That is an interesting approach, although I'd be surprised if 'average profit per unit of margin' was statistically significant. I guess it comes down to the fact I'm not really margin constrained, since I'm running my account at a risk level below where that would happen. The margin constraint is artificial and comes from the fact I'm choosing not to have the entire capital at risk sitting in cash, but instead to have a little bit of extra leverage on my entire portfolio. GAT
My approach was not intended to be "statistically significant". I was trying to find a way to identify which instruments use more margin than others, in a relative approach. That is why I suggested average profit margin, in order to make it relevant. Maybe another parameter would make better sense. You can call the margin constraint "artificial", but the fact that IB rejects your orders and your positions don't achieve your desired position size is far from artificial. Although I'm running my system at only 25% risk am I confronted every now and then with these pesky initial margin issues and have to tweak my system a bit to get it to accommodate.
I guess something like the ratio of margin to volatility per contract would make sense; but if the margin system is efficient then all instruments should have roughly the same ratio (with the exception of negative skew stuff for which standard deviation is a lousy measure of risk; eg VIX, V2X). By artificial I mean that if I fully funded my account with cash I wouldn't have any margin issues running at 25% vol; I'm currently using less than half of my notional capital for margin. GAT
"should have roughly the same ratio" I guess the key word here is "should". I am currently facing issues with margin: using more than my account value and getting warning messages from IB. So I have to do some temporary tweaking to stay below the limit and avoid that IB makes decisions by themselves and closes/reduces position sizes. If they start to interfere then my system would not understand that and try to correct the mistake again. I do some temporary tweaking in the instrument weightings until things are better in balance and then I move the weights back to where I would like them to be. It is for me not the first time that this happens. I consider it a luxury problem: too many instruments are having strong signals at the same time, resulting in large(r) positions, thus reaching the limit of available margin.
Of course in theory it would be possible to have some kind of optimisation that worried about margin, but that would cause more hassle than it was worth IMHO. By the way I have a special override parameter which multiples positions to avoid having to change instrument weights; might achieve the same effect but it's cleaner. Glad to know it's not just me that is having margin issues, although I am curious as to what your funding ratio is given you are the same vol target as me? GAT
It would make the system more complex and path dependent. An equivalent 'nice' solution would be to: Lower the vol target eg from 25% to 12.5% Because the system now has effectively less risk capital, increase the threshold required before a position is taken (in my non linear position mapping function) to avoid small positions. Of course an alternative to (2) would be to reduce the number of instruments, weeding out those that are 'margin unfriendly'. GAT
I recall that you had also included a parameter for each instrument where you could set the maximum position size. This can help as well to avoid margin issues. I was too lazy to implement that. I decided that, if need be, I would temporarily fiddle with the instrument weight. Less clean, but with the same outcome. I fiddle with the weight until I get a desired position size. We may have identical vol target (25%), but there is where all commonality stops. I believe that your account is more than 10 times larger than mine, and that you cover many more instruments than I do. I do use a non-linear position function, as you recommend in your blog about small portfolios, but have implemented it differently. This makes that the "magnification" for strong forecast values is different then your thresholding approach. How do you define "funding ratio"? I'm not familiar with that parameter.