I wrote and said a number of times, that while BCBS tried hard to set up clear rules around FRTB, it still retained a large number of ambiguities – ones that would have to be resolved by the local regulators, which would defeat some of the FRTBs goals (such as comparability).
A year after the final paper went out, BCBS finally acknowledged it, and issued FRTB FAQ to “promote consistent global implementation.” The FAQs do not resolve all the ambiguities, but the text says this should be the first set, so we will see.
The FRTB FAQ start very promising – the very first issue it resolves is the use of finite difference calculation of sensitivities. The BCBS paper explicitly prescribes this method (and even there only the “up” version). This method requires two valuations of the underlying, which, for more complex products (for example with Monte-Carlo based valuation) can be very expensive. Unfortunately, the faster methods were explicitly prohibited, which often had a significant impact on the architecture of the system. The FRTB FAQ now allows other methods of calculating the sensitivities, although the user must demonstrate that the results are very close to the prescribed formulations (which may still be tricky for barrier products).
The second important issue FAQ addresses, although we get it quite a bit later in the IMA part, is the calculation complexity. Namely, via re-arranging the formula for IMCC calculation, the ratio of undiversified and diversified IMCC can be separated, and BCBS allows this ratio to be calculated on a weekly, rather than daily basis. This means the number of calculations that is required on a per-day basis is reduced to the number of calculations for IMCC (15) as opposed to maximum of 63. The banks still have to be able to switch to daily calculations if required by the supervisor, which on the negative side increases the complexity of the system that has to be able to do both.
There are also two clarification for PnL attribution.
Firms are allowed to align snapshot times for calculation both risk and hypothetical PnLs, and it is made clear that no valuation adjustments (that are calculated less frequently than daily) and XVAs (CVA, DVA etc.) should not be included in hypothetical PnL for both backtesting and PnL attribution.
NMRFs, an area with a number of outstanding questions did receive some attention, but unfortunately, in my opinion, the answers provided still leave too much ambiguities out. The definition of “committed quote” and size have been touched upon, but to a large extent they are just re-hashing of existing document, and in my opinion didn’t really solve the ambiguity. For example, on the size for transaction, it says “non trivial compared to usual sizes”. Well, for a lot of instruments or products with NRMF, there may be only a few transactions now and then (by definition), so what is “usual size” on a small sample?
Nevertheless, BCBS trying to resolve the ambiguities is a definite step forward, and one that should be welcomed.
 As are most of the answers – clearly, it’s easier to address the simple issues than the hard ones.