From Fortune’s Formula about betting on horses:
“Kelly’s gambler …places bets according to his best informed estimates of the probabilities. When you believe that War Admiral has a 24 percent chance of winning, you should put 24 percent of your capital on War Admiral. This approach has come to be called “betting your beliefs.” In the long run, “bet your beliefs” will earn you the maximum possible compound return—provided that your assessment of the odds is more accurate than the public’s.”
Usually, for stocks, this does not work because the payouts and odds are not like a racetrack. But can stocks be adapted?
I think they can. Suppose you took SPY, GLD and TLT and you measured the returns in EXCESS OF THE MEAN!!!
Then the payouts are similar to the racetrack (with no rake) where the wins and losses sum to zero. One difference is that you generally do not lose all of you money on the days that TLT underperforms SPY and GLD. But this COULD be addressed with appropriate leverage.
So one could “Bet their Beliefs” under this scenario and be doing something similar (in many ways) to Kelly betting. The differences are important and can be expanded upon. Most can be addressed, I think. For example, leverage could be used to bring this to fully Kelly Betting (which I would not recommend).
So how does one form their beliefs? I think I will look at this: Algorithm for Finding the Best Trading Strategy Quickly
This example of Thompson Sampling is actually a Bayesian method with a uniform prior that anyone skilled at Python could program. It can be expanded to include a prior belief about how SPY, GLD and TLT will perform as well as any recent data to form the posterior belief.
In other words: form a posterior belief with Bayesian statistics then Bet Your Beliefs using Kelly Betting.
This way, one could fuse Bayesian Analysis and Fractional Kelly Betting into an easy algorithm that can be done on a laptop.
So this actually gets to the best Fractional Kelly without leverage. Or the best compounding (Relative to the Mean of these ETFs) without leverage at a minimum drawdown risk. So it should get you the best compound returns (relative to the mean) with the least drawdowns.
So, without leverage, it is a way of controlling risk. It will NOT maximize your returns.
Just the start of an idea—probably with a more than a few flaws. Okay, I am sure it has some flaws. Comments appreciated.
edit One serious flaw, I think, is that none of this gives any basis whatsoever for picking TLT, SPY and GLD. And therefore the mean is arbitrary. To some extent you would select ETFs that gave you a favorable mean to build upon. Perhaps, you would want ETFs that had little correlation or were inversely correlated. The criteria for ETF selection would need to be expanded without a doubt.
But once you have decided on those ETFs, for whatever reason, I think it might be pretty solid method for balancing the returns and the drawdown risk. Again, it does not guarantee the maximal return (without leverage): that would probably come from putting everything into SPY long-term.
Jim