Jan

26

Daily sd's 1 (1,1,1,1,1,0,0) mean variation .71 PL 2
Daily sd's 2 (0,0,0,0,0,0,5) mean variation .71 PL -18
Correct forecast, but went bust anyway, due to lumping of volatility.

Asindu Drileba asks:

What would be the best strategy to capture the return of this distribution? How would the position size be computed? Say you have $10.

Zubin Al Genubi replies:

OTM option? Don't know which direction so maybe a strangle? Its an example of a fat tail event surprising someone expecting a certain variance. Like the LTCM guys. $.20? 2%? As a hedge. Depends if its hedge or a trade.

William Huggins comments:

what you're picking up on is that variance alone doesn't describe non-normal distributions very well - you need additional tools like skewness (possibly kurtosis) to pick up on those differences. despite having a better description though, there is the presumption that the data generating process is stable across the sample period, and going forward. I've generally found (despite my poor timing record) that money is to be made when the distribution is changing, not stable (the computers rule those waves imo) so detecting breaks may be more valuable than fixed descriptions.

Peter Ringel writes:

I can confirm this from the math-undereducated trading side. Stability is boring, and boredom can lead to undisciplined trades. Shocks and short-term exaggerations are great.

Art Cooper points out:

Stability is boring, and boredom can lead to undisciplined trades. It's Minsky's Theory when this becomes widespread.

Zubin Al Genubi responds:

Thank you Dr Huggins. That is indeed the point that variance, regression, sd, means, should be used with power law distributions with extreme caution or not at all.

Hernan Avella questions:

Why is all that mumbo necessary when all you need is good entries and good stops? The house never closes and there are so many opportunities ahead. f you need that big of a stop, or it gets triggered so frequent that ruins the profits, your system sucks! It’s not a stop-loss problem.

H. Humbert comments:

I think he is saying the system did suck because it relied on improper statistical analysis, using gaussian distributions for prediction when it should have used a more sophisticated statistical analysis that doesn't make such assumption. If you know of good entries reliably without using statistics, more power to you! And maybe he needs volatility swaps in addition to variance swaps and then his system will be A-OK because that could be a simple way to hedge the fat tails. Since I don't trade, I'm just trying to interpret what's flying by.

Humbert H. writes:

Var swap vs. vol swap would be the purest expression. You could also buy a call on realized variance, by buying an uncapped variance swap and selling a capped variance swap (for historical reasons, the cap is struck at 2.5x the variance swap strike, the cap level acting as your effective call strike).

For 100k vega notional and uncapped strike at 22, and capped strike at 20, and realized vol over the period of 80:

100,000/(2*strike) = var notional = 2,272.72 var units uncapped, 2500 var units capped
Pnl uncapped 13.4mm
Pnl capped -4.1mm
Net 9.3mm for ~0.2m cost, not bad (approx (22-20) * vega not).

Some payouts were on the order of 2000:1 during March 2020. Pre 2020 you had some active sellers:

‘Amateurish’ Trades Blew Up AIMCo’s Volatility Program, Experts Say

H. Humbert responds:

Interesting. And an interesting article. You'd think that after LTCM people would realize that 100 year floods are just named that for convenience. That's why I never buy stocks in insurance companies. He whose name shouldn't be mentioned (not the fractalist but the Middle Eastern guy) always advocated buying black swan options, but I think the Chair didn't think he made money on this.

Kim Zussman links:

The hedge fund titan who’s been watching for ‘black swans’ for decades says the ‘greatest credit bubble in human history’ is set to pop—but he’s not worried


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search