A while ago I bought some historic option prices (CBOE) on a few index etfs. With this I build a sort of 'backtest' environment to test specific option strategies under specific curcumstances. The weekly options (reliable data) only start around 2009/2010. I want to expand the dataset, for example on etf spy. Spy starts trading in 1993 so a few different environments will be included. For the period of 1993-2010 I want to create synthetic option prices, in a grid of strikes 5% ranging around price(close). The way I do this is simply using interpolation based on the historic vix level (binned to int) for that week. I understand this wil not work precisely and will only be a gauge. It is partly fun/curiosity. Also keep in mind this are weekly option prices and the strategies are more 'IRA'-like. So nasa-like precision is not needed. I was wondering if this will be reliable at all, or will be totally bogus? I thought one way to test is comparing simple pricebehaviour of synthetics (each in a grid of percentage distances from price) to the underlying. Has anyone else did something comparible before? If so what are your experiences?
VIX is weighted, ofc. It works fine for OTC conventions when using CS or whatever to arrive at penny strikes for an active series, but what you're wanting to do isn't going to be reliable.
Thanks for affirming not to take it too seriously. Since it is not that much work I will finish it, probably blending it with the output of some populair pricing models, just to see how it goes. The historic starting point was once stumbling on some articles from a writer on seeking alpha, describing a simple strategy of bull put diagonals (long back (6-12m)atm, short front (1w) itm). Strike of front was modelled to price behaviour - optimize for touch. Sounds good in theory, not so in practise.
Your froward will cause you to be off by 5percent annualized. Divs are an issue too. you are prob better off running everything in vol space and interpolating off the surface.
I don’t think you can model weeklies well using a monthly surface. You will miss all the event vol from macro economic data which will massively skew your data.
Indeed a risk, and extrapolating from vix levels calculated from the upcoming 30 days -as destriero says - is also not perfect. I will finish it only for fun purposes. Btw I also looked into the systematic selling of strangles but couldn't find value in it (or yet), in contrary to the fly's (assym).
The vix will overstate the vol systematically because of the skew the big issue is the event vol. 20 vol in the one month could be 20 vol for 4 weeks or 30 vol for 1 week and 16 vol for 3. The market would have been smart and price the events properly. Your extrapolation will be dumb and you find opportunities that wouldn’t have existed.
I finished the interpolation and the results were certainly not perfect: data = spy 1993-2023 orange = synth with atm p/c weekly exp. blue = spy level t-(t-1) * 100 ( normalized) arrow = 2010-05, from then cboe data. So everything before arrow is simulated My eyes first saw a constant spread, but that was negated by looking at the spread over the same timeframe. It shows a constant drift up in spread, overstating the performance of the synth: alas, it were a few fun hours