A comment on Basel proposal for FRTB PnL backtesting and attribution
The old version of the PnL attribution test has rightfully been criticised for being overly punitive. Mainly because it could not handle well-hedged trading books, as the divisor in the test metric could be close to zero, exploding the scaled variance. In the latest revision of the rules, a more holistic approach has been devised.
By using two complementary measures namely the Spearman rank correlation test and the Kolmogorv Smirnov test, it will no longer be impossible to pass the attribution test for reasonably well matched model- and risk factor set-up:s.
See the pdf presentation Algo FRTB Kolmogorov Smirnov test for more information or view via link to Kolmogorov Smirnov video