Kelly formula is just for gamblers

Discussion in 'Trading' started by ronblack, Jul 11, 2007.

  1. ronblack

    ronblack

    You did not have to ask this question but the answer is no.

    Ron
     
    #11     Jul 12, 2007
  2. gbos

    gbos

    Each position may be 1/100th of Kopt. but what about the entire portfolio. The positions of a fund are correlated. If the aggregate portfolio has 10% mean and 20% standard deviation then putting 80% of your assets in the stock market means that you have allocated 1/3 Kopt.
     
    #12     Jul 12, 2007
  3. u21c3f6

    u21c3f6

    f = (bp - q)/b

    where

    f is the fraction of the current bankroll to wager;
    b is the odds received on the wager;
    p is the probability of winning;
    q is the probability of losing, which is 1 − p.

    The above is the Kelly formula. It is basically your edge divided by your odds. The formula you are trying to use does not include a calculation for the edge.

    The Kelly formula is perfect by the way. The problem is that it is often difficult to know the edge of a particular event. Those that use Kelly tend to use a fractional Kelly to stay on the safe side and not suffer "gambler's ruin".

    Joe.
     
    #13     Jul 12, 2007
  4. gbos

    gbos

    The formula he used is correct. Its equivalent with the one you wrote.

    The problem with the ‘high’ 0.6 result is in the input data and the simplification of the problem. Let’s take a slightly different and more realistic assumption and see if Kelly formula gives unreasonable results.

    (Probability of success) 79% … wining 1R
    (Probability of failure) 19% … losing -1R
    (Probability of something going seriously wrong) 2% … losing -5R

    The problem is almost equivalent with the one of the OP but now a third option is added. Kelly formula becomes a little more complicated than the previous 2-case one but the result now is 0.15 instead of 0.60.

    So it’s not that the formula gives unreasonable results, it’s that by taking the 2-case win/lose approach we ignore black swans etc. and it is those events that govern the calculation of the optimal risk fraction. First step to calculate a meaningful Kelly fraction is define your worst case scenario and add it to your calculations.
     
    #14     Jul 12, 2007
  5. Search wilmott.com; this subject has been heavily discussed there. Half-Kelly betting works well!
     
    #15     Jul 12, 2007
  6. Murray Ruggiero

    Murray Ruggiero Sponsor

    #16     Jul 12, 2007
  7. ronblack

    ronblack

    gbos, things for me are way simpler than that. Example:

    MSFT is trading at around $29.50 at this moment. Let's say I have 100k equity in my account and my system says to go long for $4 profit/loss.

    Also, my system has a 65% actual success rate with R =1. According to Kelly formula the optimal bet size is:

    k = .6 - [1 - .65]/1 = 0.3

    If I invest 30% of 100k in MSFT I must buy 100,000/29.50 shares , or 1,000 shares.

    My risk if the position goes against me is then $4,000. This is 4% of the account size. It is way too high. For normal 2% risk I should be buying only 500 shares.

    Ron
     
    #17     Jul 12, 2007
  8. u21c3f6

    u21c3f6

    "My risk if the position goes against me is then $4,000. This is 4% of the account size. It is way too high. For normal 2% risk I should be buying only 500 shares."

    How can you guarantee that your loss will only be $4,000? The hard part in using Kelly is "knowing" the "true" edge. As gbos correctly points out, dependent upon your set-up, you might need to add something for a black swan event.

    IMO, it works better with spreads or other set-ups that have a more defined gain or loss structure and even then I would recommend a half-Kelly.

    Joe.
     
    #18     Jul 12, 2007
  9. maxpi

    maxpi

    I recommend the 2 kelly method. I call that the turbo mode. Always have enough money to fund your account after a blowout. I have not been able to calculate the odds of blowing out twice in succession so far but I'll work on it eventually, I hope I can get it figured out before a double swan happens to me.

    I can actually generate my own swans nowadays, I make so many order entry mistakes it is unreal, maybe I could serve as an example of the multi swan effect on portfolio management. I'll work up a video for youtube.
     
    #19     Jul 12, 2007
  10. ronblack

    ronblack

    No guarantee but this is the stop loss (4 points x 1,000 shares) Thus, under normal market conditions it is the maximum position risk.

    Ron
     
    #20     Jul 12, 2007