Kelly sizing...

Discussion in 'Trading' started by Eight, Sep 6, 2010.

  1. Eight


    say I have a strategy and Kelly sizing indicates that I should allocate 80% of capital on it. Intraday my account can margin to 400% of capitol... and I have five instruments that I'm trading, can I safely put the whole account on those five instruments at the same time, equal amounts allocated to each, or does that violate the ideas inherent in Kelly's formula?
  2. drm7


    In theory, you could put more than 80% into five uncorrelated strategies, since you would get some diversification benefit.

    In practice, trading at full Kelly for even one strategy is suicidal, since your "edge" is not guaranteed to be stable over time, and the future data you base your trading on will not be the same as the data you tested on.

    I would look at one-half or one-quarter Kelly to start.
  3. drm7


    I would add that, should each of your five strategies have significant correlation with each other (say, a long trade on both GBP/USD and EUR/USD, or long Bank of America and JP Morgan), then you need to cut your bet size on each trade, since you would be essentially doubling your risk.

    I still think full Kelly is too much.
  4. Murray Ruggiero

    Murray Ruggiero Vendor

    Kelly is not valid for trading systems because average win value does not normally equal average loss value. This was why optimal f was developed by Ralph Vince. In addition these calculations are made to work on one market at a time. The standard calculations are not valid for portfiolo. Vince did develop a portfiolo version of his calculation which was discussed in his second book.

    On my TradersStudio site I have a nice tutorial on optimal f.
  5. charts


    You may find some interesting discussions here: Money Management ... :)
  6. Kelly is for loonies. You are not serious to be asking whether you should use it. Regardless, win percent and avgwin/avgloss ratio are not stationary random variables. Which value to you? My answer is a loud NO. Even if your instruments are not correlated. If there is correlation you will be optimaly screwed. This is a good paper about Kelly and its practical pitfalls. Read the section "APPLICATION OF THE KELLY FORMULA":
  7. Kelly's criterion formula takes the average win and average loss into consideration. They don't have to be equal.
  8. Eight


    moving this discussion to money management might be a good thing...

    it would be good to try to hedge long and short trades or correlated vs not but I'm just interested in the use of the kelly math... a little further reading suggests to me that kelly is about one bet at a time but i can't come to grips with whether a bet could be an "aggregated bet"... I'm thinking that probably all simultaneous bets would be the same as any single bet for kelly purposes...
  9. u21c3f6


    First, I would check my calculations and/or assumptions again before I would assume that 80% Kelly was the correct answer.

    I also agree with using half-Kelly (which I use) as it gives you protection against overallocating (a very bad thing with Kelly) and also cuts down on the volatility and drawdown of your bankroll while maintaing about 75% of the profit potential of the optimum or full-Kelly.

  10. empirical


    Murray is right, Kelly only applies if there are only 2 outcomes,1)win amount is the same, and so is 2) the loss which does not apply to trading.

    Consider optimal f instead, (read the mathematics of money management by Ralph Vince)
    Did anyone test it out before discrediting it?
    #10     Sep 18, 2010