I have a current trade in SPX. I noticed today that, when OV modeling is set to Variable, the current price is well below the T+0 line. When I use EIOIO as the modeling framework, the Graham dot is spot on the T+0 line. I have experienced this any number of times in other trades as well. Does this mean that the EIOIO is actually doing a better job of modeling how the market is pricing my overall position in this trade? I have attached two screenshots to show the difference.
I asked Ken Dole about your question and here is what I got: Analyze is all theoretical values based on OptionVue7 modeling and the matrix is your actual G/L. They are not intended to match. You can look at the Analzye t+0 as OptionVue7’s fair value for your position. That’s the short answer. Here is the full description of the modeling which creates the theoretical lines on the Analyze graph: The Variable Volatility Modeling, which generates the projected volatility and the Analyze graph time lines is created in this fashion: The Bid/Ask generates the Mkt Price and from the option market price the MIV (Mid Implied Volatility) is calculated. The next step is the IV’s of all options for any given option series are used as the raw data to build our Variable Volatility Model. From this model a projected volatility can be created. The projected volatility is used to create the theoretical prices for options at any given underlying price and time to expiration. That’s what you’re seeing in the Analyze risk graph and the matrix for the Th.Pr column data. The current IV does not depend on these theoretical prices. Everything in the matrix is based on real bid/ask prices. Everything in the Analyze graph (except the black dot) are theoretical projections based on the Variable Volatility Modeling. Each individual option IV is small part of the data input for the theoretical price (fair value), which is generated by the Variable Volatility Model. The skew graph for each option series is where you can view the volatility curve, which the Variable Volatility Model is based on. (see below screenshot) The Variable Volatility Model is also taking into account the Horizontal variance in IV shift over time to expiration and price of the underlying, which is our CEV (Constant Elasticity of Volatility) model. There are 15 CEV’s used in the model, 5 of which can be viewed from the Model\Volatility\Horizontal Skew window, as shown below. The variance from 1 in the CEV shows the amount of the shift in IV as the underlying price shifts, 1 being 0 shift. CEV numbers below 1 show an IV shift which has a negative correlation with the underlying price and CEV numbers above 1 show an IV shift which has a positive correlation with the underlying price. You can see from the below screenshot that the CEV (hence the IV shift) decreases dramatically for the same move in the underlying price as you move out in time to expiration. Let me know if you have any other questions.
James, Thank you for this thorough reply. An assumption behind my question is the idea that any model is attempting to mimic how an options position will behave in the marketplace as conditions change. I do understand that model is theoretical in some sense, but is not the goal in anymodel to, as often as possible, anticipate how the position will actually be priced in the market place? Hence, my observation that I am surprised how often the EIOIO model agrees with how the market is pricing a position at a given moment in time. Since the Variable model is evaluated (by people who know much more than I) as being superior, I am wondering why the EIOIO is matching the market price as often as it does. Thank you again for taking time to consider my question
Hi Jay, The EIOIO model takes the Market Price of each option price at the current moment and models it as fair value whereas the variable model is taking a snap shot of all the options in the matrix and plotting them on a "line" and assigning a best fit/fair value based on these calculations. I hope that explains the difference better.
EIOIO is "accurate" for estimating the current P&L in the position. Variable variance is much more accurate than EIOIO for modelling what will happen if price were to move up or down. This should help:
Thank you, Ron, for the video link. Very helpful. Has anyone done any analysis of the predictive accuracy of OV going "forward". In other words, if I used the Variable model and asked OV to show (on day T+0) the T+1 line, could I then wait until tomorrow, note where the market actually moved to on that day, and then see if the market value of my position at that level is what OV predicted the day before? Assuming there was not a large volatility change, should not a good model affirm that the T+1 value predicted by the model (on day T+0) is close to what the market price actually is for the position at the end of day T+1? And even if the vol did change from day T+0 to day T+1, presumably the Variable model could incorporate that as well in the T+1 prediction by simply adjusting the vol inidicator on the Analyze graph in OV when the prediction is made on day T+0. Is it valid to look at the model this way, and, if so, has any done this and published any results? Thanks, Jay
Yes, Jay, I've looked into exactly what you're mentioning. I even wrote a formal report on it, comparing the current versions of OptionNet Explorer (which uses the same model as ThinkOrSwim) to OptionVue. My results showed that OptionVue had the most accurate options modelling system. In OptionVue I was using: the Yates pricing model True Delta, Gamma Variable variance I also experimented with comparing the results when enabling "combine call and put skews" in OptionVue. My results showed slightly more accurate results when leaving this setting turned off, when trading the main indexes (note that this goes against the advice given by John Locke on this issue). The report generated a lot of......let's call it "activity". This then had the result of being able to make much closer contact with the creator of the OptionNet Explorer software, and offering to assist with testing of more advanced options modelling within this software. I've now reviewed a number of private betas which includes variable variance, and the results are looking promising. Once a new version of OptionNet Explorer is officially released, I'll re-run my tests and release an updated report (I've currently pulled the original report I put out in order to give the OptionNet Explorer developer a fair chance to improve the software). So although there are many things about OptionVue which I hate and drive me nuts, the fact remains that it currently has the most accurate options modelling available to the general public.
Hi Ron, I noticed on your round table tonight that you have the combined call/put skew box unchecked, although you´ve found it´s slightly more accurate unchecked. Is that because even if it´s less accurate the guidelines for certain trades (e.g, M3, rhino) are based on combined call/put skew deltas? I have an SPX trade on that has a delta value that varies between about minus 125 to 145 without the box checked but only about minus 45 to 50 with the box checked. The trade is very similar to Tom and Jims current Kevlar but with the calls slightly more ITM. That´s a notional difference of $180k depending on which deltas are correct, and this obviously has a massive effect upon the shape of the t=0 line. Interestingly I´m comparing it with a M3 paper trade of similar size (2 calls, 12/24/12, 100 point wings) and the delta of that trade has only marginal difference with combined put/skew checked or unchecked. Here´s screen shots of the Kevlar type trade with and without the combined call/put skew checked, note the difference in deltas, gamma and vega. Differences are so great have no idea what T-=0 line to believe, in addition to the widely diverging P/Ls and black dot positions (using variable model)
Hi David My initial testing found that having the checkbox unticked seemed to give slightly more accurate projections. That said, I'm currently trading with the checkbox turned on. The main reason for this is that I'm following trade guidelines (Rhino and John Locke trades) which were developed by having that checkbox turned on. It therefore makes sense to stick with the same settings used by the trade creators in order to make sure that you are adjusting at the correct location (based on hitting a particular delta limit).
Ok thanks Ron, that´s what I figured. Regarding the issue I mentioned I now see if I set up an identical 100 point wing M3 in SPX but use an iron butterfly instead all puts, the deltas are vastly different with the iron butterfly combined call/put skew unchecked (minus 150 versus minus 55). However the all put M3 has only marginal differences between checked and unchecked. My conclusion is that you have to check the box in OV for iron butterflies (at least in SPX). I´ll speak to OV about this as I know they recommend unchecked.
Do this "experiment": Go to Backtrader (date and time does not matter). Each time you click on "reload". The delta changes significantly, although the option prices stay the same...
Uwe, I don´t have backtrader but minor changes in delta is not the issue.. The issue is a widely diverging delta on an iron BF depending on whether the combined call/put skew box is checked. Here´s the delta values for the identical position as an all put M3 or an iron BF. (15/30/15 x 1925/2025/2125) with 3 x 2000 calls. M3 box checked:-minus 51 M3 unchecked :- minus 68 Iron fly checked:- minus 68 Iron fly unchecked:- minus 142 It´s not an issue using the same set-up on the RUT, only SPX
Sorry to ramble on about this but if anyone´s interested the issue seems to be the "Define Strike" option. If I choose "large" for OTM calls, it more than doubles the delta value of the position if the combined call/put skew is unchecked versus "Small". Big difference so good to be aware of this. I assume missing delta values on far OTM strikes are affecting the modelling but not sure.
Yes, your assumption is correct. By having data for more strikes, you'll get more accurate greeks. Also, if you're doing backtesting, and you wanted to rewind to a particular date, (depending on your settings in OV) then you should go to 3 days prior to that date, and then cycle forward (day by day) to the date you want. This will allow OV to "cache" the data and be able to leverage the 3 days of prior options data in the calculations it uses. In my case, I'm only using the last 1 day of options data (as John Locke recommends), but I know various people who recommend using the last 3 days of data.
Thanks for input Ron but I´m getting the opposite, the more strikes I use the worse it is. Or more specifically the OTM calls on "large" setting are doubling ++ the position delta. If I keep that on "Small" I get a position delta on the iron fly close to the all-put equivalent. I wrote to OV about this and got the following reply:- "It’s true that changing the strike preferences will change the Variable Volatility modeling, which in turn affects the greek calculations. The default of Small ITM and Moderate OTM AutoStrike settings will typically give the best modelling results"
Thanks for the feedback.....although that's quite concerning. John Locke recommends keeping the strikes set to Large, and you would also think that more info would give you more accurate modelling, rather than the other way around. More OV "awesomeness" in action.....
I just tried switching the config for the strikes from all being set to Large over to ITM being set to small and OTM set to moderate. You're correct in that it makes quite a substantial difference.......which again, is pretty concerning. I guess I have to keep going back to the fact if we follow any particular mentor (such as John Locke), then we should stick to the OV settings which he has used in the trades he has created. In this case, it would mean setting all strikes to Large.