Volatility Filter

Discussion in 'Strategy Development' started by dima777, Sep 30, 2008.

  1. dima777


    I wonder if you are using volatility filter in your systems. This is very important issue give the current price action. Please let me know what you think on this subject.

    Here is a quote from the Market Wizards:

    You mentioned before that you used increased volatility as a signal
    to stop trading a market. How many days of past data do you use to
    determine your volatility filter?

    Anywhere from ten to 100 days.

    When you say ten to 100, are you trying to be deliberately ambiguous
    or do you mean you use different time frames within that range?

    We look at different time windows in that range

  2. MGJ


    Some people, including me, use betsizing algorithms that cause position size @ entry to vary inversely* with "volatility". Higher "volatility" means smaller position @ entry; lower "volatility" means larger position @ entry.

    "Volatility" could be some kind of average true range, or it could be some sort of width-of-Bollinger-Band calculation (i.e. a standard deviation), or it could be some kind of width-of-Donchian-Channel calculation (i.e. HH(n) minus LL(n), the n-bar Hi-Lo Range), depending on the system.

    When using this approach, the scariest environment is when volatility is LOW. You could put on a huge position and then get creamed by an anomalous gap (or a sequence of limit moves against your position). HIGH volatility isn't so bad because your position is dinky.

    *approximately inversely
  3. I believe you are talking about tom basso from market wizards right????

    I think the idea behind scaling out of positions that have too much volatility was to reduce the variability in returns (smooth out the equity curve). I think this concept would cut some winners short & reduce overall gains, but make them easier to handle. This may have some merit if you have outside investors & they get angry about more volatile returns, even if they are higher on average.

    I personally like to be in markets that are moving and appear to be volatile based on some simple math. Thats where I make most of my money.

    I size positions by setting recent volatility for a market equal to a desired % risk of the account balance. Basically this is the same as what the last poster said.

  4. dima777


    thank you guys for your detailed answers....that helped! BTW that quote was from the interview with Larry Hite..I am myself gravitating towards more volatile and more profitable trades as opposed to less volatile and more even ones..one more quote:

    Standard Deviation Measurement A standard deviation was used to determine the volatility threshold level. For example, for a highvolatility filter, a 1-standard deviation threshold means that no trades were taken if the volatility was above the average volatility plus 1 standard deviation, the top 16%. A 2-standard deviation threshold puts volatility in the top 2.5%, and a 3-standard deviation threshold means the top .5%. A 0 standard deviation would filter all trades above the average volatility. Because only 35 days are used, actual volatility can jump well beyond the normal 3-standard deviation maximum. Filter values above 3 standard deviations (up to 7 standard deviations in these tests) must be used to isolate the most extreme volatile price movements.

    I think the last filter can be useful in filtering out extremely volatile times - using filter values above 3 standard deviations...

    Frankly i am more concerned about non-directional volatility and love to see more of directional one...
  5. I remember writing a trading simulation program that could change position size exponentially as a function of volatility. For example, say I am using average true range (ATR) to measure volatility. I use a parameter that I call PositionSizeExponent. I raise ATR to the PositionSizeExponent power. If volatility increases then position size values can become much lesser or greater depending on the value of PositionSizeExponent.

    The system is a Donchian price breakout system.

    I remember conducting trading simulation studies using a few stock daily price histories. My records show at least one of those studies used 15.19 years of historic price data.

    I recall different PositionSizeExponent values show about the same trading results as judged by growth rate, greatest draw down and information ratio.

    I do not remember finding an advantage to trading this way. I do not now research this area.

    I suspect high price value volatility sometimes associates with a big trend, sometimes not, sometimes increasing price trends, sometimes decreasing price trends. I do not recall finding a consistent relationship between price volatility and trend.

    My research in this case is limited to stocks only. I do not remember testing futures prices or currency exchange rate histories.
  6. How do you intend to differentiate between directional & non-directional volatility?

  7. dima777


    directional volatility is mostly seen during the trends while non-directional volatility - when prices jump up and down - as can be seen on some of the last stock action - is better seen in the range-bound markets...I do not have anything against the volatility as such....nice, energetic, volatility-crazy impulse waves are all what the elliott theory is about....and I would love to see more of))
  8. dima777


    why did you use exponential coefficient to your position sizing algorithm instead of the simple multiplier?
  9. I do not use this system for real trading. For real trading I use a simple multiplier times range.

    I remember wondering about volatility: What happens if I take very large position size when price volatility increases? What if I reduce position size during times of increased volatility? With exponential volatility based position sizing I can use the square of ATR, or the square root of ATR to calculate position size. This experiment tests nonlinear position sizing.

    To my surprise I found that the growth rate, greatest draw down and information ratio do not change much from experiment to experiment.

    I remember testing only a few stocks; futures or currencies may test differently.

    These simulations model long term position trades, not day trading.
  10. That sounds about right, the information ratio shouldn't change as the info coefficient and breadth-per-unit-time wouldn't change either, or are you interpreting this differently from Grinold's classic paper? Remember position size doesn't affect breadth, since these are meant to be independent trades.
    #10     Oct 2, 2008