construction and use of adaptive averages

Discussion in 'Technical Analysis' started by mind, May 13, 2004.

  1. mind


    a while ago i heard about adaptive averages, like jurik, tilson et al. i did not study teh subject in depth so far, but i have a starting point i would like to discuss in principal and in terms of usability.

    in principal, i currently use the following concept.
    1. take the last eleven close prices and derive the ten percentage changes.
    2. calculate the mean percentage change (mpc)
    3. take the last percentage change (lpc) and calculate its likelihood if you assume a normal distribution with a mean of the value derived under 2. and a standard deviation of 1. call this P(lpc,mpc,1)
    4. calculate (1-P(lpc,mpc,1))

    now you have a figure between 0 and 1, which is nearer to 0 if the last percentage change was very near the mean percentage change of the last ten days and which increases towards 1 if the last percentage change is far of the mean of the last ten days.

    5. define adaptiveCoefficient as a positive number
    6. multiply it by (1-P(lpc,mpc,1)) in order to derive numbers greater than 1 for bigger adaptiveCoefficients. calls this product final adaptive coefficient (fac).

    now we have a number that can vary between 0 and the adaptiveCoefficient

    6. define E as the exponent for an exponential moving average.
    7. calculate a daily simple EMA using E
    8. calculate and adaptive EMA by multiplying E with fac, followed by the normal EMA procedure.

    it is just a thought that i developed in a couple of minutes in excel. there are several weaknesses, like the use of a normal distribution which flattens out at the tails, while we might like a different feature for this purpose. and the adaptive coefficient is a little strange as well. but i think this must be somehow the basic concept of adaptive averages. if this is true, now my question. do people here have experience with using them for bollinger bands for example? or macd or stuff of that kind?

  2. mmillar


    I have been looking at using Adaptive MA's in trend following strategies. I've looked at Kaufman's KAMA, Chande's VIDYA and Ehlers' MAMA and compared them against Simple, Exponential and Weighted MAs.

    There wasn't much in it but if I had to choose the 'best' it would be an Exponential MA.

    Obviously only relevent to the particular strategies I was testing.
  3. rognvald


    Do you yet have any chart examples of the result of applying your idea?

    The three problems with ama's seem to be Smoothing, Phase lag and Overshoot

    This article from TASC may be of interest
    Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson
  4. Here's a thread from last week where I was asking about using adaptive averages in Excel:

    A couple of members were nice enough to post some spreadsheets and information. Do you have a copy of your formula that you could post?

    My intent is to make a spreadsheet with an upper and lower moving average. In an uptrend I would send a buy signal on a touch of the lower moving average and sell when it hits the upper average. To determine the trend I would use a calculation of the slope of a longer term moving average.
  5. MartinD


  6. rognvald


  7. mind


    i was afraid, someone would write this post. nevertheless, thanks mmillar for sharing your experience.
  8. mind


    i will come back with diagrams and formulas. i am currently a little afraid of the postSpike effect. i think i would have more vola in the indicator after a spike.
  9. mind


    i will post a sheet. just do not have it with me. tomorrow.
  10. mind


    i will try to address all issues. i thank everybody for participating and the useful comments.
    #10     May 13, 2004