Working Paper No. 287
By Raymond Knott and Marco Polenghi
This study considers how the probability of exceeding central counterparty (CCP) initial margin levels can be estimated, in order to provide a timely and informative measure of risk coverage. Previous studies of CCP margining have largely focused on the unconditional distribution of returns, estimating margin coverage on a long-term average basis. The present study extends previous work by estimating conditional margin coverage using a GARCH (1,1) model, so that variations in coverage can be tracked over a much shorter time frame. The model is applied to estimating non-coverage probabilities for two heavily traded derivatives contracts, the Brent and FTSE 100 futures. To account for the well-documented fat-tailed characteristics of distributions of futures returns, several variants of the GARCH model are estimated. These assume that innovations are distributed according to either normal, Student t, extreme value or historical distributions. Backtesting is used to select the best performing distribution. During the sample period, margins are found to provide a coverage level generally in excess of 99%, over a one-day time horizon. It is noted, however, that the coverage probability implied by the model is likely to fall under more volatile market conditions; under these circumstances central counterparties will reset initial margin more frequently and call for margin intraday.