backtesting a deltahedged option portfolio
- Thread starter TskTsk
- Start date
One route you can take is using Amibroker. Import the iv as an indicator , then fire off backtests using the iv as a buy signal i.e. buy if Iv> x . model the pnl expectation outside iei TOS or hoadley then plug in pnl curve back to amibroker using their portfolio object to mimic the pnl of a specific option position. Using Amibroker in this manner has a very steep learning curve, multiple import/export requiring serious programing help.
It's pretty much a pointless endeavor. You're in the swap; mitigated by the friction associated with your dynamic hedge. I know guys who have blown-up close to expiration from missing the sigma-trigger and watching the underlying revert. Any backtest would assume optimal conditions. Perhaps there is some value in showing that it's fruitless.
You will have a huge sampling error if you just use one option.. Even if you use a straddle, It might show profit for that particular straddle. But how much of that is due to your hedging frequency/strategy, and how much of that just has to do with that singular implied vol/realized vol that might never play out again.. As well over time based on the assumptions made from this singular sample you could give back well over what you have made.. You would need to build a distribution of hedging of options that you have modeled as over valued or undervalued, and test over longer time frame. And see if that distribution has a positive skew.. meaning it makes money over time.. and then you can't even assume that will be what happens in the futures..
I'm sorry folks, but every time I see these detailed, obtuse discussions, I am reminded of an old Atticus quote that made THE difference in my options trading.
"None of this amounts to a hill of d*cks if you can't forecast price or vol with some degree of accuracy."
"None of this amounts to a hill of d*cks if you can't forecast price or vol with some degree of accuracy."
OK, my buddy CD described Atticus as cryptic, so for the benefit of everyone who is not up to speed on what he meant, here it is.
You make the big bucks by being right, not by your skill at putting band aids on positions that are going downhill. You can make the most of a bad job, but it seldom is worth the cost and effort.
My worst loss ever was trying to repair a position, rather than just take the small loss at first.
You make the big bucks by being right, not by your skill at putting band aids on positions that are going downhill. You can make the most of a bad job, but it seldom is worth the cost and effort.
My worst loss ever was trying to repair a position, rather than just take the small loss at first.
I won't be seeing if the results transfer to trading as (obviously) they will not. It's more a way to study the options from a purer vol (no delta noise) standpoint and see if some assumptions I'm making regarding behavior holds up.
Anyways I found the solution. The historical prices themselves contain all the info needed to back out IV, delta, theta, gamma and anything you need via the Excel solver function. Via the two variables price and time left to expiry (easily calculated) one can back out everything from even the most granular option prices (1 sec). Only problem is it's brute force, so takes some time & cpu power.
Anyways I found the solution. The historical prices themselves contain all the info needed to back out IV, delta, theta, gamma and anything you need via the Excel solver function. Via the two variables price and time left to expiry (easily calculated) one can back out everything from even the most granular option prices (1 sec). Only problem is it's brute force, so takes some time & cpu power.
This is the sort of thing at which MATLAB (and others, I' sure) excels...
Unless you are doing something where you are gamma-weighted (e.g. have multiple options of various stirkes), my first suggestion is to use implied and realized volatilities rather then try to back-test the actual delta-hedging process. Otherwise, the path-dependent nature of your P&L will produce plenty of false positives or false negatives. However, if you really want to do do it (e.g. you want to back-test a collection of skew trades), you should take a rolling frame of a specific maturity, reprice the options and calculate the delta. Matlab or R (or Python) would be more useful then Excel.
Thanks for the reply, yes it will be used for multiple options. Could you explain what you mean with path-dependant nature of p&l? if one assumes frictionless deltahedging, shouldnt final pnl derived from iv-rv be the same as the continously deltahedged (frictionless) final pnl? again it wont be used for any trading but more of a way for me to study options...
by the way how do you back out pnl from looking at difference in implied and realized? i.e. is there a formula for it?