Predicting HV - Thoughts on a Trading Model

Discussion in 'Options' started by neuralmarket, Sep 11, 2007.

  1. Hi All,

    I'm a relatively newbie to the world of options but I have a strong mathematical background.

    I've built a pattern recognition neural net model to help me predict the direction of next week's HV in the QQQQ's an SPX. Its accuracy is between 60-70%.

    How do I translate this into a trading system that minimizes risk? Is something like this useful at all?

    I've been fooling around with OptionsXpress virtual account using volatility strategies from Natenberg's book (straddles/strangles) which isn't panning out all that well.

    Any help or thoughts on the matter would be of great help!

  2. Tom - how are you measuring HV? The annualized SD over the last X periods on a rolling basis? If so, what is X?

    You have to be careful that you are not predicting some element of the past. For example, if you measure HV over the last 20 trading days, and are predicting it 5 days ahead, your neural net will appear to have very good results because it is already incorporating 15 out of 20 days in the prediction.

    But if you could predict the realized 5-day volatility, 5 days from now, that would be very useful information. You could use it to trade options vs the underlying. Example - current IV is 30% but your predicted 5-day HV (5 days from now) is 15%. That means a quieter period for the actual market than the options are predicting. You would take advantage of that by selling straddles/strangles (say) and delta-hedging them for 5 days using the underlying. If the model is correct and the market is quiet, you'll lose less money delta hedging than you'll gain in time decay.

  3. Yes I'm using 5 day realized HV and then predicting one week (5 days) ahead. The model is re-optimized on a weekly basis because volatility, is well, volatile. :)

    Ah so in your example then, I could sell a straddle and then delta hedge the underlying.
  4. You should be very careful trading HV/IV alphas ( short or long) . Besides , how did you got 60-70% of accuracy ? you can have overall losses if only ONE stock out of ten ( when shorting) moves big against you , while still maintaining 90% of "accuracy"
    Good luck
  5. IV, there in lies my quandry. My accuracy has been from taking samples from my data and overlaying the model on top of it.

    From those samples, the directional predictions of HV week (Up or Down) have been forecasted one week forward and has been correct 60-70% of the time.

    In my virtual trading experiment , I've seen the model prediction be correct (say it predicted down) and the my short straddle get crushed because the underlying asset's price simply drifted lower than the strike price of the straddle.

    It doesn't do me any good if I can predict the direction of HV 1 week forward and be right 60-70% of the time if I lose money 100% of the time.

    How do I use this and build a solid trading system to minimize my risks and maximize any gains?
  6. in declining future HV environment ( and hopefully IV to follow) I would probably trade flies. Search on riskarb fly journal
  7. Do you mean Butterfly trades or something else?
  8. With your math background, you must know about variance swaps? With strikes at each dollar and penny spreads like on QQQQ this may actually work out. I haven’t analyzed it, but it is very interesting.
  9. GTG


    Are you dynamically hedging the deltas on your straddles in your simulation?
  10. sjfan


    Chances are he can't trade variance swaps. They are OTC.

    That being said, 60-70% accurate is a warning sign. You need to perform a binomial validation to make sure that your "accuracy" can't simply be accepted as a random realization of 50/50 probability because of small sample distortions (christ - I sound like Jack Hershey. Thank god what I write should actually make sense).
    #10     Sep 11, 2007