Fully automated futures trading

[Kernfusion]

Following a comment on my blog I spent quite a bit of time this morning trying to think of a good way of doing things along these lines of a more heuristic optimisation, but really struggled to come up with something that wasn't really complicated (I had something along the lines of allocating risk budget to asset classes depending on their forecasts, and then within those taking positions based on the highest forecasts until the risk budget was filled... really messy).

GAT

I imagine defining asset-class groups might be a problem too, even if we do it statically based on our intuition, is it going to be Equity\bond\Commodity\FX or do we go deeper into commodities and split it into grains, metals, energy.. Doing grouping dynamically based on the current correlations would probably be even messier..
Maybe some static grouping and simply say that none of the asset-classes can ever take more than 50% of the risk and then it's first come, first served (I think stock people mostly do their allocations to the first X while there's capital, maybe with randomization in the backtest..).
Also, with prioritization by forecast strength, if we're using capped forecasts, there's still a possibility of a situation when we have more +20 instruments than available capital and we're still forced to choose arbitrary from that group. Prioritization by the asset class first might help solving that, but there's still a possibility that there might me more equally-worthy candidates than available capital. Or maybe I'm focusing on the edge-case too much..
Would still be cool to have something more logical than just first come first served..

 

[HobbyTrading]

I imagine defining asset-class groups might be a problem too, even if we do it statically based on our intuition, is it going to be Equity\bond\Commodity\FX or do we go deeper into commodities and split it into grains, metals, energy.. Doing grouping dynamically based on the current correlations would probably be even messier..

I decided to forego all the statistics and simply use a (1/n) split of the available risk capital over 6 asset classes: bonds, equity, agriculture, forex, metals and energy. Within some of the classes am I using subclasses: agriculture is split between grains and meat. Similarly have I split bonds in short and long term bonds. Short term bonds tend to have much lower volatility (because of smaller contract size) and would therefore prevent any long term bonds to have an open position. Within each (sub)class have I then defined how many open positions I allow. This is based on account size and the volatility of the contracts in this (sub)class.
All in all a handicraft approach, without any mathematical or statistical foundation.

 

[djames]

I decided to forego all the statistics and simply use a (1/n) split of the available risk capital over 6 asset classes: bonds, equity, agriculture, forex, metals and energy. Within some of the classes am I using subclasses: agriculture is split between grains and meat. Similarly have I split bonds in short and long term bonds. Short term bonds tend to have much lower volatility (because of smaller contract size) and would therefore prevent any long term bonds to have an open position. Within each (sub)class have I then defined how many open positions I allow. This is based on account size and the volatility of the contracts in this (sub)class.
All in all a handicraft approach, without any mathematical or statistical foundation.

My kids would call this the OG approach

 

[HobbyTrading]

My kids would call this the OG approach

I had to look up that abbreviation. But I think I can live with it

OG is a slang term for someone who's incredibly exceptional, authentic, or "old-school."

 

[globalarbtrader]

I've posted some more on this small account problem, this time on using a more systematic method to pick a static subset of markets.

https://qoppac.blogspot.com/2021/06/static-optimisation-of-best-set-of.html

I am going to have a stab at the more heuristic method of picking up positions until you are 'full' and see how that goes. Then I'm probably done on this subject.

GAT

 

[tradrjoe]

Rob, I would like your opinion on the value of adding stocks such as VIRT (Virtu Financial) and FLOW.AS (Flow Traders) as a portfolio hedge. These stocks represent HFT market making firms and have strongly negative betas compared to stock indices. Surely, by MPT, they should have a hefty weight in a balanced diversified portfolio?

Any comment on this?

 

[globalarbtrader]

tradrjoe said:
Rob, I would like your opinion on the value of adding stocks such as VIRT (Virtu Financial) and FLOW.AS (Flow Traders) as a portfolio hedge. These stocks represent HFT market making firms and have strongly negative betas compared to stock indices. Surely, by MPT, they should have a hefty weight in a balanced diversified portfolio?

Any comment on this?

Clearly, no...

I haven't done the analysis, and I don't really use MPT anyway, or trust Beta. But yes in theory, if you did this optimisation and found you should allocate more to these kinds of stocks then go for it.

GAT

 

[wopr]

Backtesting: I don't think so
I could very easily backtest the above code: reselecting a group of instruments every single year. However I don't see the point. I'm not expecting it to add performance value compared to a benchmark of just using my current set of instruments, or a randomly chosen set of instruments - performance per se isn't one of the things I'm considering.

I gotta say I admire this. I'm 100% certain I could not resist running a backtest. I wouldn't change or tweak anything depending on the result, but I would run one.

You did mention in a post in this thread a few days ago:

Trying to solve the problem that:

- diversification across instruments is the only free lunch with huge benefits to expected SR
- it consumes capital, thus making it hard if you only have a small account

Why not run a backtest and see if the backtested SR is better?
And second, I didn't understand why would you not expect this to add value over your current set of instruments? Is it because you were picking your current instruments with an eye towards diversification and removing 10 markets and adding 2 news ones isn't that big of a change?

 

[globalarbtrader]

I gotta say I admire this. I'm 100% certain I could not resist running a backtest. I wouldn't change or tweak anything depending on the result, but I would run one.

You did mention in a post in this thread a few days ago:


Why not run a backtest and see if the backtested SR is better?
And second, I didn't understand why would you not expect this to add value over your current set of instruments? Is it because you were picking your current instruments with an eye towards diversification and removing 10 markets and adding 2 news ones isn't that big of a change?

Yes it's basically a systematic method for what I was doing manually.


GAT

 

[Elder]

i

I've posted some more on this small account problem, this time on using a more systematic method to pick a static subset of markets.

https://qoppac.blogspot.com/2021/06/static-optimisation-of-best-set-of.html

I am going to have a stab at the more heuristic method of picking up positions until you are 'full' and see how that goes. Then I'm probably done on this subject.

GAT

Loved the post and the latest TTU episode. Very impressed with the incredible output - and all whilst the Euro Cup is on!