities for each bank in terms of numbers of counterparties. In contrast with other works, we find little impact of the 2008 crisis on the structure of the credit market. The derivative market however exhibits a peak of concentration in the run up to the crisis. Globally, both markets exhibit large levels of stability for most of the network metrics and high correlation amongst them.Finally, we analyze how banks interact with the real economy by investigating the network of loans from banks to industries in Japan. We find evidence of a particular structure of interactions resulting from the coexistence of specific strategies both on the lending side and the borrowing side: generalist agents and specialist agents. Generalist banks have a diversified portfolio (i.e., they provide liquidity to almost all industries) while specialist banks focus their activity on a narrow set of industries. Similarly, generalists industries obtain credit from all banks while specialist industries have a restricted number of creditors. Moreover, the arrangement of interactions is such that specialists tend to only interact with generalists from the other side. Our model allows to structurally characterize highly persistent, and economically meaningful, sets of generalists and specialists. We further provide an analysis of the factors that predict whether a given bank or industry is a generalist. We show that size is an important determinant, both for banks and industries, but we also highlight additional relevant factors. Finally, we find that generalist banks tend to be less vulnerable. Hence, how banks position themselves in the network has important implications for their risk profile. Overall the results presented in this thesis highlight the complex role played by financial interlinkages. Therefore, they demonstrate the need to embed the network dimension in the regulatory framework to properly assess the stability profile of financial systems. Such findings are relevant for both theoretical modeling and empirical investigations. We believe that they also shed light on crucial aspects of systemic risk relevant for policy making and regulation of today's complex financial systems.

Documents en relation

DI-fusion