Fully automated futures trading

Same Lazy Element said:
The latest episode popped up on my iTunes yesterday while I was walking the dog :)

Here are some comments on using signal strength for position sizing (that's what I do for my strategies). It has good and bad sides to it
- it is more suitable to reversion/normalization strategies since you increase your signal as your expected alpha becomes larger even if it has not been working
- pnl/tradevalue usually improves since you do not have fixed size round-trips as you do with fixed thresholds, it's also much easier to think about transaction cost control in this setting
- you have to have some sort of min and max signal parameter (min so you don't trade the noise and max so you don't go overboard), so it's yet another parameter to fit (and possibly curve fit)
More...

My next blog post will be on this subject, but very quickly:

- it also works for momentum, although perhaps the intuition makes more sense for mean reversion
- yes agreed
- disagree, you don't have to have a min signal parameter and the max signal parameter doesn't need fitting you just set it at some sensible value (in my case twice the average signal).

GAT
 
Same Lazy Element said:
- you have to have some sort of min and max signal parameter (min so you don't trade the noise and max so you don't go overboard), so it's yet another parameter to fit (and possibly curve fit)
More...
If your maximum position size is a limited quantity then you'll automatically have some sort of filter against noise from your signal. Example: if your position size at maximum signal level is 5 contracts, then 1 contract represents 20% of your maximum signal. All noise which is less than 10% (less than 0.5 contract) will be suppressed and will not result in additional trading. It's a form of quantization. If your maximum position size is very large then this approach could be insufficient.

Alternatively you could feed your signal through some sort of S filter: small signals will get dampened, for middle sized signals there is a linear relationship from input to output, and large signals get dampened.
 
HobbyTrading said:
If your maximum position size is a limited quantity then you'll automatically have some sort of filter against noise from your signal. Example: if your position size at maximum signal level is 5 contracts, then 1 contract represents 20% of your maximum signal. All noise which is less than 10% (less than 0.5 contract) will be suppressed and will not result in additional trading. It's a form of quantization. If your maximum position size is very large then this approach could be insufficient.

Alternatively you could feed your signal through some sort of S filter: small signals will get dampened, for middle sized signals there is a linear relationship from input to output, and large signals get dampened.
More...

... which is related to how I currently deal with limited capital right now (not an S shape)
https://qoppac.blogspot.com/2016/03/diversification-and-small-account-size.html

GAT
 
You do need a either some form of minimum signal threshold or a hysteresis of some sort. Most of the noise is random flickering near zero signal - if you going to trade an epsilon amount every time as your signal turns on and off, you gonna pay a lot of TCs.

In fact, you can think of lower signal threshold as part of your signal processing while your upper threshold as part of the risk management process.
 
Same Lazy Element said:
You do need a either some form of minimum signal threshold or a hysteresis of some sort. Most of the noise is random flickering near zero signal - if you going to trade an epsilon amount every time as your signal turns on and off, you gonna pay a lot of TCs.

In fact, you can think of lower signal threshold as part of your signal processing while your upper threshold as part of the risk management process.
More...

Perhaps I should have said, but I have position buffering in place (which is indeed a form of hysteresis). So I won't trade any small change in signal (wether it be from a signal of 0 to 0.01, or from 5 to 5.01) that is less than a certain threshold. Hence I don't need a minimum signal threshold to reduce trading costs.

Plus, to an extent, this is all academic as a retail futures trader since the forecast has to be pretty strong before you have enough risk for a single contract.

GAT
 
How about quantpedia as a source of trading ideas ? Not exactly a Slack channel, and there's still almost 500 filtered strategies to go through, lots of which could be not feasible for one reason or another, so you probably still need to have a sense what to look for in advance..?
(this thread with all of it's related material is of course the best by far :) )
 
Last edited:
globalarbtrader said:
Perhaps I should have said, but I have position buffering in place (which is indeed a form of hysteresis). So I won't trade any small change in signal (wether it be from a signal of 0 to 0.01, or from 5 to 5.01) that is less than a certain threshold. Hence I don't need a minimum signal threshold to reduce trading costs.
More...
Makes sense. I have hysteresis built directly into the portfolio formation for cross-sectional strategies, but for single asset strategies I do it directly at the signal level.

globalarbtrader said:
Plus, to an extent, this is all academic as a retail futures trader since the forecast has to be pretty strong before you have enough risk for a single contract.
More...
Hmm, I did not think about it but indeed - when I trade a single VIX contract it's epsilon for me while it could be a sizable chunk of the budget for a smaller trader.
 
globalarbtrader said:
In fact there is a general dearth of good quality discussion on the subject (dare I say this thread is one of the better places for talking about it?).
More...

As few others have mentioned, this thread *is* the best place for these discussion. Perhaps I was just looking for more distraction, as refactoring this PriceCalculations object isn't too interesting at the moment :)

Kernfusion said:
How about quantpedia as a source of trading ideas ? Not exactly a Slack channel, and there's still almost 500 filtered strategies to go through, lots of which could be not feasible for one reason or another, so you probably still need to have a sense what to look for in advance..?
(this thread with all of it's related material is of course the best by far :) )
More...
I probably didn't explain it well, but I wasn't thinking of discussing trading ideas or strategies (we don't even talk about those all that much here from what I can see), more like for example the discussion that Rob, Same Lazy Element and a few others just had few posts above, seems to me that's a lot easier to do somewhere where there's also proper threading/conversation support.
But I think we can also stick around here. :)
 
You must log in or register to reply here.
Back
Top