18 January 2010
Simulations show full Kelly betting wins large gains most of the time over long horizons but is very risky short term, and bad runs can still lose most capital.
Leonard MacLean, Edward O. Thorp, Yonggan Zhao and William Ziemba simulate the full Kelly and fractional Kelly betting strategies over medium horizons in three cases: the Ziemba-Hausch betting model and two cases from Bicksler and Thorp (1973), one with uniform returns and one with returns like those of US stocks. The results show the superiority of full and close to full Kelly strategies over the long run, the high risk of Kelly and high fractional Kelly strategies in the short run, a consistent trade-off of growth against security as the bet size changes, and that a sequence of bad results can lose most of an investor's capital however favorable the odds.