I had a very bad experience with multi period optimisation.... so you can count me out of that discussion thanks very much.
GAT
Fully automated futures trading
- Thread starter globalarbtrader
- Start date
I had a very bad experience with multi period optimisation.... so you can count me out of that discussion thanks very much.
GAT
Dynamic optimisation is now live
I've also set up a system for publishing my reports: https://github.com/robcarver17/reports
Suggestions for stuff that go in there are welcome.
GAT
I've also set up a system for publishing my reports: https://github.com/robcarver17/reports
Suggestions for stuff that go in there are welcome.
GAT
Hi all,
I have created a github repo with a quick explanation of the idea and python and C# implementations:
https://github.com/golforado/tracking-portfolio
After reading Rob's post I thought I probably should try to pose the problem in a different way, but I think is similar enough to understand the idea.
I used cvxpy for python, it seems really good. As you can see, I'm using CPLEX in both python and C#. Changing to other solver in python is just changing the line prob.solve(solver='CPLEX'), but I don't think cvxpy default solvers are very good for this problem. I haven't tried all of them though.
@Kernfusion Yes, you would need to install CPLEX IDE afaik. There's no Nuget package, but you just just need to add a reference to the .net libraries in the cplex/bin installed files (ILOG.CPLEX.dll and ILOG.Concert.dll) and make sure that path is accesible at runtime.
Thanks,
Luis
I have created a github repo with a quick explanation of the idea and python and C# implementations:
https://github.com/golforado/tracking-portfolio
After reading Rob's post I thought I probably should try to pose the problem in a different way, but I think is similar enough to understand the idea.
I used cvxpy for python, it seems really good. As you can see, I'm using CPLEX in both python and C#. Changing to other solver in python is just changing the line prob.solve(solver='CPLEX'), but I don't think cvxpy default solvers are very good for this problem. I haven't tried all of them though.
@Kernfusion Yes, you would need to install CPLEX IDE afaik. There's no Nuget package, but you just just need to add a reference to the .net libraries in the cplex/bin installed files (ILOG.CPLEX.dll and ILOG.Concert.dll) and make sure that path is accesible at runtime.
Thanks,
Luis
Hello Rob, I have three questions:
You said in one of the interviews that you usually hold around 15 different futures positions at the same time, does this mean that your futures notional value (at 25% risk target) is multiple times of your capital in trading account?
What is the point of bond futures? I backtested some moving average pairs on daily data (ZB contract) and results are disappointing. Do you include them as portfolio diversifier in a more passive buy&hold strategies?
Do futures contracts specifications ever change with time (new rollover dates, new multiplier, new trading code...)? If so, how do you track these changes since you follow so many different contracts in multiple asset classes on multiple exchanges?
You said in one of the interviews that you usually hold around 15 different futures positions at the same time, does this mean that your futures notional value (at 25% risk target) is multiple times of your capital in trading account?
What is the point of bond futures? I backtested some moving average pairs on daily data (ZB contract) and results are disappointing. Do you include them as portfolio diversifier in a more passive buy&hold strategies?
Do futures contracts specifications ever change with time (new rollover dates, new multiplier, new trading code...)? If so, how do you track these changes since you follow so many different contracts in multiple asset classes on multiple exchanges?
Yes. In fact right now it's 500% of my capital.
I'd make three points:
Firstly, it is very presumptuous to test one instrument (ZB which is the 30 year US bond) and write off an entire asset class.
Secondly I'm extremely surprised by your results. Although the results from trend following US30 aren't as good for shorter maturities, they're still above average. You may want to double check something.
Finally, even if your findings were correct it's extremely unlikely that we'd find that an instrument, or asset class, has results which are so bad that their statistical significance implies we should not trade them at all.
Expiry dates change with every contract, I get these automatically from the IB API. The rollover pattern hardly ever changes, the only recent example I can think of was when VIX introduced weekly expiries. My code broke and I had to introduce a (hacky!) fix to deal with this.
Trading codes and multipliers also change very occasionally. If this happens my code would break ('identifer not found' type messages) and I'd have to manually tweak my configuration files. Most recent example would be BITCOIN mini futures (as in this thread).
Things like contract specifications changing (eg bond futures referencing different coupons) also happen, but this doesn't affect me as I use backadjusted prices.
Basically the lesson is to code taking nothing for granted, better that your stuff breaks rather than proceeding with an incorrect assumption.
GAT
Now I know why I've always had a very low margin percentage, currently about 8% of account balance. Gross futures value is about 150% of account.
Admittedly, the system is currently neutral all equities, Canadian dollar, corn and soybeans, and there is nothing (for me) to replace them with, but I can't remember the margin ever being above 12%.
Its the conservative position
KH
Now I know why I've always had a very low margin percentage, currently about 8% of account balance. Gross futures value is about 150% of account.
Admittedly, the system is currently neutral all equities, Canadian dollar, corn and soybeans, and there is nothing (for me) to replace them with, but I can't remember the margin ever being above 12%.
Its the conservative position
KH
There's a daily updated report here. Current value is actually 629%.
A lot of that comes from Eurodollar (about a third). If I took away short term bonds and rates I'd be a lot less.
What's your annualised risk target?
GAT
Annualised risk target: 25%, but IDM of 1 until last weekend, now 1.1. I'm only trading the micros, plus some minis. 12 in total.
The system ran via spreadsheets until I decided that a good night's sleep was more important than putting on positions. Fully automated since mid July.
I'm not too worried about the conservative position, just letting the system work itself in until I'm 100% happy that it is working properly. I'll make a decision in the new year about allocating more capital (edit: and more risk) to the system.
KH
Sorry for the second post, I should have added that I saw your Risk Report over the weekend, and I've already set about doing something similar. It will help me see how the system is performing.
I did notice that many of your positions are single positions, whereas I am still targeting a minimum 4 positions at Forecast 20. This rules me out of virtually all full contracts, and some minis.
KH
I did notice that many of your positions are single positions, whereas I am still targeting a minimum 4 positions at Forecast 20. This rules me out of virtually all full contracts, and some minis.
KH
Conservative is good! To be clear then, each instrument is targeting 25% *1.1 vol = 27.5%, but because of correlation you will undershoot that.
Yes, the many single positions are because I'm using the new dynamic optimisation method with over 100 instruments; of which I've actually got positions in 26 right now.
GAT