Quote from peregrinecap:
I am looking for a logical/theoritical approach to position sizing. I have read about the kelly criterion in articles and books, but I find it hard to apply in practice. how do you guys know of any other mathematical/logical methods of position sizing other than kelly?
Thanks.
Even if I understand why they exist (easy to understand, apply and at least give some thinking about risk into the process) simple rules of thumb like ânever risk more than 2% in any positionâ, I think they often miss the point.
First, they only refer to the risk of one position whereas normally people normally wants to manage risk for their account. So this kind of rule only transforms the question into how many positions and which positions to take. And then the correlation/co-movement between instruments becomes a factor.
Second, if you take it literally it means that you can only invest 2% in a position, since there is always a (minute but still existing) risk that the position becomes worthless. If you want to use this kind of rule then I believe it should be restated to something like âno more than 95% risk of losing more than 2% on any dayâ, i.e. you need to introduce some kind of probability. Maybe this is what is actually meant with these rulesâ¦
One option is to implement Value-at-Risk (VaR) and manage this. A good description of this is given in âRed blooded riskâ by Aaron Brown, but there are probably many other places to find information about this. Depending on what instruments you trade this could be easy or very hard. This gives the risk on the account level but you still need some way to push this down to the individual positions. One way is to look at how much a new position contributes to the VaR.
Of what Iâve seen the best formal work that is usable for trading is the work of Ralph Vince. I often recommend to take a look at his work. I especially like the book âThe Leverage Space Trading Modelâ. One good thing with this approach is that it actually works with the risk people normally care about - account drawdown risk.
But âcaveat lectorâ - it is also often misunderstood. The reason behind this is probably that it is rather mathematical and sometimes not so easy to understand. So he has taken a lot of heat by people that claims that using his methods will blow up your account. In my opinion the main problem is not in the methods per se, but in the assumptions we make around the input data used for the calculations.
Using your backtest data to size real trading positions will very likely result in overbetting your system. Normally the returns distributions from your backtest are susceptible to overfitting and/or selection bias. In my opinon every system, also those systems that has not been created through automated optimization or learning suffers from this, simply because they have been tested against market data and found to be working. So always assume that the returns distribution of your walk-forward activities will be more adverse than those of the back-test.
So, I think more focus should be on the how representative the input data really is. In the end, if the returns distribution for real trading is so different from the ones used for positioning that you blow up, you cannot really blame the method, can you?
