Executive Summary
CCPs set margin requirements and call margin on a wide range of financial instruments to insure against the potential default of members. Typically, margins are calculated and called on a portfolio basis — i.e., they reflect the price risk across the member’s portfolio of positions, rather than being set at the level of individual positions without any reference to any other positions in the member’s portfolio. Portfolio margining encourages better risk management and more efficient allocation of collateral to the greatest risks.
Calculating margins on a portfolio basis requires robust capture of correlations (or the lack of correlations) across a range of market conditions. Current regulatory standards allow portfolio margining across assets that are significantly and reliably correlated. They do not however specify the meaning of significance and reliability in exact quantitative terms.
The purpose of this paper is to describe a portfolio margining standard that gives precise meaning to the type and level of reliability required in the risk modelling of interdependent assets. It is too simplistic to assess the significance and reliability of individual correlations separately. Instead, the level and reliability of portfolio margining techniques depend on the entire correlation structure embedded in the portfolio, and require a portfolio-level assessment standard. This paper lays out the conditions under which the correlations embedded in a portfolio model are jointly significant and reliable enough to yield robust portfolio margin requirements.
To achieve this, we begin by recognizing some of the key issues underlying portfolio risk management and margining, including the nature and impact of time-varying correlations on hedging and portfolio diversification. The paper then expresses the desired model robustness in terms of the probability of under-margining due to model risk, and proposes that models can be deemed reliable if this probability remains below a certain tolerance. This form of model risk is called a type-II error, and the suggested standard is that this error should be <5% to ensure model robustness.
The adoption of such an industry standard would provide a balance between the desire for specificity and consistency on the one hand, and the reality that any standard needs to be compatible with a wide range of risk management and modelling practices on the other. Indeed, the quantitative analysis underlying this approach supports the idea that both strong and weak correlations can be valuable for risk reduction. This in only true to the extent that the margin model captures these variations robustly, including the extent to which they are reliably present during times of stress. The need for correlation stability is greater for the strong correlations underlying hedging than for the weak correlations typically exploited through portfolio diversification. Thus there is a trade-off between the desired correlation strength and the allowable instability.
Given that it is not practical to set precise numerical criteria (e.g., caps or floors) for individual correlations, we propose that each CCP should rigorously establish the isoquant curves — combinations of correlation strength and stability — that represent comparable levels of model risk. The precise definition of correlation will be portfolio- and model-dependent, and it would remain the responsibility of each CCP to develop, implement and document model risk assessment procedures to ensure the risk of type-II errors remains below the 5% standard.
The intent of this paper is to contribute to the industry debate on Portfolio Margining and to encourage debate amongst CCPs and Regulators as to how best to address this difficult issue.
Please click here to download the document.
