Critique this position sizing method

[Martinghoul]

There is a LOT of variants discussed out there on them internets... For example, this:
http://www.distressed-debt-investing.com/2010/04/kelly-formula-and-event-driven.html

 

[Daal]

There is a LOT of variants discussed out there on them internets... For example, this:
http://www.distressed-debt-investing.com/2010/04/kelly-formula-and-event-driven.html

What are your thoghts on the Secure F?

 

[tdazio]

Daal,
kelly f is: (r-Rf)/v, were r = expected return, Rf= free-risk ret, v= expected variance of ret
Tipically you use history to forecast r and v. If r and v come from past real trading 1/2 kelly is ok, but if they come from backtesting I suggest 1/4 kelly or less.
Better to have not max log-grow of money, than broke

 

[Daal]

Daal,
kelly f is: (r-Rf)/v, were r = expected return, Rf= free-risk ret, v= expected variance of ret
Tipically you use history to forecast r and v. If r and v come from past real trading 1/2 kelly is ok, but if they come from backtesting I suggest 1/4 kelly or less.
Better to have not max log-grow of money, than broke

That's the idea of sizing to the max drawdown thing. If your assumptions are off, depending how frequent the system trades, it still might take a long time to produce the 99% drawdown that hits the max drawdown level. So from that perspective, its conservative. Not only you have to be off by a significant amount, you also have to be unlucky enough that the streak of losses necessary to produce the max drawdown happens in your lifetime. But I'm just speculating and thinking out loud here. Was wondering if there is a flaw into this thinking

 

[Martinghoul]

What are your thoghts on the Secure F?

I am relatively familiar with F-Secure, but I have never heard of Secure F...

 

[Daal]

I am relatively familiar with F-Secure, but I have never heard of Secure F...

Its a variation of the Optimal F that takes into account max tolerable drawdown
http://signaltradinggroup.com/wp-content/DCSArticles/Securef.pdf

 

[Martinghoul]

Its a variation of the Optimal F that takes into account max tolerable drawdown
http://signaltradinggroup.com/wp-content/DCSArticles/Securef.pdf

All these are variations of the same idea, which can't address the fundamental uncertainty inherent in estimating the inputs. Sure, you can run any number of backtests or Monte Carlo simulations to get some flavor of F, which will fit in that particular instance. In other words, you can data-mine to your heart's content, with whatever personal constraints around the drawdown. I don't see anything too exciting about that.

FWIW, if you read the comments on Ernie Chan's blog post I mentioned earlier, you will find a similar discussion.

 

[Daal]

All these are variations of the same idea, which can't address the fundamental uncertainty inherent in estimating the inputs. Sure, you can run any number of backtests or Monte Carlo simulations to get some flavor of F, which will fit in this particular instance. In other words, it's just data-mining.

I'm not sure how off one's assumptions have to be for a 1/2 kelly/optimal F size to generate the drawdowns of the full kelly amount. I believe they have to be large. And if the person is making those large errors, isn't it likely he is a losing trader and hence no position sizing can save him?

By sizing to the max drawdown with some safety nets, the trader can extract the max profit per unit of tolerable drawdown

Now, I don't necessirely think this is a holy grail, your idea of using stress levels as one of the inputs also deserves consideration. Some 25% drawdowns might be tolerable in some instances but not at others due life circumtances, etc

 

[Martinghoul]

I'm not sure how off one's assumptions have to be for a 1/2 kelly/optimal F size to generate the drawdowns of the full kelly amount. I believe they have to be large. And if the person is making those large errors, isn't it likely he is a losing trader and hence no position sizing can save him?

By sizing to the max drawdown with some safety nets, the trader can extract the max profit per unit of tolerable drawdown

Now, I don't necessirely think this is a holy grail, your idea of using stress levels as one of the inputs also deserves consideration. Some 25% drawdowns might be tolerable in some instances but not at others due life circumtances, etc

I agree and, if you look at the link in my other post (the Tiburon one), I think those guys broadly agree with you as well.

 

[globalarbtrader]

Let's say you have $100,000 and your max tolerable drawdown is $25,000. You would then assume $25,000 is your total capital

Then you would calculate the Kelly/Optimal F for your trades. You can then use 1/2 Kelly/Optimal F but you apply the fraction to the $25,000 not the $100,000

Correct me if I'm wrong but in a lot of systems, the Kelly/Opf, will generate a 99% drawdown if traded to infinity (but depending of the number of trades per year, that might take decades/centurities). By using 1/2, not only you protect against the usual stuff (estimation errors, fat tails) but also you make sure it becomes unlikely you will breach your max drawdown. At the same time you will maximize your capital growth given your risk tolerance (which might not happen if you use a fixed percentage rule of thumb like 0.5% regardless of the trade)

Thoughts?

One downside is you will do a lot more position size related trading

So if you lose 12.5K you'd cut your position in half; then double your positions when you make it back.

IMHO it's better to trade an eighth (half times a quarter) optimal F on the full amount you're willing to lose. If your trading capital is 100K that ought to be what you're prepared to lose.

This gives you the same effective risk as using half optimal F on 25K, but if you lost 12.5K you'd only reduce your positions by 12.5%, so less trading.

GAT