Econometrics of Central Counterparty Risk Management

A Central Clearing Party (or Central Counterparty, or CCP) is responsible for the settlement of transactions between buyers and sellers on financial markets. It carries the so-called counterparty risk, i.e., the risk that one party cannot fulfill its obligations, during the settlement period of typically two to three days. During this period, in case of the insolvency of a clearing member, the CCP has to step in and has to guarantee the settlement of the transaction. In order to cover exposure from market movements and counterparty credit risk, a CCP requests collateral and margins. CCPs are designed to reduce counterparty default risk, financial distress contagion as well as the frequency of financial crises. Accordingly, they are central pillars for financial market stability and are required by financial regulation. They need to be equipped with significant resources originating from collaterals, margins and default funds to effectively withstand periods of stress. Consequently, CCPs face a tradeoff between a thorough risk management (requiring high margins and collaterals) and the risk of an over-collateralization (and thus impairment) of the financial system. Hence, an efficient management of risk exposure originating from clearing member portfolios is crucial for the stability and the well-functioning of a market. Despite the systemic importance of CCPs, surprisingly little research is performed on the quantification of CCPs risk exposure and the use of econometrics for efficient CCP risk management. This project utilizes innovative disaggregated (anonymous) data of end-of-day net positions of clearing members of CCP Austria, the central counterparty for the clearing and risk management on the Vienna Stock Exchange. The project aims at bringing together recent innovations on the econometrics of risk measurement and the (regulatory) needs of CCPs. An effective risk management of a CCP requires to account not only for the (time-varying) riskiness of the underlying instruments but also for (time-varying) correlations between instruments and member portfolios as well as trade- related risks due to intra-daily changes of members portfolio allocations. A central task will be the construction of prediction models of aggregated as well as member-based and instrument-based risk exposures and underlying correlation structures using daily and intradaily data. A focus will be on trade-related risks, which induce an extra layer of risk beyond price risks. The resulting analysis will be used to shed some light on major risk channels for CCPs and will provide guidance when to optimally exhaust margin buffers in stress periods.

The most important outcomes of the research project can be split into the following major fragments. 1.We have developed a new econometric framework for handling high dimensional data. The framework is aimed to measure uncertainty and risks on financial markets in the presence of non-trivial and complex dependence structure between a large number of market instruments. Our framework is based on the block model for covariance (or correlation) matrices which provides a considerable dimension reduction for high dimension matrices. The model implies a homogeneous covariance (or correlation) among the variables within pre-specified groups, but allows for heterogeneous dependence structure between groups. We provide a simple representation for such class of block matrices that can be obtained by an explicit rotation, thus avoiding costly numerical algorithms. As a result, the framework is easily scalable to arbitrary high dimensions, but remains analytically and computationally manageable. For large correlation and covariance matrices with the block structure, the new representation greatly eases computation of various matrix functionals, evaluation of corresponding likelihood functions, estimation and modeling covariance matrices in high dimensions. 2. We proceeded to investigate various properties of correlation structures which are useful for empirical applications involving high dimensional data. In the paper "A New Method for Simulating Random Correlation Matrices" (joint work by I. Archakov, P. Hansen and Y. Luo) we propose a new technique for generating random correlation matrices based on the log-matrix transformation , where the location and dispersion of the generated correlations can be easily controlled while the positive definiteness is always guaranteed. The new method enables to generate correlation matrices of very high dimension and avoid producing ill-conditioned matrices. The new method can be useful for various simulation-based analyses and estimation procedures. The paper has been conditionally accepted in the Econometrics Journal. 3.We conducted further revision of the paper "Local Mispricing and Microstructural Noise: A Parametric Perspective" (joint work by T. Andersen, I. Archakov, G. Cebiroglu and N. Hautsch), and have investigated a number of aspects related to statistical identification of the parametric models for intraday financial prices. We inspected origins of the potential identification failure and conducted a range of extra analyses to verify that all the qualitative findings in the paper are not affected by a possible lack of identification. We report the extra results in the short corrigendum that complements the main paper published in the Journal of Econometrics.