A fresh take on risk and valuation
Back
Article

Non-Cleared Derivatives: Approaches towards initial Margin Calculation

By François-Xavier Duqué, Managing Consultant

and Marc-Louis Schmitz, Partner Valuation Services

Introduction

Since February 2017, the big players have been subject to the obligation of exchanging initial margin (IM) on non-cleared derivatives, while other players will see these IM obligations phased in for them over the next 4 years. In this article, we take a step back to reflect on this important evolution. We discuss the merits of an industry initiative to develop a standard initial margin model (ISDA SIMM) and the extent to which it echoes the Sensitivity Based FRTB Standard Approach.

The European Market Infrastructure Regulation, better known under its EMIR acronym, aims at reducing systemic risk arising from OTC derivatives. The regulation walks on three legs, each materialized by a set of Regulatory Technical Standards (RTS): Reporting, Clearing and Risk Mitigation. Variation Margin (VM) and Initial Margin (IM) are at the core of the Risk Mitigation effort.

While VM covers current exposure, IM is about the potential exposure that could arise following the default of a counterparty – and hence the last VM call – before the contracts can be replaced on the market.

More specifically, under EMIR, exchanging IM should protect both sides to an OTC derivative against a slippage of the MTM during a 10-day margin period of risk (MPoR) with a confidence interval of 99%. Putting a number on such a metric is challenging. The EMIR roadmap presents two paths: either use a standardized schedule-based approach or implement an internal margin model approved by supervisory authorities.

Scheduled-based method vs internal models

The scheduled-based method has the advantage of simplicity. Contract notionals are weighted individually according to their asset class (see table below) and added up. Some offset is allowed by applying the familiar net-to-gross ratio (NGR) to the resulting total. However, calculating IM under schedule-based method proves to be too costly in capital for many market participants. (The NGR is the ratio of the sum of positive and negative MTMs in the netting set (floored at zero) divided by the sum of positive MTMs in the netting set. Hence, the bigger the offset in MTMs, the lower the NGR).

Internal models can reduce the cost of posting collateral on IM significantly by considering all deal features and allowing for less-crude diversification effects. As for VaR and CVA, there are a number of modelling options, extending from parametric models to Monte Carlo stochastic diffusion, not forgetting historic simulation, which is a popular choice among clearers (LCH, CME, Eurex. etc.). Internal models however come with the burden of model governance (model development, validation, approval & maintenance). Besides, market participants are left exposed to the practical issue of managing collateral disputes: checking the correctness of the margin call a participant receives, ideally requires the participant to develop the same IM models developed by all its counterparties. So, a situation where a multitude of models coexist implies a huge operational complexity, arguably threatening the ability to comply with the regulation.

ISDA SIMM

Understandably, the industry initiative led by ISDA to develop a proprietary standard initial margin model (SIMM) has created a great deal of interest among market participants. The ISDA SIMM aims at striking a balance between conflicting objectives: non-procyclicality, ease of replication, transparency, calculation speed, cost, extensibility to new risk, predictable capital allocation, and, last but not least, being an appropriate measure of the risk involved. Next to these desirable features, participants using the SIMM model for each other’s margin calls benefit from more transparent dispute resolution. They also find relief on the governance front, as the ISDA owns the responsibility of maintaining the model, for example by performing an annual calibration of the model parameters. The model consists in a relatively simple three-step calculation:

  • Firstly, risk sensitivities (“greeks”) are allocated across different product classes, risk classes, risk factors and risk buckets.
  • Secondly, risk weights representing a 1-in-a-100 10-day market stress are applied.
  • Thirdly, weighted sensitivities are aggregated via a sequence of nested correlation matrices.

What to expect?

Applying the ISDA SIMM model might potentially decrease the initial margin requirements by 5 compared  o the schedule-based method. The significant advantage of the SIMM model lies in the possibility to 1) offset sensitivities of different deals within risk classes and 2) account for diversification across risk classes (via the use of correlations in the aggregation) within each product class.

However, the advantages of the SIMM model over a schedule-based approach are not indisputable. If trades are long-dated, always in the same direction or denominated in high volatility currencies, then the schedule-based methodology may be the simplest and most efficient solution.

The advantage of the SIMM model over internal models is real but comes with a caveat. Sensitivities are not easy to calculate; different models will produce different sensitivities. Because ISDA SIMM does not cover this aspect, it will not eliminate all collateral disputes. But the existence of shared and standardized templates for sensitivities will facilitate the reconciliation of discrepancies.

Akin to FRTB SA?

ISDA SIMM shares many similarities with the Sensitivity Based Approach in the FRTB Standard Approach. Both SIMM and FRTB aggregate weighted sensitivities by using a sequence of nested correlation matrices.

As popular wisdom goes, the devil is in the details. Below is a list of points illustrating that SIMM and FRTB are “similar but somehow different”:

  • SIMM and FRTB share 4 risk classes (IR, FX, EQ, CO), but credit is split across 2 risk classes in SIMM (qualifying versus non-qualifying) versus 3 risk classes in FRTB (non-securitisation, securitisation, correlation trading).
  • Both rely on delta and vega sensitivities. However, SIMM computes curvature as a transformation of vega, while FRTB requires an additional sensitivity to be calculated.
  • SIMM does not allow offset between sensitivities in the same risk class across different product classes (e.g. DV01 of an IRS does not offset the DV01 of a CDS).
  • SIMM has two additional interest rate buckets (1W, 1M).
  • Weights and correlations are obviously different (calibration for capital versus margining).
  • Unlike SIMM, FRTB calculates the risk charge with three sets of correlation (high, medium, low) and takes the largest of these.
  • Unlike FRTB, SIMM uses concentration thresholds to account for concentration risk.

Still, the analogy between ISDA SIMM and FRTB contributes to the appeal of the ISAD SIMM approach. Overlapping of SIMM and FRTB requirements may minimise implementation costs significantly and allow banks to benefit from a systematic approach to risk management.