From September 1, the new rules for margining uncleared derivatives are live – at least for US and Japanese companies (EU delayed the implementation earlier this year). The rules now cover only the initial margin, but March 2017 all participants will have to post variation margin as well.

One of the best known models for the initial margin model is SIMM – Standard Initial Margin Model by ISDA.

The interesting bit on this model is that it has changed quite a bit in the last year, and the final version very closely resembles FRTB SA.

Just as FRTB SA, it uses sensitivities to calculate (in effect) a parametric VaR with a few variation-like formulas.

That said, there are some subtle, but important differences.

Firstly, and very importantly, the asset classes aggregate trades, not risk factors. This is a very important fact, because it means that your CDS interest risk will not be offset by your IR swap. I believe that ISDA is in discussion with regulators to bucket risk factors, not trades, which would be a substantially better solution (amongst others, the decision of whether a trade is a credit or IR or equity etc. feel be a bit arbitrary).

SIMM also has a different number of asset classes than FRTB – there are only two credit asset classes, and FX and rates are treated as one asset class.

Last of the major differences is that SIMM of course has no default or residual charges – from that perspective, it would be more accurate to compare SIMM to Enhanced Delta Plus rather than FRTB-SA.

There is also a number of smaller differences in detail, such as

•  S values are calculated differently (ISDA claims its methodology is more precise, and that FRTB is more conservative – but capital should be conservative).
• The risk factors is very slightly different from FRTB, for example the IR has more buckets towards the short end of the curve (in addition to FRTB’s tenors it includes 2 weeks and 1 month).
• FRTB includes correlation (high/medium/low) adjustments and taking the worst, SIMM adjusts risk factors with a concentration factor (admittedly, these two aren’t comparables, but are differences in methodology)
• Curvature calculation in FRTB must use full revaluation, while SIMM relies on delta/gamma (with gamma derived from vega) approach. The reasons for this are actually quite theoretical (using normality assumptions on a non-normal distribution for curvature PnLs), but the added benefit is that it is also faster and easier to calculate. Of course, if the portfolio is full of products with discontinuous prices, it may create problems of its own.

The above can invite a comparison where SIMM and FRTB’s EDP could share implementation. We believe this should be approached with some caution, for a few reasons:

• The implementation of the variance-like formula across a few buckets is very easy, even if it has to be implemented couple of times.
• There is still a number of details that are different, and special code would have to be written to handle it, with potential of making the code less readable
• ISDA’s SIMM is “patent pending”. In our view it’s unlikely that it will get the patent, but one never knows – we are not patent lawyers (strangely enough, FRTB could fall under prior art here). Should it succeed, having SIMM and FRTB “joined at the hip” may prove costly should a separation be required for whatever reason down the road.
• We believe that implementing the SIMM or EDP formulas is relatively small part of the whole project implementation, so the relative saving is small.

Where we believe the approaches can provide saving is sharing the process calculating the standardised sensitivities, as there is a significant overlap.