Traditional measures of both counterparty credit risk and liquidity risk fail to account for the fact that banks are tangled in complex webs. Adjusting for these “network risks” is a key focus for regulators and practitioners alike, and the subject of much ongoing research.
Here in the UK, the Bank of England has made strides in recent years to explicitly account for these types of risk and ensure that banks are capitalised against “network losses” on a subset of their exposures through its annual concurrent stress test.
The Bank of England’s model captures so-called solvency contagion: The value of a financial asset is linked to the health of the counterparty; when the probability of default (PD) of a counterparty increases (its solvency declines), the value of that asset should decline — spreading contagion.
More complexly, in a network of claims, if my counterparty’s counterparty suffers an increase in PD, this should also change the value of the asset — the probability of default of my direct counterparty has increased slightly. Banks currently fail to take account of this “second-order” counterparty credit risk.
Funding Contagion Risk
The same maxim holds true on the other side of a bank’s balance sheet. Banks are funded by a number of counterparties, each of which is funded by its own counterparties. Shocks to a bank’s counterparties’ counterparties will have implications for its own funding. Current measures of liquidity risk fail to account for these higher-order effects — systematically underestimating liquidity risk.
During times of stress, what looks like a well-diversified funding base could turn out to be heavily reliant on one or two key nodes to which a bank is only indirectly exposed.
Furthermore, regulatory risk metrics include strong assumptions about the behaviour of funding counterparties during a stress. These assumptions have been crudely calibrated based on limited historical data. For example, the LCR (Liquidity Coverage Ratio) makes simple funding outflow assumptions based on the type of liability.
Quantifying Network Risks
Agent-based models are able to capture network risks by explicitly linking agents together through causal relationships. Furthermore, these causal relationships can be calibrated based on observed behaviours.
This approach means that the propensity of individual funding counterparties to pull funds can be modelled directly. The resulting models can be used to explore the sensitivity of a liquidity risk assessment to a range of alternative behavioural assumptions.
This can facilitate what-if analysis around changes in funding counterparty behaviour — or even allow risk-managers to perform reverse stress tests to understand by how much behaviour would have to change under stress for a bank to get into difficulties.
