kelly wagering

Discussion in 'Strategy Development' started by saultnutz, Jul 17, 2006.

  1. as i understand it, the basic concept of kelly is to bet the % of your bankroll equal to your percent chance of winning

    ie if you have a ten percent chance to win, bet ten percent of your roll. but what do you do if your win amount is bigger than your loss amount

    ie you have a ten percent chance of winning, but youw wins are 4x your losers. do you then bet 2.5%?


    i posted this in the tradestation forums, and only got smartass non answers - i hope ppl here are smarter
     
  2. Hello

    There is a long history of literature on the Kelly System available to you in print, on the net, and from various sites.

    You can find some here, some at Wilmott and some in print from an old friend named Bill Ziemba.

    You comment is correct in principle but is only one aspect of the core kelly system. If I get a moment later I will add more information

    Good luck
    Steve

    Edit:

    On a related subject, Dr. Ziemba has written a book that I really like called "Beat the Racetrack". If you have a chance to skim through you might like it.
     
  3. Luenberger's book "Investment Science" has a chapter on this.
     
  4. thx - the article he posted was perfect
     
  5. Please don't take this as a smartass answer, but if you have a 10% chance of winning only 4 units of profit and a 90% chance of losing 1 unit, then don't take the trade! You will lose an average of 0.5 unit on every trade and you will most certainly go bankrupt.

    If only 10% of the trades generate a profit, that profit on a win had better be at least 9X the amount you lose on the other 90% of trades. This is breakeven ratio -- you need an even higher amount per win (or a better % of winners) if you want to avoid ruin.

    As far as the Kelly system is concerned, the size of the win is less important. Allocation schemes such as Kelly seek to reduce the chance of going bankrupt. Thus, they are more concerned about the potential for numerous consecutive losses. With consecutive losses, the amount of the win is irrelevant because no wins occur during a prolonged losing streak. For example, with only a 10% chance of a win, you face a 35% chance of 10 consecutive losses, and a 12% chance of 20 consecutive losses. Thus is an average year of making one trade per day, you face a very high chance (more than 80%) of seeing a 20-consecutive-loss streak.

    In intention of something like Kelly is to ensure that you survive the losing streaks to be there for the wins. If you know that long streaks of losses can easily occur, you know you need to lower the amount risked on each trade regardless of the potential size of the win.
     
  6. good point - i just pulled those numbers out of my azz
     
  7. maxpi

    maxpi

    Kelly came up with a simple formula for how much to bet if what you know is the ratio of winners to losers and the ratio of win percent to lose percent. Kelly sizing is pretty aggressive but it supposedly optimizes risk and gain. Some commentators have called it oversimplified compared to the realities of trading.
     
  8. rosy

    rosy

    i think the kell criterion works well for a paramutuel betting system where who know the odds are off. you still need the edge somehwere.
     
    #10     Jul 17, 2006