So Many Margin Models
So Many Margin Models
The focus on clearing and risk management is creating an expanding need for mathematical models to calculate margin requirements in many contexts. In the exchange world SPAN has been the staple calculation method, and widely implemented by software vendors, whereas since the arrival of SwapClear and the global take up of central clearing, various variations of Value at Risk (VaR) have become necessary both for margin, capital and portfolio modelling reasons.
Each Central Clearing service (CCP) has adapted portfolio margin models to suit their own purposes, and in doing so created a need for users of CCPs to replicate these margin models, to validate and explore the drivers for Initial Margin (IM). Most CCPs for Rate Swaps adopted a similar approach to SwapClear with a historic VaR, details of which are in our earlier article, but with variations on the holding period, confidence interval and scenarios. For Credit products ICE developed a proprietary multi-factor model to take into account the increased complexity of the pay-out on a CDS portfolio, as did LCH. Clearnet for CDS with a Monte-Carlo approach to portfolio margin.
On the horizon now is the new requirement to calculate and exchange Initial Margin for non-cleared OTC products. Firms are free to use any model provided it meets the regulations and is explicitly reviewed and approved by their regulator and is agreed with each of their counterparts (this is to avoid disputes and additional capital charges). A model for this purpose must either use a simplistic schedule of a percent of notional, or use a VaR model with a minimum of 99% confidence level and up to 5 years history with embedded stress scenarios. The regulations also ban the use of scaling such as GARCH or EWMA which is commonly used to reduce the significance of aging volatile scenarios.
Additionally CCP’s are increasingly looking at ways of reducing the impact of Margin on their members whilst maintaining or even strengthening the risk positions. By way of illustrating this, we can look at the work done at Eurex.
Eurex offers clearing service for Interest Rate Swaps (IRS) using its PRISMA methodology, which calculates IM on a portfolio level using volatility filtered historical simulation VaR (FHS VaR) with 750 historical observations, and also floors the IM by a Stressed Period VaR (SP VaR) with 250 stressed scenarios that are identified from more than 10 years of history. Alongside adopting these methodologies Eurex introduced cross-margining (XM) of Fixed Income (FI) Derivatives and IRS. Both FI and IRS positions in the same account with time-to-expiry over 5 days are merged into the same IM calculation group in order to reduce the overall IM requirement by summing up the VaR scenario P&Ls.
Elsewhere in the Prime Brokerage (PB), Stock Loan and Repo businesses margin requirements are still used to cover the risk on those portfolios, but each has different approaches. Many firms who were active in the PB business pre-Lehman had a range of methods to measure risk on the client portfolio, some of which would have been proven to be inappropriate given the collapse that occurred, and will have now moved to a VaR based approach.
Regulations have provided an opportunity for CCPs to launch hybrid products which allow the underlying OTC market to be traded as a future, sometimes referred to as “swap futures”. During their development there was debate about the appropriate holding period for such a product, with some arguing that a future on a swap should have a similar holding period to holding the actual swap, but the argument didn’t hold, so the holding period on a swap future can be as low as a one day period, compared to a typical five day period for cleared OTC swaps.
Variations on a Theme
Each margin calculation needs configuring to suit the context, sometimes for CCP portfolios, sometimes for limit checks and sometimes for internal calculation needs. Common configuration options on a VaR model include:
- Scenarios: This is usually a historic time series going back between three and give years. Sometimes this can be a set of random scenarios generated by a Monte Carlo engine. Some debate takes place on the number of historic scenarios, with CCPs currently choosing periods between 10 years and 3 years.
- Scaling: CCPs in general apply weighting to the volatilities to reduce the significance of the oldest scenarios. This has two outcomes, one making recent scenarios drive the IM outcome directly, and two, making the effect of highly volatile periods reduce gradually over time, rather than dropping off suddenly. Market participants like this approach as it helps make margin requirements stable, but in a low volatility environment can allow IM to sink ‘too low’. In this case some firms add a volatility floor to maintain a minimum level of IM.
- Holding period (or Margin Period of Risk): Over what time period does the model calculate losses? The longer a portfolio is held in a stressed market the greater the potential profits or losses can be. As mentioned above the holding period for an exchange traded product is short, justified by the ability of an exchange to sell off a default portfolio rapidly. For longer term OTC positions, most CCPs allow for a five day holding period, or longer, to allow time to hedge the portfolio.
- Product scope: The major CCPs provide clearing around the liquid OTC products, such as Rate Swaps or Credit Default Swaps, and also some Equity products. Regulators have specified that for non-cleared OTC products firms must not mix portfolios across the major asset classes and keep Rates, Credit, Equities, FX and Commodities separate, in their believe that there is little correlation across those assets classes in a stressed market. In addition a typical firm may store trade data from multiple business lines in a single warehouse, and the risk platform needs to be configured to retrieve specific assets classes or product types as inputs to a calculation run.
- Outcome selection: In a typical statistical VaR, the output of the VaR model is produced purely mathematically using standard deviations, but with the arrival of CCP margining, the underlying scenario profit and loss results allow for a more complex process. SwapClear used a Worst Case Loss method for many years, taking the worst loss of any scenario. Recently SwapClear moved to an Expected Shortfall approach, which takes an average of the six worst scenarios, giving a more stable outcome over time.
There are many reasons to use a VaR model on a portfolio, some for margin purposes and others for capital and funding purposes. Implementing all these needs individually would be expensive, which suggests that the fewer platforms for these calculations the better. Here are a number of business needs:
- Validating CCP IM calls using your own implementation of their models
- Demonstrating independent CCP IM validation to investors
- Comparing portfolio IM across multiple CCPs
- Optimise the CCP venue or bilateral trade choice to minimise capital, collateral and funding costs (IM)
- Reducing margin costs across CCPs by selecting which CCP should clear a trade
- What-if analysis on a cleared portfolio by adding, or removing trades
- Simulating the take-on of a Client portfolio in a default scenario for a backup Clearing Broker
- Simulating the movement of time on a portfolio to estimate margin costs
- Bilateral margin calculations for non-cleared portfolios, either as part
- Use Monte Carlo/Historical Simulation method to provide profiles over time (forward or backward)
- Calculate IM based on SIMM or internal models between bi-lateral counterparts
The Many Margin Models
I’ve picked a selection of the margin models in use which most firms will need to support as they engage with different markets and their counterparties (see Table 2).
This table doesn’t mention the many other CCPs, or Client clearing or the complexity of some bespoke margin agreements that banks provide to clients. When you combine all these venues with the permutations of margin calculator configurations, a firm needs a flexible calculation platform to avoid being hit by high costs to support the range of margined business relationships.
All the above point to the need for a platform which provides a high level of flexibility at the outset, many different permutations of calculations and many different business calculation scenarios. The data requirements for the calculation platform rely fundamentally upon trade feeds from around an entire enterprise covering as many assets classes as is relevant, giving Risk teams the ability to drill down and analyse portfolios. Building and operating such a platform is a major investment, and needs a carefully selected partner who brings solutions ‘out of the box’ for integration into your own systems.
This article was first published in edition 3 of Rocket, our magazine. Download your copy here, or save your address in your profile to receive a printed copy.
To save your address into your profile:
- Visit the home page
- Click Account (in the middle of the row of black buttons)
- Click Edit Profile (in the row of buttons at the top)
- Click Reader (top right)
- There you can see your profile, with a box for your address - complete it accurately, and click Save