Fully automated futures trading

AvantGarde said:
Hi GAT,
I too would like to use forwards, but more so that I can run my account in USD and hedge back to base, which outperforms changing my system to run positions in base currency.
I am curious, if you were to take your FX trades using forwards how would you deal with the constant mismatches in settlement dates? E.g., if you open an exposure today using a 1 month forward in e.g., EURUSD, and if vol and spot moved tomorrow, you would have to adjust your position. In this case, would you open a new 1 month forward the following day or adjust the forward you opened yesterday (and do this everyday until settlement, at which point you'd open a new 1 month forward)? From my experience forwards also have wider spreads as you have counterparty risk due to this being an OTC exposure. What are your thoughts here and how do CTAs deal with this generally?
More...

Ideally you'd trade forwards which always start on quarterly IMM dates. This is what they do in CTA land. Then you have something that looks almost exactly like a future (assuming you also cash settle the forwards on expiry).

Last time I checked some spreadbetting firms offer something similar to this, although as usual the spreads are too wide for my taste.

GAT
 
isotope1 said:
A quick question about correlation. I have rebuilt the system from scratch using your suggested instruments. It all works well; just looking at the correlation between instrument returns (weekly resampling, 1992 onwards):

View attachment 166285

My question is related to correlation/causation. According to your book/blog posts, my understanding is that I should now add instrument weights to offset any correlation. Does adding in those weights make much difference for you, versus an equally weighted portfolio?

(Current Sharpe for this as an equally weighted portfolio 1992:today is ~0.45)
More...

Non equal instrument weights don't "offset" correlations exactly but I think I understand what you mean.

On a portfolio like this where most of the correlations are pretty similar it wouldn't make much difference.

GAT
 
globalarbtrader said:
Ideally you'd trade forwards which always start on quarterly IMM dates. This is what they do in CTA land. Then you have something that looks almost exactly like a future (assuming you also cash settle the forwards on expiry).

Last time I checked some spreadbetting firms offer something similar to this, although as usual the spreads are too wide for my taste.

GAT
More...
I see, thanks. What is the benefit to trading an OTC forward vs a listed future?
 
I see, this makes sense. You could still control the frequency you're trading at to keep costs in check, but obtain greater diversification.

I was wondering, where do you source your Korean 3yr, 10yr and Kospi200 futures data from? I see none of this is available on Quandl...
 
AvantGarde said:
I see, this makes sense. You could still control the frequency you're trading at to keep costs in check, but obtain greater diversification.

I was wondering, where do you source your Korean 3yr, 10yr and Kospi200 futures data from? I see none of this is available on Quandl...
More...

Interactive brokers

GAT
 
GAT, how do you calculate your individual instrument volatility forecasts (for risk targeting)? I've tried various timeframes for rolling stdev of daily returns, and I find that the results from my backtest are sensitive to how I calculate this volatility (sharpe could vary from 0.3-0.8 depending on how I measure the vola).
 
tradrjoe said:
GAT, how do you calculate your individual instrument volatility forecasts (for risk targeting)? I've tried various timeframes for rolling stdev of daily returns, and I find that the results from my backtest are sensitive to how I calculate this volatility (sharpe could vary from 0.3-0.8 depending on how I measure the vola).
More...

Exponential weighting 25 day halflife
 
tradrjoe said:
How did you pick this value? Do you know why the results would be so sensitive to how long the lookback is?
More...

Same as riskmetrics (to be precise I use a=.054 where v_t=a*x_t + (1-a)*v_t-1 and x_t = r_t ^ 2)

I'm very surprised the results are so sensitive. For lookbacks from 1 week up to around 20 weeks I found almost no difference in pre cost SR. Of course the faster lookbacks cost more to trade. For cheap futures using the default is fine. For expensive ETF's using a slower lookback makes sense, up to a point.

If you're interested there is much much much more on this in chapter 12 of my book.

GAT
 
You must log in or register to reply here.
Back
Top