Quantcast
  1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Options Modeling: The Long And Winding Road - Begining

Discussion in 'General Discussion' started by Marcas, Jan 12, 2018.

  1. Marcas

    Marcas Well-Known Member

    Can one trade options without any model? - Yes.
    Can one trade options with bad model? - Yes.
    Can one trade with good model? - Absolutely.

    Which one do you choose?

    I'm aiming for #3. That is fine but how to get there? How will I know when I see good model?
    One of methods is to test models for accuracy and consistency. How to do it? Look at model's prediction and compare it to what happened in reality.

    For example if you have ONE you can do it like this: choose trade for testing, choose time period. Place trade at start date and apply your model. Convenient way to proceed is to open another instance of ONE for moving time forward and leave fist one with your model at starting date.

    Then, on second instance, go 1 day forward , grab SPX level and realized p&l, go back to ONE with model on starting date. Set the t line to t+1 (top left corner of risk graph window with only one t line projection) and position cursor to match grabbed SPX level. Compare real and predicted p&l. You may wish to take notes about IVs as well. If you cant have two instances of ONE you will switch back and forth in single window.
    Repeat this as many times as you wish. Graph your data.
    Then change trade and repeat. Then change it again and repeat. And again, and again. (Or limit yourself just to one type of trade.)

    Then change starting date and start all over.

    When you finish you may have an idea about accuracy and consistency of your model.
    Then change model.

    In ONE we are limited to built-in models. At least I don't know how to test 3rd party's attempts to project fate of the trade. Of course you can use OV or any other backtesting software, ONE is just general example. On-side note: I was doing above a little, mostly to valuate my results, and found that ONE's elasticity model sometimes is unbelievably accurate; and sometimes it is not. I wish I know it but it's propriety.

    As with all manual testing it is a lot of tedious work. Looks like right job for computer. I put together program that can do above in semi automated manner.
    Here are some results for your thoughts if you are interested, I'm all ears.
    Firstly the trade.
    I used (1) BF 50/50 with shorts close to ATM and (2) BF that has shorts placed about 100p below ATM. I used only x25 strikes. Both trades are set about 56 DTE.

    I chose 3 periods for testing.
    #1 with flat market and stable IV – start date: 10 Apr 2015.
    #2 up-marker with IV going from higher to lower – start date: 8 Nov 2016
    #3 down-market with IV going from low to high – start date: 14 Aug 2015

    Now modeling. I used 3 simple modeling methods.
    * iliv (individual leg's IV) when I grabbed IV for each leg from starting date options' chain and apply that for all calculations.
    * wutec (walking up the expiration cycle) when I tried to predict changes in IV by applying IV based on strike distance from ATM in single expiration cycle. I applied simple extrapolation between strikes if captured SPX price fell in between.
    * wutch (walking up the chain) when I tried to capture IV for model from other chains based on DTE's. Here instead of extrapolating DTEs I simply chose chain with the DTE closest to tested t+0.

    Used model is Black-Scholes. IV chain is by others. Free rate is 0.25%. Dividend is 0.
    I know that mature modelers dismiss above inputs on the spot. It definitely belongs to category II (bad modeling) but by starting here we will see how bad they are (if at all) and what improvement can be achieved by application of some work to our inputs.

    I presented data on graphs as difference between t line and real p&l in a way that if t line was 'optimistic' it is shown above zero-line and its distance shows how much it's prediction was of in $ (It should be shown as % but I thought it will be easier to read this way especially that we looking nly at SPX at about same levels), and opposite: if model's p&l 'prediction' was below real data it is shown below zero-line.

    x - line is DTE (weekends are included).

    That's all.

    And disclaimer: I don't insist that my code is free of mistakes.
    ---

    First graph: 50/50 ATM BF in flat market (look it up for yourself on your platform I suppose it isn't necessary to post corresponding SPX and IV graphs).
    Flat_50_50_atm.png

    Second: 50/50 OTM BF in flat market (and low IV I should say).
    Flat_50_50_otm.png

    Now we switch to Up-market (with the same trades).
    Our ATM:
    Up_50_50_atm.png

    Up market OTM trade:
    Up_50_50_otm.png

    And here are our models in severe down-market conditions:
    ATM:
    Down_50_50_atm.png
    Down market OTM:
    Down_50_50_otm.png

    I kept x and y scales fixed for easier comparison.
     
  2. Marcas

    Marcas Well-Known Member

    As addition, risking of dumping too much on forum, just snapshots of some others points of view.
    Here is graph as above: flat market, atm trade, but placed at shorter DTE:
    Flat_50_50_atm_35DTE.png

    and here is brief look at 'iliv' (cute name, hugh?) method for the same trade at different DTEs in down market conditions:
    DifferentDTEs.png

    y scale on this graph is not exactly the same as before - I throw away weekends and holidays as non existent, I think it doesn't make a difference as all graphs are just to get a grasp of issue.
    And remember about disclaimer :)
     
  3. Mark17

    Mark17 Well-Known Member

    You lost me, Marcas. :)

    But before I got lost, the first few lines of your original post got me thinking. With regard to developing a better model by looking at how existing models performed in the past, couldn't this be an AI-related project? If you look at all cases of 1 SD downmoves in the last 10 years, for example, place butterfly trade, move ahead a few days when the market has rebounded by x points and y% IV, evaluate actual PnL vs. modeled PnL... etc. This could probably be broken down into all kinds of conditions with lots of occurrences that an automated machine could pore through?

    I'm not a coding guy. I haven't the faintest how this would be done but it seems to reverberate with a lot of things I've read over the years regarding AI, genetic algorithms, and the like.
     
  4. Marcas

    Marcas Well-Known Member

    Hi Mark17, it is easy to get lost in calculations especially when you forget why are you doing all this work :)
    As per AI - I don't know much about machine learning and I'm not ready for it.
    When you have tool ready (not AI) it is just matter of asking right questions. Examining particular scenarios, like your 1SD example, would be a step on path of creating better modeling method.
    I only tried to roughly evaluate model I'm dealing with, how it behaved for different structures in different DTEs, to have better grasp of 'mental adjustments' that need to be made for t-lines and for creating better, more precise trading plans.
    I'm not a coding guy either, not in popular sense. I just had to learn part of this business because it was the easiest way to proceed.
     
  5. Kevin Lee

    Kevin Lee Well-Known Member

    If the objective is to forecast options prices given a move in the underlying, machine learning would be a great tool to use. For the past couple years, I have tried to forecast IV and IV skew movements by learning and building more mathematical models. I realize that builds up complexities very quickly without increasing much accuracies. After learning machine learning, I realized the task could be done much more easily.

    Actually, I am having a different thought now. Instead of forecasting IV and then calculate option prices, why not forecast option prices directly ?

    Can it be done? I think so. I'm not 100% sure (yet), but I've done some ML work to give me to confidence to say it should work (just a question of what margin of error). That's my goal. I want to totally skip the IV modeling. Instead, I'd like to be able to draw the T+n line without using black scholes at all but rely on deep neural network or some other ML techniques to forecast options prices directly given an underlying price movement. That'll take care of all the complexities in IV and IV skew once and for all.

    With ML, there are many other things that can be done that will be helpful to options trading. But that's a discussion for another day.
     
    Jason A likes this.
  6. status1

    status1 Well-Known Member

    Interesting thoughts but I am not sure why there is a need to calculate the price of an option unless it is just for your own reference
    If the bid and ask is between lets say 0.25 and 0.75 cents and the mid price is around 0.50 cents but your calculation is 0.35 cents what are you going to do ? Are you going to argue with your broker because your option was not filled at your exact calculation ? You are lucky if you get filled at the mid price

    What would be more interesting if you could predict the trader reaction to a certain event and in turn how that would affect the volatility and option prices Of course nobody knows stuff just happens
    Did anyone predict the SPX would go up 150points in one month ? No
    Did anyone predict that the vix will go up as SPX is going higher ? No
    I did hear someone predict SPX to be at 3000 by the end of the year but that was based on some valuation calculation
    Is the SPX going to go up another 150 points in February ? Who knows ?
    At that rate we could be looking at SPX at 4500 by the end of the year
    Are the companies growing that fast or are the traders just buying because the market is up ?
    If someone can predict the collective traders or fund managers psychology to blindly just keep buying for no reason other than because the stocks are going up than you might have something
    Than you might have a chance to predict where the option prices will be but without that it's just a guessing game in my opinion
     
  7. garyw

    garyw Well-Known Member

    The interest is in "forecasting" as Kevin notes, based on some realistic assumptions. Assumptions are easy to come by, but "realistic assumptions" typically have us chasing our tails more than we like.
     
  8. ACS

    ACS Well-Known Member

    Isn't it true that IV = option price? If so, can one be easier to "predict" than the other? I use parentheses because I'm skeptical it can ever be done beyond a certain level of accuracy thanks to the vagaries of human psychology, the ultimate driver of markets. That accuracy seems even harder to obtain when volatility is at the extremes, both high and low.
     
    Ice101781 likes this.
  9. Marcas

    Marcas Well-Known Member

    Answer is correct, question is not. At least it is not he question I'm asking (as much I wish to know what SPX will do in the future).
    I rather ask: what my p&l will look like if SPX move 150 points. You can deal with this problem via 'bended handle bar' method.
    In general I see that for very small changes in inputs BSM model works fine. For little bigger moves 'mental adjustments to t-line are kind of ok (if you are kind of ok with kind of ok accuracy) for 150 points moves... it depends. (Are we talking about 150 points in one day or in one month. Did market was relentlessly climbing up in past month od was dropping like a rock. Are we in bull or bear market... etc) If you are trading small you don't have motivation to pursue what will really happen at 150p move, you are fine with risk of loosing it. When you start scaling in story changes. You pretty much would like to know quite precisely your p&l to take appropriate actions without shooting at dark.

    Kevin, I'm all ears :).
    Your idea of going directly for price is fine, I believe it even may be ultimate goal when you have final working trade set (a la JL if I understand what he is doing in M3) but 'IV' provide good normalization for different strikes and different DTEs. I.m convinced that new metric can be used, based on price in relation to atm and DTE, metric that might be better than IV, but it need to became accepted in community.

    No, it is not true. You have to include probability as well. And you are right about your skepticism, still I think it can be done way better then applying standard BSM. Depends on your motivation...
     
  10. Kevin Lee

    Kevin Lee Well-Known Member

    Let's first clarify something - in this discussion, we are NOT trying to predict SPX or IV. Meaning - the question is not "tell me what SPX or IV will do tomorrow". Instead the question is "IF SPX moves 1% up or 1% down tomorrow, tell me what will the price of SPX options be". These are very different questions. The first one is a prediction question. The second is a modeling question.

    Don't we know that already ? Nope. Not as accurately as we need. Proof point is with the T+n line. How often do you find that the T+1 looks great but then when the market moved the next day, the shape of the T+0 line doesn't look like the T+1 the day before? Why did it change? That's because the pricing modeling isn't accurate. That causes unnecessary losses. Don't you wish the T+n lines to be more stable? A good T+n line should be stable. Need not be perfect but should be better than what it is today. That is what we are talking about in this discussion - or at least that what I thought we are talking about.

    So... if we want the T+n lines to be stable, we need a better forecast or modeling of options prices "GIVEN" an underlying price move. Not trying to predict what IV or SPX price will be the next day.

    Not exactly. Theoretically, if you know IV, you can just plug it into the BS formula and you get options price, but in practice it doesn't work that way. The issue is with interest rate and dividend. What do you use as input? Unfortunately, we cannot just look it up on a website (eg. libor or fed rate) and use those numbers. If we do, we'll get the discrepancies in the Put / Call IVs we see in TOS and OV. I already had numerous discussions with OV on this topic. To be fair, they did improve on it. Issue is it's not fully automated in OV - I have to manually adjust it once in a while.

    Once the IV is wrong, all the greeks are wrong. The interest rates and dividend has to be reversed engineered from the option prices. One example is how IB does it. Their IV is the most correct. If you look at IB's interest rate navigator below for SPX, you'll see the interest rate used for every expiry is different.

    So my point is, even if I get the IV modeling exactly right, there is still one more step to convert IV to option price in order to draw the T+0 line. Why not go straight to model options price directly ?

    upload_2018-1-28_23-16-56.png
     

    Attached Files:

    Last edited: Jan 29, 2018
  11. status1

    status1 Well-Known Member

    That's kind of the point I was trying to make
    If you can't forecast what will happen in the future how can you make a model that will help forecast the t+0 line?
    I mean it would be nice if it can be done but I doubt it can be done with any accuracy
    There are too many variables that is out of our hands that cannot be controlled it's all dynamic with almost infinite possibilities as the market makers constantly adjust the price

    Even if you can make a model to forecast a 1% move I believe that it will have different t+0 line outcome based on the moves mentioned which most likely will not repeat so you may have to have a model for each scenario
    I doubt that you cam make a model that covers all the scenarios
    Also the model for 1% up is probably going to be different than a 1% down and as SPX keeps going higher the 1% move that was at a few years ago around 1800 is probably going to be different here at 2870
    Also is the model going to work the same on all the indexes or just the SPX so than you would need a different model for RUT or any other index
     
  12. garyw

    garyw Well-Known Member

    Steve S's "chain bootstrap paper" does a good job of getting the interest rates error to a satisfactory level (time adjusted Libor rates if your timeframe is 6 months or less), and resolves the dividend issue. The solution for dividend is as you infer by extraction from the option prices. --
    The issue with the T+n line accuracy, is due to the error in the expected IV for each option n-days out! -- The guesses used in modeling are either sticky strike (EIOIO and TOS individual IV) or (Variable, sticky moneyness, sticky delta, TOS Vol smile, <need a better solution and more insight here>).
     
  13. Marcas

    Marcas Well-Known Member

    I don't need to know what will happen with SPX to draw t line. If I new where SPX will be in 2 days I wouldn't need t+2 line at all, what for? I would need only my option prices for where SPX will be. That's all. With modeling you try to cover wide range of possibilities and the problem we are struggling with is how good is our t line comparing to reality.
    You are right it is not easy and requires time end effort. Question is is it worth? I'd say that answer depends on individual case. Most likely end result will not be a simple universal formula that covers all securities.
     
  14. Kevin Lee

    Kevin Lee Well-Known Member

    Thanx for letting me know... Yes, there are ways to handle the IV issue. For me, I did it programmatically by recursion until the Put Call IV converge. I was trying to say it's an extra step to get the IV right first and then the options price. Instead, I'm now attempting to get rid of IV as an intermediate step. Let's see where that leads me.... Still trying and learning every day :)
     
    garyw and Marcas like this.
  15. garyw

    garyw Well-Known Member

    Pardon this "out of band" question.
    OptionVue (which I do NOT have access to) can provide a Volatility (identified below) {near bottom of graphic}. Does anyone know the derivation for the Volatility being used there?
    It appears to be about 26% greater than the 20 Day Historical Volatility, but would like to understand it better (considering replicating it for other uses).
    upload_2018-1-29_9-51-35.png
     
  16. status1

    status1 Well-Known Member

    I understand you can simulate or look on the risk graph what the p/l may be at a certain SPX price but that is assuming that nothing else changes or it's all static view
    Once the trading starts everything starts moving in all directions , time is shrinking, volatility may be going up or down price may be going up or down , individual iv is changing theta is changing at different rate depending on how close you are to expiration
    I maybe wrong but are you trying to find the Holy Grail for modeling ?
    Good luck to you if you are
     
  17. Kevin Lee

    Kevin Lee Well-Known Member

    Hmm... never noticed that before. Sorry I don't know what that means. Why don't you send an email to Ken (OV tech support) to ask ... ken@optionvue.com
     
    Martin likes this.
  18. garyw

    garyw Well-Known Member

    @Kevin: Thnx, email sent. Hopefully reply will be other than "that information is proprietary"!
     
  19. Mark17

    Mark17 Well-Known Member

    I thought that was HV. I always change it to "Avg IV" before looking at a risk graph.

    I'll be interested to hear what Ken says.

     
  20. garyw

    garyw Well-Known Member

    Ken is out of office, so maybe later this week he may respond.
    For those with OV, it would be possibly helpful if someone could report some of the values using SPX and perhaps end of day (time stamps), and note the date, then I can try revers-engineering the derivation (and/or validate, should Ken report favorably). {5 or 6 dates in last one or two years with very differing HV values would be nice to have}
    Send info directly to me, if you think we may be cluttering this thread.
     

Share This Page