Reconciling Kelly Criterion derivatives

There are various derivatives of the Kelly Criterion, depending on the context  (e.g. Betting, Investing, other)

I'm trying to reconcile the different approaches to get a Kelly Criterion for a case where:

  • There's a 55% chance of winning, 45% chance of losing.  
  • A win will result in 35% increase (e.g. 100−>100−>135).
  • A loss will result in a 30% loss (100−>100−>70).

Equation 1. Generally: all variations seem to derived from:
$ f^* = \frac{(bp−q)}{b} $

p: probability of a win
q: probability of a loss
b: a win of "b" units for every unit waged

Equation 2. Per this:

$ f^* = \frac{(p \times \text{win} - q \times \text{loss})}{win} $

Equation 3. From Wikipedia, per Thorp (page 7, lines 25-28)
$ f^* = \frac{bp − aq}{ab} = \frac{p}{a} - \frac{q}{b} $

where a unit wager wins "b" and loses "a"

The issue is that these equations seem to not properly reconcile, and leave me uncertain what the correct formula is. I'd like to understand why, ultimately get to a formula to confidently use.

Notice for the initial example I posed:

Table 1: output of equations for above posed case

Eq # p q a b $f^*$
1 55% 45% n/a

$\frac{.35}{.30}$ per this

(I'm not certain if correctly captures partial losses)

$ \frac{\frac{0.35}{0.30} \times 0.55-0.45}{\frac{.35}{.30}} = 0.164 $
2 55% 45% 0.30 0.35 $ \frac{(0.55 \times \text{0.35} - 0.45 \times \text{0.30})}{0.35} = 0.164$
3 55% 45% 0.30 (uncertain if these are correct values given the case, see table 3) 0.35 (uncertain if these are correct values given the case, see table 3) $ \frac{0.55}{0.30} - \frac{0.45}{0.35} = 0.547 $

If we assume a full loss however, everything matches up.

Table 2: adjustment to posed case, to show a full loss reconciles equations

Eq # p q a b $f^*$
1 55% 45% n/a $\frac{.35}{1}$ -0.7357143
2 55% 45% 1 0.35 -0.7357143
3 55% 45% 1 0.35 -0.7357143

Further adding to confusion, Wikipedia had a revision (see under "Investment formula", which I've for now corrected on the latest revision) indicating that the values of "a" and "b" for the formula of $ f^* = \frac{bp − aq}{ab} = \frac{p}{a} - \frac{q}{b} $ be determined by adding a 1 to the percentage price chanages. Thus this previous revision indicates:

Table 3: wikipedia revision suggests a, and b should be derived differently

Eq # p q a b $f^*$
1 55% 45% n/a $\frac{.35}{.30}$ $ \frac{\frac{0.35}{0.30} \times 0.55-0.45}{\frac{.35}{.30}} = 0.164 $
2 55% 45% 0.30 0.35 $ \frac{(0.55 \times \text{0.35} - 0.45 \times \text{0.30})}{0.35} = 0.164$
3 55% 45% 1.30 1.35 $ \frac{0.55}{1.30} - \frac{0.45}{1.35} = 0.089 $


Answers can be viewed only if
  1. The questioner was satisfied and accepted the answer, or
  2. The answer was disputed, but the judge evaluated it as 100% correct.
View the answer
Blue Blue
  • Thanks. Can you show the translation of Eq1 into Eq3?

    • Just the clarify my question, I understand eq 1 is eq3 if a=1. But my understanding is that the original kelly was focused on complete loss. At some point the partial loss was introduced. Trying to understand how that partial loss derivative was incorporated given that original kelly "b" was all or nothing.

  • @Blue: Please respond to Dforootan comment above.

  • Blue Blue

    The Kelly Criterion is more general than the complete loss case. You start with the partial loss and derive the complete loss formula, not the other way around.

    • My understanding is that the original Kelly Criterion, from the paper (pdf below) only has full loss per, αs, not partial, is that incorrect?

    • Blue Blue

      Sure, but there exists another proof for the more general result, why you can find sketched and cited on the Wikipedia page you linked earlier. The historical order of the results need not affect which one is the more general criterion.

    • Oh, I see. Could you please share that proof? there's a lot in the wikipedia and Thorton paper.

  • Blue Blue

  • Thank you for your help!

The answer is accepted.
Join Matchmaticians Affiliate Marketing Program to earn up to 50% commission on every question your affiliated users ask or answer.