Your question made me curious myself. So I went back to my data. It turned out that I only started to collect initial margin by June 2021. Before that was I only collecting maintenance margin data (started in April 2017). A chart of the relative margin requirements are (margin as percentage of NLV), since the start of collecting initial margin: The result surprises me a bit: this fluctuates much more than the impression I had of it.
Hi guys. I'm currently setting up my own system based on Rob's books / blog posts and tweak it for my own needs. I'm wondering about the following. According to IB there are quite some limitations regarding historical futures data: Expired futures data older than two years counting from the future's expiration date. How / from where did you guys collect data going further back? I've just started to collect data. If that was already answered somewhere in the 313 pages of this thread, mea culpa Still catching up on that one.... cheers
I use barchart.com (paid for subscription which I will keep whilst I'm doing my 'add loads of markets to my system' project), but in the past I've used quandl (when it was still free). Rob
Personally I wouldn't risk this since the cost of doing a few extra optimisations isn't very much, especially in production. Rob
1. No 2. I'm only running one strategy at a time (currently), but with multiple sources of forecasts. Adding another forecast doesn't chew up any capital. Adding another instrument would also chew up capital under the old system, but not with dynamic optimisation of course. GAT
I back-tested it and the number of trades and everything else is pretty-much the same., Basically what is happening is that most of the time when the system wants to change position in an instrument, the required adjustment with my small capital is only 1 contract, which results in no change if the adjustment factor is <0.5, and it's always below 0.5 when the error-difference is less than 2x buffer by construction, so why bother with the optimization if we already know we're not going to change the position anyway (except if the difference in some instrument is more than 1 contract, but somehow the risk-difference is less than 2x buffer which almost never happens). It's just that I'm doing it all in real-time in live-trading so there's a ton of unnecessary computations.. And the absolutely-worst that can happen is I'll effectively double my buffer..
I'm also thinking that maybe a good ongoing report for DO would be to first find all instruments which currently have high "before optimization\natural" position but "small" actual position (which is unusual in the first approximation), and then for each of them print 3 currently most highly-correlated instruments and their current positions., The expectation here will be that if everything is running normally, then for all such high-natural\low actual position instruments we should see high positions in the most-correlated with them instruments..