Money Management

Discussion in 'Risk Management' started by cnms2, Nov 22, 2005.

  1. gbos

    gbos

    Hello I would like to add some comments to the above…

    The formula is not simple at all. Only if you assume a Bernoulli distribution for your trades it takes this simple form.

    Then why are we trading?

    Uncertain but quantifiable within limits. Nothing is 100% certain but you can use your data to take an estimation for a range of your edge.

    Correct.

    Wrong. Kelly optimizes for the median wealth. Taking additional risk will optimize your average wealth. The problem is that rare events are the main contributors for this average wealth so in the vast majority of scenarios what you are interested in is the median (most probable scenario). Additional risk doesn’t always compensate with extra return but on average and I explained from where this average originates.

    Wrong. It is applicable in every case. An example of a ‘system’ with 3 possible outcomes…

    R-multiple Probability
    -1 ----- 1%
    -0.05 ----- 10%
    +0.13 ----- 89%

    1/10 Kelly = 9%

    Monte Carlo of Wealth after 100 trades (median 248, starting capital 100)

    [​IMG]

    Again for anyone interested how Kelly is implemented in the real world, read Ed Thorp’s paper. He has managed billions of dollars for decades using Kelly and is one of the best mathematicians around.

    Regards
     
    #161     Dec 20, 2005
  2. Those input are other people's insights/ ideas, not mine. I'm still learning.
     
    #162     Dec 20, 2005
  3. ES335

    ES335

    For someone who is "still learning" you sure have a lot of opinions...

    (sigh)
     
    #163     Dec 20, 2005
  4. Thanks for pointing us to the generalizations, variations and practical aspects of applying Kelly principle, rather than a static Kelly value. :)
     
    #164     Dec 20, 2005
  5. Murray Ruggiero

    Murray Ruggiero Sponsor

    #165     Dec 20, 2005
  6. Murray Ruggiero

    Murray Ruggiero Sponsor

    #166     Dec 20, 2005
  7. Would you mind to point out anything wrong with the opinions, so that I could learn more from this borad (that's why I come here)? Thanks in advance! :)
     
    #167     Dec 20, 2005
  8. gbos

    gbos

    :) This was an answer to

    If you don't like the static framework there are papers on the net for a Bayesian dynamic updating of the Kelly optimal allocation.

    Anyway no-one can force you to accept as useful the idea of optimizing the median of your payoffs. I am convinced that this is the best way and I use it to size my positions.

    Regards
     
    #168     Dec 20, 2005
  9. When you mention "1/10 Kelly = 9%", are you implying Kelly = 90%? Why use 1/10? Are you using 1/10 of Kelly all the times? I'm just curious, and wondering. TIA.
     
    #169     Dec 20, 2005
  10. gbos

    gbos

    Yes I imply that in this example Kelly=90% so 1/10th Kelly is 9%. The fraction of Kelly you are going to use depends on many factors. First is how tolerant are you to drawdowns. For example there is 34% chance that you are halved before doubled if you are using 1 (full) Kelly.

    [​IMG]

    Other considerations are how uncertain you are about your system statistics. Uncertainties in system’s expectancy, variance etc. must be taken into account before making a meaningful choice.

    Also you need to make considerations for black swans by assuming a small but non zero probability to a very damaging event. This also must make you more conservative in choosing the appropriate Kelly fraction.
     
    #170     Dec 20, 2005