Discussion in 'Option Risk Calculator Forum' started by CC, Jul 8, 2017.

  1. CC

    CC Active Member

    Enjoying trying out the tool. :)
    Very eye opening to backtest different systems.
    What are your recommended settings for volatility?
    Adjusting volatility can make or break profitability significantly.
    If I use the "calculate" button, which option's volatility should I use?

    Here's an example trade:

    The volatility for the 4 different positions are different.
    And trying each volatility number yields different results (profitable to unprofitable).

    Would appreciate your advice, thanks!
  2. Volatility is a hugely important input: it drives the computation of theoretical option price before expiration, and the price probability curve throughout the trade. The tool lets you supply your own value, or lets you compute the volatility implied by current option premiums (i.e.: what volatility would cause the model to match the current price?).

    There's a good page about the inputs on the Macroption Web site. The best thing you can do is educate yourself about what it means and how it affects the model. Any specific advice about how to choose it I will leave to greater minds.
  3. Kevin Lee

    Kevin Lee Well-Known Member

    You asked one of the most critical question of income options trading. Unfortunately this is a really difficult question because IV changes over time and changes due to market sentiments. Not only does IV change, IV of each strike along the option chain and across different expirations change differently, thereby creating the vertical and horizontal IV skew shifts. Ability to forecast such IV skew shifts and its impact on an options position is one of the most critical task for options trading. Yet, there isn't an accurate way to predict IV that I know of, but if others disagree, please share your thoughts. However, although predicting IV is extremely difficult, my belief is that accurately modeling the IV of individual strikes given an underlying price movement is possible through machine learning.
    CC, garyw, Chaitanya and 1 other person like this.
  4. Jim N

    Jim N New Member

    I believe you have carried out extensive research/analysis on estimating/forecasting IV, so can you tell me if you have researched the approach of Dumas,Fleming and Whaley where volatility is regressed against strikes and time to expiry? I am about to download option chain data to perform an analysis of their approach, however if you have been down that road and formulated a conclusion as to the merits or demerits of their approach I would be grateful for your input. No sense in pursuing a lost cause. Thanks
  5. Kevin Lee

    Kevin Lee Well-Known Member

    Hi Jim,

    Sorry, I'm not familiar with their work. I googled and found their 1998 paper. Is that the one you're referring to ? I scanned through quickly. Seems interesting. Has bookmarked for further study. Thank you for highlighting.

    For whatever it's worth, here are my thoughts on using regression analysis on IV :

    Firstly, I have been down the road of trying to "patch" the inaccuracies of Black Scholes. Read many papers that attempts to do that theoretically. Personally, I feel that's going down a rat hole. Now, I'm moving towards a new approach, ie using machine learning to perform regression on individual underlying. I don't care about the mathematical formula any more. But instead one underlying, one regression and I'll be using machine learning algorithms. I think that will greatly simplify the problem statement because we only need to worry about getting it right for one single underlying and not having to generalize.

    In addition, there are some advancement in neural network algorithm which is very promising in market analysis. Let me give you one example. The IV skew change is not consistent, ie in the recent couple years, most of the time when market moves down, the IV skew flattens. But this relationship is not consistent. Some other times, the entire IV curve will move up vertically with little skew change, and in the remaining cases, IV skew will steepen. Clearly there are other factors influencing it. Therefore, if volatility is only regressed against strikes and time, the result will likely not be good. The different outcomes will be averaged instead of being split. We need additional features other than time and strike as well as some algorithm that regress based on what has recently happened.

    Recurrent Neural Network, for example, has the ability to learn from recent pricing history and incorporate that into the regression. You can select how long a history to remember and how to weight the influence. In addition, we can also easily add more features (eg. market sentiments, technical indicators etc..) as input into the network. I believe that will make the analysis more accurate.

    But... this is just my speculation. I don't have the answer yet because unfortunately, my attention is somewhere else for the next few months. For now, it's my educated guess. I hope to get to it by year end. But if you find anything with your approach or anyone in this forum has tried using ML on IV, please do share.
    Last edited: Aug 13, 2017
    Murphy Tan likes this.
  6. CC

    CC Active Member

    Hi Kevin,
    What ML package do you use? And what language do you code in?

    Sent from my iPhone using Tapatalk
  7. Kevin Lee

    Kevin Lee Well-Known Member

    Mostly sklearn and tensorflow in python
    CC likes this.
  8. ACS

    ACS Well-Known Member

    Since volatility reflects the emotions of an ever-changing population of participants to an ever-changing market environment, it seems there will never be a formula that is 100% accurate in its predictions. I wonder how accurate any formula can become and at some point do you have to say good enough?
  9. Marcas

    Marcas Well-Known Member

    It goes along the line: you can't predict precisely behavior of single individual but you can be very accurate in prediction of crowd's reaction. That comes with ability to manipulation. We are manipulated as a society and this shows in volatility behavior. What we can do to get some edge here? I distinct 3 approaches (ad hoc): experience, statistics and ESP (extrasensory perception).
    Experience - imo, good results but requires time and focus. Each of us uses experience in narrow field of consciously trading single strategy for longer period of time. It is usually limited to specific strikes and DTE's. Vast knowledge of this topic is lifetime quest. Is it worth?
    Statistics - ideal or close to ideal - if we can make it work. I'm very interested in Kevin's progress with machine learning. Kevin does results look promising?
    ESP . . . I remember scene from "Reminiscences of stocks operator' when author had impulse to sell some stock (Western Pacific?). He was selling it into uptrend and good lookup. He couldn't explain why he was doing it but felt strong impulse to sell. Few days later big earthquake proved him right. Take whatever you want, I'm not touching it.

    I think that thinking while trading is the way to go for retails.
  10. Kevin Lee

    Kevin Lee Well-Known Member

    Trading is not an exact science like Physics. 100% accuracy is never the goal. Markets are so complex and the rules are reflexive, like social sciences. The best we can do is to model them stochastically. Just like weather forecast, it's never perfect, but it doesn't mean it isn't useful.

    I'm still very early in the discovery journey. I'm prepared to spend the next 1 to 2 years just learning whatever I can. However, what I've seen so far is very exciting and promising. There are some low hanging fruits. IV forecast is one of them.

    ML is still relatively new in trading, especially options trading. Most of the progress thus far has been in other fields, such as translation, object recognition, human speech recognition etc.. A lot of new technologies and tools were developed in the past few years in these areas. Some of them have good potential to be adapted in trading, I think.

    15-20 years ago, retail options trading was still very new. It was made possible by internet technologies and then subsequent lowering of transaction cost. People who started then and committed the necessary 10,000 hours of hard work eventually became the veteran traders today. Another disruption is coming. This time brought by AI. I believe it will once again change how trading will be done and once again open up lots of new opportunities.

    However, I don't think AI will replace human traders. That is not the intent. Some people said AI will never think like human. That's true, but that is also not the intent. AI should not think like human but rather complement human thinking. AI can do things that human can't do and human can do things that AI can't. Therefore, the best solution is to collaborate. In the future, in many industries, it's not that AI that will replace human but rather human who partner with AI who will replace those who don't. Instead of calling AI, artificial intelligence, perhaps it's more appropriate to call it "Augmented Intelligence".
    ralphwindsor likes this.
  11. Paul Demers

    Paul Demers Well-Known Member

  12. Kevin Lee

    Kevin Lee Well-Known Member


    You're right. I take that back. There are certain type of trading, such as HFT, where machine has a unsurmoutable advantage over human.

    I was thinking more from an options trading perspective. Where the edge isn't in speed and there is some element of human judgement that is required. In this kind of situation, a human / AI collaboration is a reasonable model. But again, if one just trade by following a trade plan, I suppose that can be fully automated too. With reinforcement learning, AI can even develop their own trade plan without human intervention.

    Good point.
  13. Paul Demers

    Paul Demers Well-Known Member

    Some food for thought.

    My thinking is that even though all these computer programs are trading stock (HFT, Algo trading, or AI portfolio management) have a direct influence in the options trading world because the options are just a derivative of those stocks. What if with all these computers that have no fear or greed the volatility gradually continues lower. These computers are so fast that there is no need to hedge as they will just liquidate their positions in milliseconds and step aside. Will the option become more affected by delta changes rather than volatility changes as more and more computers trade the markets and less humans?

    I do not know the answer to that question.
  14. Kevin Lee

    Kevin Lee Well-Known Member

    Interesting thought. So far, options volume is still trending up.


    I think one key information needed is what's the split between options being used as hedge vs as speculative/leverage tools and how's that trending. That will drive different demand profile in the future and influence the answer to your question.

    Attached Files:

  15. onyxperidot

    onyxperidot Well-Known Member

    “If you've been playing poker for half an hour and you still don't know who the patsy is, you're the patsy.”

    Warren Buffett

    Just a random idle conjecture: Speculation in financial markets shares many similarities to an arms race with a dangerous prey as examined in this paper, https://goo.gl/sxMpfx

    A more realistic model would be an arms race with many different dangerous preys. But that was too complex.
    Last edited: Aug 16, 2017
    Paul Demers likes this.
  16. Marcas

    Marcas Well-Known Member

    Kevin, thanks a lot for sharing experience. I got an answer I was afraid of. I was afraid because it means another project on my to-do list. I wish I could jump in right now but I have to finish more basic stuff.
    Kevin, you said you like to work with Py. Do you have any suggestions about GUI? So far I'm working with Jupyter (iPython) alone and I think it is great but I'm at point that simple interface would be useful. I think I'd prefer something browser based. Do you have any advice?

    About break through. Well, one may think when AI takes over all trading is done. I don't think so, market will adopt. As a byproduct, or collateral rather, of this process may be that retail trading will become history as it happened with horses and combustion engines and many other things in past. But even here I'm not pessimist. True, market will change, probably become more chaotic (for human eye), but trading longer term I really don't care much how many billions of trades HTFs placed between my entry and exit. Even if I have to give up on execution I might be ok. Definitely strategies will change. I think that simple IV metrics won't be sufficient to provide enough data for recognizing state market is in, we may need something to deal with periods of very low vol mixed with bursts of high vol. Or may need something quite different - my hope is that we will be able to develop smth to deal with situation. So, it may be viewed through biological arms race model, to me though it is bit overcomplication.

    Paul, if your scenario will play out it wont be to bad. I could place very wide trade and wont worry about huge neg vega : )

    As per being pastry (I didn't know it was Buffet's). Anyway there is simple way to figure out if you are pastry or not - monitoring p/l (over long period).
  17. Marcas

    Marcas Well-Known Member

    Just come across piece that is relevant to our talks about future of markets. Sorry for quoting but piece is rather long and I'd like to post only relevant part as 'some food for thought'.
    It is, again, Ann Barnhardt presenting alternative future for algos. Realistic? Probably not this moment but . . .

  18. Paul Demers

    Paul Demers Well-Known Member

    Is there a date on when that statement was made?
  19. Marcas

    Marcas Well-Known Member

    Today or yesterday.
  20. Kevin Lee

    Kevin Lee Well-Known Member

    Python based web development is quite popular nowadays. For me I don't really care about GUI. Jupyter is great. I do most of my prototyping on Jupyter and for more complex stuff I do it on PyCharm (it's free too). Much easier to debug than Jupyter.

Share This Page

  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice