IMA-FRTB – The Computational Challenge
A Benchmark Study of Fast-Pricing Solutions
For banks applying for the Internal Model Approach under the coming FRTB rules, P&L attribution tests (PLATs) are proving extremely challenging. Among those who have managed to pass these tests in simulated setups, there is common agreement that full revaluation is the best way to succeed in the P&L attribution tests. However, IMA-FRTB via the Front Office prices without any tailoring imposes outstanding hardware costs and IT challenges. The only practical solution seems to lie in numerical techniques that provide quasi-full-revaluation: very accurate and fast proxy-pricing
We run a case study with 60 trades (IR swaps, IR Bermudan Swaptions and Exotic Barrier Options) with 5 underlying NMRFs to compute the IMCC charges, VaR and P&L attribution tests. We compared different numerical pricing techniques: Taylor series expansions, linear interpolation Grids and “Smart” Grids via MoCaX Objects. This simple portfolio required 0.7 million revaluations for a full IMA-FRTB computation. All pricing techniques were much more computationally efficient than “brute-force” full revaluation, around 50 times, but the only technique that was able to pass the difficult PLATs was “Smart” Grids via MoCaX Objects. That was the case because it is the only one of the tested pricing alternatives that is both ultra-fast and ultra-accurate.
In this paper we explain the details of this comparative study.