The problem with short gamma

[sle]

Quote from RedEyeFly:

Nitro,

Have you ever done any cointegration or RV testing for VS trades?

Variance is it's own little animal with a lot of fuzzy fur and wrinkles, so understanding how trade relative value there is a bit tricky. Now, for first order approximation, you can download the history of VIX, rescale it for the business day count and here you go, you got a rolling history of 1 month variance from the beginning of times.

As a retail trader, however, you exposure to var swaps is a bit limited (unless you are running a fairly large book) and you are better off looking into relative value trading of VIX futures and options. There is like a tonn of things you could do there, from simple spreads and flys to fairly complex option strategies. If you add VXX and VXX options (or other VIX etfs), sky is the limit.

 

[sle]

Quote from nitro:

Right, that is why you want something like a VS. The whole is almost equal to the sum of the parts in VSs, so aggregrating accurate book risk is possible.

Running a large var swap book is about as much fun as a root canal. Yes, you don't have the strike risk but there is soo much other stuff that comes into play (basis, convexity etc) and liquidity is nowhere near the vanilla options.

 

[nitro]

Quote from nitro:

...

What I am trying to say is, trading vola purely is a pipe dream unless you are a bank and have access to exotic variance or vola swaps, or perhaps hyper options. Peter's example show this beautifully.

In summary, you cannot escape movements in the underlying with vanilla options (or even simple spreads), and beginning traders should throw just about all options books in the garbage can that claim you can just trade vola statically even by putting on a spread and forgetting about it, especially from short gamma point of view in the presence of hedging. This is very dangerous misinformation.

"Introduction to SPX Variance Strips

CBOE has enhanced its trading systems with a new technology called BasketWeaver that will enable market participants to trade a large and complex portfolio of SPX options in a single transaction. This basket of SPX options, which we are calling "SPX Variance Strips," is intended to replicate implied variance exposure and will be quoted under the ticker symbol VSTRP. The quoting convention for VSTRPs is similar to that used in OTC variance trading: prices will be quoted in volatility terms, and quantities will be in contracts that have a multiplier (e.g., $25,000 or $50,000) that represents the aggregate vega of the SPX options comprising the VSTRP. CBOE intends to make VSTRP trading available each trading day."

http://www.cboe.com/data/variancestrips/intro.aspx

:eek:

 

[atticus]

Quote from nitro:

We have seen markets enter a regime where there appears to be only one direction to price movement. A more accurate way to express this is to say that the return autocorrelation, or price mean aversion [reversion], is high. This was stated in another way in this thread:

http://www.elitetrader.com/vb/showthread.php?s=&threadid=193829

Textbooks on options trading will tell you to sell a straddle when you expect the vola to be low. Is this correct? What about in markets that have very little mean reversion and in one direction like we have seen for at least a month? Those sure have low vola. The answer is no, with an explanation detailed below.

The problem is that models like Black-Scholes assumes that autocorrelation is zero (it assumes that markets are efficient, i.e., high AC would mean there are predictable trends), and that you should therefore trade as continuously as your trading costs make reasonable. Well, outside of MMs and high frequency firms, costs usually suck the big one and eat into any premium profit pronto. To add insult to injury, the premium you got was probably too little to finance the replication costs of hedging. [Remember, you as an option trader assume markets are efficient. If you don't believe it why the hell trade options? Just trade the underlying if you believe you can predict trends!]

These two things taken together spell doom for any strategy that is short gamma in this regime for all but the most sophisticated, well capitalized, low cost structured firm/trader. If this structure is not true for you, guess what, you are also delta trading.

If option traders assume markets are efficient, by the definition provided by you (why the hell trade options?), then any and all trading (by them) outside of the purview of market-making would involve cognitive dissonance. All trading would involve adding liq and anyone taking liq would be the dumb-money.

Predictable trends? An corr of zero in dispersion would imply betas of 1 on components. Do you see that expressed in the stuff you watch? I don't.

Let's make a fwd-test. You hold a portfolio of short flies and I'll hold the same port of long flies. Let's see who does better (held to exp, no hedging).

 

[sle]

Quote from nitro:
Introduction to SPX Variance Strips

Yeah, yet another waste of an effort - nobody really trades full strips that much anyway, usually people trade lower strikes or ATM against their var swap positions. I think they should have concentrated on giving exchange MMs an incentive to trade their variance futures instead.

 

[nitro]

S&P 500 Variance Futures

http://www.cfe.cboe.com/Products/Products_VA.aspx

Variance Futures combined with VIX futures and the recently launched S&P 500 Variance Strips completes the circle of volatility products.

This thread appears to have good information:

http://www.elitetrader.com/vb/showthread.php?threadid=255405

 

[nitro]

Quote from nitro:


...
There is no way to just trade volatility. In the presence of a skew, and a position that has puts and calls and mixes months, you are most definitely trading the underlying, no matter how much you twist and turn. Variance swaps could help to ditch the mean, but they don't exist as a unit for the retail trader, and replication is expensive.

This has changed slightly. See above posts.

 

[piezoe]

Quote from Martinghoul:

As I have mentioned to you before, nitro, it's all been done already. The whole point of stoch vol models is to treat volatility as a random variable. How to parameterize its behavior is the million dollar question with a whole variety of possible answers. You pick whichever one suits your view.

Once you can correctly parametize the collective, future human psyche and also know the value of the parameters, the problem is solved.

Good luck, Nitro!