If I have a huge index option position including long and short calls and puts ,how do I know my skew risk ?25DRR increase 1% or decrease 1%, what is the effect to my P&L??

i would bucket my vega by strike and maturity and see where the holes are

Outrun, how'd you design your scenarios in that case?Currently I measure my exposure to smile as a sum of all vega by strikes, but that gives me poor sense of risk (as I see only a result of moving the whole smile 1% vol towards my positions). Think I should use expectations in RR and BF changes, but how can I do this properly?

You can take implied volatility curves and run a PCA on their daily changes to see the typical patterns and even assign some probability to different moves. You can shock your current curves by the first few principal components (alone, together, ...) to see how the position performs. You could perform regressions of the components against underlying returns to account for a relationship between the underlying moves and the vol moves.

Shifts in the volatility surface are quite dependent from:1. Your current surface2. Your current volatility regime3. The shift on spotIt is reasonable to say that you can fit a GARCH model to the index.Now imagine that your long term vol is around 16% and your recent vols are around 10%.if you have an 1.5% move, this might increase your spot vol, and subsequent large moves may cause the curve to invert.But if your recent vols are around 35% and the curve is inverted, an 1.5% move could start flattening your volatility curve.This is why the impact on implied volatilities is "state dependent" and dependent on the size of the spot move as well.Large moves in skew tipically come together with spot and ATM moves, so modeling just a shift in skew is interesting for:1. Explaining P&L2. Showcasing portfolio mismatchesAnd the best way to calculate it is to bump your inputs one by one (although here the new curve with just one input in one maturity being different might present a very unstable local volatility, so you don't want a large bump).Try to map your strikes in log-moneyness when comparing different maturities as well.

Thank you, gents, for the good food for brain.I think I need something easy and reliable like AK47 - one "greek" to measure this (imagine we have only one tenor).BTW, I use SABR, what if i take all sum of vega by strikes and mix it somehow with ATM vol (alpha) and volvol? Not representative?PS for me its quite strange that given question is far from being widely discussed, as many of us are making markets and have saws in positions

Adding vegas from different maturities is equivalent to adding DV01s from different maturities; because the dynamics are not dominated only by parallel shifts (short-term vols are more volatile), you cannot just add different maturities.Using SABR at each maturity is not going to help you in linking different maturities; look for the most recent Gatheral presentations on arbitrage-free SVI for a suggestion of a parametrization that seems to be able to use the same parameters across different maturities (of course, they are the same until they aren't).

Nonono, imagine we have only one maturity - let's solve from easier.

Map your desired skew change to a vector of changes in the SABR parameters (dependent on the actual curve and SABR fit).Aplly the changes, recalculate the portfolio with the new curve.Risk is the difference between MTMs.

Related question. What is a reasonable way to calculate historical VaR scenarios for swaption smiles. Though smiles are SABR (like) ones, due to system restrictions I can not do this in terms of the model parameters. What do you think about the following.- separate atm from smile, - for atm generate percentage scenarios from historical data for each (option,swap) point in the atm matrix,- for smiles, interpret the strike scale in terms of delta (or log moneyness / or moneyness) even if usually defined in absolute strike spreads,- on top of the atm scenarios apply percentage smile scenarios in this sense, again separately for each (option,swap) point in the atm matrixWhat is your experience with different scales, i.e.delta, moneyness, log moneyness, ... ?Thank you, Peter