Fully automated futures trading

[HobbyTrading]

the values are the same for all instruments for the same permutation (e.g. EWMA_8_32 is 5.3 and EWMA_16_64=3.75 for all instruments )., Actually looks like I took these values from Table 49 of "Systematic trading"

Same here. This is what I implemented in my software.

 

[traider]

Thanks a lot, I think the scaling is made consistent by the denominator term when dividing by std deviation of price diff.

 

[wopr]

After trading paper money for more than a year while saving some real play money I finally turned on the live system on Friday
Base capital 100k USD, 14 instruments, most of them with thresholding enabled (selected based mostly by the smallest margin requirements but also tried to cover as many asset classes as I could), volume knob is at 25.

Good luck!
I'm in almost exactly the same situation. I went live about 3 weeks ago, same capital, but I only got 8 instruments, not sure how you managed to squeeze 14 and keep the capital allocations somewhat equal (or did you not try to do that?).

Also, bad luck (timing wise) with MXP , I'm in the same boat there too.

 

[globalarbtrader]

Let's assume that you are not facing any scalability issues (and thus not forced to change any parameters due to liquidity or risk constraints, which for AHL was not always true, I'd recon?). If your primary objective is to preserve the smoothness of your PnL profile, you can actually optimize a scaling schedule based on your Sharpe. I recall deriving something along these lines a few years ago, but my weed-infused brain is blanking on the exact formulation right now.

If your expected SR improvement>0 and costs=0 then surely the optimal schedule is to do it immediately? Though not sure what your definition of 'smoothness' is, some kind of tracking error between p&l before and after the change?

GAT

 

[globalarbtrader]

If anything, something being very prominent and liquid means that it's pretty efficient and not worth trading. IMHO, of course.

For the record, I disagree with this. I've never found any statistically significant pattern between SR (pre or post cost) of a trading system and eg volume of an instrument.

GAT

 

[globalarbtrader]

Does anyone remember if we estimate forecast scalar for a ema rule for all instruments or for each individual instrument? I was thinking if there was a way to make the ema rules more consistent across multiple instrument such as using percentage returns.

Dividing by vol does indeed make the rule consistent across instruments (and across time). I've experimented with other rules that take vol scaled or % returns cumulated into a price series as their input, and they provide a little diversification, but they aren't neccessary to achieve consistency across instruments.

Others have replied already, but basically it's better to use a scalar that is estimated across many instruments to avoid overfiting and there is no good reason why the scalars should be different across instruments.

You can even estimate the scalar using a random walk price series, you get pretty much the same answer.


GAT

 

[Same Lazy Element]

If your expected SR improvement>0 and costs=0 then surely the optimal schedule is to do it immediately?

If you assuming geometric SR, no serial correlation of your returns, no alpha decay etc, probably so. Once any of these assumptions is broken, I am not so sure. For example, if you are beholden to dollar Sharpe (many prop shops do that), it becomes a very different decision.

Though not sure what your definition of 'smoothness' is, some kind of tracking error between p&l before and after the change?

Sorry, "smoothness" was a wrong word here, rather a "desired risk metric". In my specific case (a standalone book at a fund) it is avoiding a draw-down of a given dollar magnitude. More specifically, it would be a scale-up rate that would not put me below zero on the year given PnL YTD.

 

[Same Lazy Element]

For the record, I disagree with this. I've never found any statistically significant pattern between SR (pre or post cost) of a trading system and eg volume of an instrument.

Well, I'd imagine that your strategies are exploiting a very specific feature, so it's hard for me to comment. In general, however, liquidity and efficiency go hand in hand. Something like the spooz are very liquid and extremely efficient, while something like SET50 futures have all sorts of quirks you can exploit.

PS. Actually, this make me wonder if that's a sign that your primary source of alpha is cross-sectional risk premia rather than market inefficiencies.

 

[globalarbtrader]

Well, I'd imagine that your strategies are exploiting a very specific feature, so it's hard for me to comment. In general, however, liquidity and efficiency go hand in hand. Something like the spooz are very liquid and extremely efficient, while something like SET50 futures have all sorts of quirks you can exploit.

PS. Actually, this make me wonder if that's a sign that your primary source of alpha is cross-sectional risk premia rather than market inefficiencies.

My source of 'alpha' is definitely risk premia.

GAT

 

[Same Lazy Element]

My source of 'alpha' is definitely risk premia.

It would make sense and it's mostly a good thing. Biases tend to be more persistent and are easier to capture with limited infrastructure. This said, it's hard to attribute these things well.