I see. But the Kelly criterion accounts for this very case. ie. Betting on two (or any number of) assets.
Let's say you believe you have information (in the information theory sense of the word) on two assets. I'll use your two strategies A and B. If the kelly fraction for A is 30, and for B is 60, then in order to determine how to size your bet you would need to know how correlated the returns are between A and B are during your backtests. If A and B are perfectly correlated (corr = 1), then the math comes out where you'd reduce each bet for A and B by half, since the correlation between the strategies implies A and B are the same bet. If A and B are perfectly anti-correlated, then the math comes out that you'd bet *more* than 30x and 60x since the correlation between strategies implies a reduction of variance due to the fact the two returns of the two strategies are anti-correlated. If A and B are perfectly non-correlated (corr = 0), then the math comes out that you'd in fact bet 30x and 60x for A and B respectively. In all cases, you need to know a) the correlation between strategies of A and B, and b) the size of your bankroll in order to determine the size of your trades in order to maximize growth given the information you believe you have for A and B. In the continuous case for a single asset, the kelly fraction is = mean(returns)/variance(returns) In the continuous case for multiple assets, the kelly fractions are = covarianceMatrix^-1 * meanReturnsVector, where ^-1 means matrix inverse, and the returns are the returns generated in your backtests. mean(returns) would be the average single-bet return gained using some strategy, and variance(returns) is the variance of those single-bet returns. So if you're using 1min ticks, then the return over 1min would be a single bet. If you're using 1hr, then that 1hr would be considered a single bet, and so on.. In either case, calculating the kelly fraction, and *not* using it to determine bet/position size is a pointless calculation. So I guess there are no plans to create something like, AccountManager to execute queries to return liquidation value of the account's portfolio? On Sep 25, 6:18 pm, nonlinear <[email protected]> wrote: > You are right in that the Kelly Criterion is typically used to calculate > the optimum bet site (or in the case of trading, the position size). > However, in JBT it's used for a different purpose, which is ranking the > strategy with respect to a given set of parameters. Think of it this way: > suppose you have two strategies, A and B. Let's say they make the same > profit, and have the same max drawdown. What's different is that the Kelly > for strategy A is 30, and for strategy B it is 60. Which one would you > trade? The answer is obvious: you want to trade strategy B, because the > higher Kelly indicates that the performance is more stable and predictable. > In JBT, the optimizer can be configured to use Kelly as a way to rank the > optimization parameters. So, Kelly in JBT is used as a strategy performance > metric (along with other metrics, such as Net Profit, Profit Factor) during > the strategy optimization. > > > > > > > > On Tuesday, September 25, 2012 10:58:43 AM UTC-4, B wrote: > > > Hi I noticed that the kelly fraction is calculated. > > > But, I can't seem to find the code in JBooktrader that returns any > > account information in order to use the kelly fraction to size my > > positions. > > > Does that code exist? > > And if not, what's the point then of calculating the kelly fraction? -- You received this message because you are subscribed to the Google Groups "JBookTrader" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/jbooktrader?hl=en.
