Here is a quote from Leonard C. MacLean† , Edward O. Thorp‡ , Yonggan Zhao§and William T. Ziemba¶:

But how do full Kelly and fractional Kelly strategies that blend with cash actually preform in practice?

To investigate this we revisit three simple investment situations and simulate the behavior of these strategies over medium term horizons using a large number of scenarios. These examples are from Bicksler and Thorp (1973) and Ziemba and Hausch (1986) and we consider many more scenarios and strategies.

The results show:

1. the great superiority of full Kelly and close to full Kelly strategies over longer horizons with very large gains a large fraction of the time;

2. that the short term performance of Kelly and high fractional Kelly strategies is very risky;

3. that there is a consistent tradeoff of growth versus security as a function of the bet size determined by the various strategies;

4. that no matter how favorable the investment opportunities are or how long the finite horizon is, a sequence of bad scenarios can lead to very poor final wealth outcomes, with a loss of most of the investor’s initial capital.

Hence, in practice, financial engineering is important to deal with the short term volatility and long run situations with a sequence of bad scenarios. But properly used, the strategy has much to commend it, especially in trading with many repeated investments.Found here:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.225.4036&rep=rep1&type=pdfAn interesting strategy that has just occurred to me while reading this excellent paper, is to earn a "Kelly Bankroll" from profit and then attempt to use it to achieve "Kelly Wealth

^{TM}", and then go back to regular betting and repeat. O_o

Maybe this is what they mean by "financial engineering"?

I definitely am a fan of the full Kelly because the divergence from the fractional Kelly's in total profit is truly immense!

Another excellent paper by some of the same authors:

http://onlinelibrary.wiley.com/doi/10.1002/9781119206095.app1/pdfThis quote is gold:

Logs abound in information theory and Kelly argued that maximizing the expected log of final wealth was a good idea. The idea of using log as a utility function was not new to Kelly and dates at least to Daniel Bernoulli in 1732. But Kelly, in an ad hoc math way, showed that it had good properties. Later Breiman (1960, 1961) cleaned up the math and showed the great long run properties:

1. Maximizing E log maximizes the rate of asset growth asymptotically, and

2. it minimizes the time to reach arbitrarily large goals.

This means that a log bettor, who is in competition with another bettor who bets differently infinitely often, will have arbitrarily more money than the other bettor as time goes to infinity.

So the longer you play, the better log is, but we know from Chapter 4 that in the short run, log betting is extremely risky. Indeed the Ziemba-Hausch (1986) example discussed in Chapter 2 shows that you can make 700 bets on assets with a 14 % advantage, all independent, all with chance of winning of 0.19 to 0.57 and turn $ 1000 into $ 18.

This is part of the Merton, Samuelson critique. Even if you play a long time and have a good advantage on every bet, you can still lose a lot.

But, Kelly advocates like I am point to the great gains most of the time and the corrective action that you can take should you have a sequence of bad scenarios and the fact that in practice trading is financial engineering not pure financial economics. So my version of "financial engineering" is to obtain 3x one's table bank and then use 2 full table banks for Full Kelly.

This will allow me to start doubling my bets much earlier and then allow me to incrementally raise my bets as by "Kelly Bankroll

^{TM}" grows.

I currently already double & triple my bets (and even higher) but none of my betting is based on bankroll.