HelloPlease could someone inform me about growing importance of Risk-Based P&L? RBPL is as I understand it a P&L predict based on the sensitivities method (Taylor expansion). I understand its growing importance due to the Fundamental Review of the Trading Book (FRTB) but am unsure as to why. Thanks

I am not an expert on this, but from my reading it serves as a diagnostic for your risk models. If the P&L predicted by the risk models differs greatly from the actual P&L, i.e. you have 'unexplained P&L', it suggests that your risk models are not satisfactory. Probably there are other risk factors at work that your model does not capture and/or you are not even aware of. Even if those factors are helping you now (positive unexplained p&l), they could reverse and bite you in the future. Keep in mind that the regulators [still] allow banks to use their own internal models, but are increasingly concerned about the quality of those models, in view recent experience. So they are taking a "show me that it works" attitude. But again, I am not an expert on this...

I didn't know that FRTB meant that it was becoming more important...I thought it was already pretty useful. Places that I have worked at have had P&L Explain for years, the key metric being how big is the unexplained portion. I have worked places where we broke the P&L down either by a Taylor Series expansion, or by bump/bump reset as they give good metrics.

Hi Rmax - is there any chance you could send me your doc on P&L Attribution? i'm really just looking to udnerstand the "bump and rest" idea from someone i trust! Thanks in advance (not sure best way to send my email)

Hey, you can read about this in Andersen and Piterbarg vol. III. It's a critical topic on the buy side. It's conceptually quite simple, but getting the details right is not necessarily trivial.Just think of your portfolio as a big mulit-variate function P. You want to decompose your PnL between an initial market state (say, yesterday's close) and a final market state (say, today's close), both known. On one hand you need to evolve your portfolio value P variable by variable. So from this could get the PnL due to every variable move. On the other hand you perform a Taylor expansion of the order of your liking and calculate the risk-based (risk-predicted) PnL variable by variable. The differences between the true PnL and the risk-based PnL is the explain. For options, you can go quite crazy with this stuff and get into vanna, volga, etc...It's the kind of thing you just have to go through and implement...

Thanks so much mtsm (and Rmax). i'll go check that vol.111 out. thanks again

hi againFeeling a little foolish, but I am very clear on:(1) P&L attribution from sensitivities methods (Taylor Expansion) and (2) P&L Attribution via revaluation methods-> bump&reset or bump& run. But when we hear of a front office "Full Revaluation" and compare this to description number (2) listed above - the P&l attribution via a revaluation method - are the 2 the same? So Front Office Full reval is just bump&reset or bump&run?thanks, riskcube

My 2 cents is yes they are the same.

under the new FRTB regulatory regime ,P&L attribution is one of the tools in conjunction with backtesting that must meet rigid criteria for desk level performance and approval to be included in the application for the use of Internal models at desk level ,i.e FI must meet the rigid requirements at desk level (whatever that means...vague ) in order use internal models at desk level under the proposed FRTB guidelines which are set to be complete and the official text to be out in january 2016 , FI need to reconcile actual P&L to theoretical P&L by meeting a regulatory proposed threshold,which is difficult to satisfy :let sigma_actual ,sigma_theoretical and sigma(Theoretical -Actual ) be the respective P&L vols ,your theoretical p&L should be a very good predictor of your actual P&L and this hold if the correlation between the 2 p&l's is at least 90% ! ....it follows thatsigma_(theoretical -Actual ) = sqrt( ( sigma^2(theoretical ) - 2 * correlation (sigma_(actual),sigma_(theoretical))*sigma(actual)*sigma_(theoretical)+sigma^2_(actual))if we assume that sigma_(actual) = sigma_(theoretical) rearranging and factoring we have :sigma_(theoretical -Actual)/sigma(actual) = sqrt( ( 2-2 * rho(actual,theoretical)),so to meet the rigid threshold (proposed) : then (sigma_(theoretical-actual)/sigma_(actual))^2 <20% ,we need rho to be >90%