what is the best way to trade skew while hedging delta, gamma, vega, and theta?
- Thread starter njrookie
- Start date
Main source of P&L in skew trades is usually the actual change in implied volatility when underlying moves over to the new level. I will send you a PM, don't want to clutter this space.
If you see skew correlated with volatility, you are not measuring it correctly. Let me guess - are you looking at the 25 delta risk reversals or something of that sort?
Excuse me, njrookie1, I obviously don't want to ask you the secrets of your quantitative model but...
In the past I plotted the skewness of some underlyings (*) and I noticed that variable isn't so easy to deal with (e.g. it's not stationary in covariance, maybe autocorrelated but this is always a problem).
(*) Expressed as SKEW = [IV(90%) - IV(110%)]/IV(100%), same expiry.
In the past I plotted the skewness of some underlyings (*) and I noticed that variable isn't so easy to deal with (e.g. it's not stationary in covariance, maybe autocorrelated but this is always a problem).
(*) Expressed as SKEW = [IV(90%) - IV(110%)]/IV(100%), same expiry.
Are you talking about same constant tenor or same maturity? Also, the series above would be unstable by definition. What you probably want is either in root-time moneyness space (most people use 10% sitrke spread):
SK10 = [IV(90%) - IV(100%)] * sqrt(t)
or skew in distributional space
SKEW = [IV(25delta) - IV(75delta)] / IV(50delta)
Both will yield slightly different results and there are reasons to preferr one over the other, but they would be independent of the ATM volatility, as skew is a whole separate risk premium.
My preference is the root-time expression of skew, it is the most applicable to the equity world and gives you the most insight into the volatility dynamics.
Thank you for your help, sle.
I don't see the need to use the first equation you wrote. I mean: let the trader wants to operate with the same T, so not calendar spread; therefore he will look for skewness which is imbalanced according to what happened in the past.
We could even define a probability density function of the skewness if we look at the past, but doing this we will fix T and then we will examine the second formula you wrote.
So SKEW = [IV(25delta) - IV(75delta)] / IV(50delta) would be stationary? This is very interesting
The reason everyone re-normalizes the skew for root-time is to bring the term structure of skew into the same basis. My advice, in general, is NOT to think about the skew as an indicator of the actual implied distribution, but rather as an indicator of the expected volatility dynamics with respect to spot and the vega convexity position that ensues.
It would be (a) stationary and (b) it would be independent of the level of implied volatility.Quote from Cren1:
So SKEW = [IV(25delta) - IV(75delta)] / IV(50delta) would be stationary? This is very interesting
Thanks sle for bring us back to skew discussion. We need adult supervision here
Normalization by ATM IV effectively assumes vol of vol is proportional to the current vol level.
You want to separate strike skew from the calender skew. skew at fixed maturity fixes this problem (by weighing IV of different maturity appropriately).
I model the whole vol surface. I.e. from historical surface to forecast future IV surface.
If you think ATM IV is the level (or some weighed average like the way VIX is constructed), and the 30 delta - 70 delta IV as skew, then those two processes are not orthogonal. Correlation is about 0.4, ie days where whole IV curve shifts up, it tends to get steeper as well. The elevated vol and steepened skew, of course, is also correlated with the direction of underlying.
I believe I have read somewhere the 2nd normalization you proposed leads to almost zero correlation.
njrookie
PS: sle I somehow cannot review my PM. It says "blocked by administrator".
Normalization by ATM IV effectively assumes vol of vol is proportional to the current vol level.
You want to separate strike skew from the calender skew. skew at fixed maturity fixes this problem (by weighing IV of different maturity appropriately).
I model the whole vol surface. I.e. from historical surface to forecast future IV surface.
If you think ATM IV is the level (or some weighed average like the way VIX is constructed), and the 30 delta - 70 delta IV as skew, then those two processes are not orthogonal. Correlation is about 0.4, ie days where whole IV curve shifts up, it tends to get steeper as well. The elevated vol and steepened skew, of course, is also correlated with the direction of underlying.
I believe I have read somewhere the 2nd normalization you proposed leads to almost zero correlation.
njrookie
PS: sle I somehow cannot review my PM. It says "blocked by administrator".
Bob,
Can you elaborate on "Trading volatility changes in strikes that get out of line from a large order"
I trade options in conjunction with underlying position, I always found that pure options traders just disguised their directional bets in options positions that eventually lead to a larger position that they would not have taken anyway from the beginning??
Thanks
According to this trick, I think that you could explain me (us?) another trick just a bit similar to yours. I've read about it in Sinclair's book, but I did not realy understand how it really works and what's the equation to set the IV surface as the author does.
See the attached image, please.
Not true. In general, "pure" options traders trade volatility, which can either be a true directional bet (e.g. if you are long vega, you are short the market) or could be truly market neutral (e.g. a collar skew position is a bet on volatility dynamics with respect to spot and not much else).
All he's doing is re-normalizes the vol surface by the ATM volatility X = Vol(K)/Vol(50d).
Anyway, trading skew is a very "adult" topic, since it's a very shady risk premium. I, personally, have worked out a lot of gimmicks and models for it, but your mileage may vary.