2 GB of memory is insufficient for me, I sometimes encounter some memory leak issue in IB gateway that eats up my memory: java.lang.OutOfMemoryError: Java heap space. Rather than figuring out the cause (doesnt affect pnl of my strategies anw), I just allocate more memory to IB gateway and reset it every weekend.
Fully automated futures trading
- Thread starter globalarbtrader
- Start date
2 GB of memory is insufficient for me, I sometimes encounter some memory leak issue in IB gateway that eats up my memory: java.lang.OutOfMemoryError: Java heap space. Rather than figuring out the cause (doesnt affect pnl of my strategies anw), I just allocate more memory to IB gateway and reset it every weekend.
Really? I have never seen that error message. I switch on Gateway on Sunday evening and let it run unattended until the following Saturday morning. Thus a bit more than 5 days continuously. On the other hand, my NUC computer has 8 GB memory, so maybe that's why it is not running into problems.
Co-incidentally I have backtested trading the second dec for the oil market mainly because I was trying to find a way to mitigate the effects of the vol curve in oil futures. Since WTI Dec 1 yr out typically has a lower vol than 1st or 2nd nearby, the volscalar goes up with time and the optimal position size declines. This to me seemed to be undesirable when trading non-institutional account sizes. My thought was that trading 24 mths out and rolling whilst it is still 12 mths out, you would suffer a much smaller increase in vol.
However, the tests showed that this came at a cost of decreased performance. I suspect this may be because very long dated commodity contracts do not trend as well. Other performance metrics such as Portfolio skew, return and worst drawdown all worsened when I switched to trading the 2nd Dec. So I found no real benefit to making the switch. Of course, anyone sufficiently curious would want to verify this themselves, and correct this impression in case I have erred somewhere.
I did have the passing thought that Dec-Dec spreads might do better as an alternative, but I never got around to testing this thought. With spreads there is of course the extra challenge that the returns are not lognormally distributed.
Yes, it's a strange feeling when I have a electricity blackout with no way to check on the VPS. Normally I at least follow the log in realtime.
It makes sense (Ernie is a smart guy who I have a lot of respect for) but it would be too complicated for me and little value outside of something like Crude and the other seasonal commodities. I prefer to keep things as simple as possible, and ideally use the same methodology across all markets. Having said that, if I tested it and found a huge benefit, of course I'd use it (but my intuition is that it won't give much advantage over using the nearest contract and smoothing).
GAT
That's a one year regression of prices to get a slope, right? Yeah I've heard of it, and also played around with something similar, but I don't formally use it as a benchmark. Andreas Clenow also has something similar in his second book, but he uses an exponential regression and if memory serves also takes into account the R squared. The advantage of the exponential regression is it's smoother (when a large data point drops out you don't suddenly change your fit - imagine what will happen to this indicator through March 2021!). The vol scaling in the AQR method is a bit strange. Finally I'm unhappy only using a single time period; I prefer several (of course this is easily fixed).
Ultimately, there are millions of ways to pick up trends, and they all achieve pretty similar results. I wouldn't say 'oh this method is definitely the best', since it's extremely likely that everything is pretty correlated and the difference isn't significant. Best of all is to trade several of them; but don't get carried away and add dozens of trend following models - at some point you are better of diversifying across other sources of return!
GAT
I've had this discussion many times and I have to face up to the fact that I just really like looking at physical hardware, and showing it to people; and no matter how cheap the cloud is, or how many times my internet goes on the blink, I'll probably always feel that way.
https://cpcrulez.fr/book_english-alan_sugar-the_amstrad_story.htm
"Later on, Sugar looked at the new electronics and hi-fi industry with the eye of a "working class punter". He didn't know everything about the innards of component systems and he didn't want to. He knew what people would buy. Putting all the different parts of a music system together, Sugar invented the hi-fi Tower. It was to revolutionise the industry, and it was to sell in its millions. Sugar's name for this appeal to the masses was "a mug's eyeful". He believed that most people wanted something that had lights, dials and features on it."
I guess that makes me a working class punter...
GAT
It's not even a regression haha. They only compare 2 price point (t and t-261) and rebalance at a monthly interval. For each instrument, they assume annualized vol of 40%. There's no vol control at portfolio level. And it ticks me that such a simple model with 30 lines of code is better than the model that I'm building currently. But of course, simplicity may triumph complexity in some cases.
I left out carry effect from the model that I'm building because of the nature of my dataset. Wondering how much impact would there be on sharpe ratio?
Yes, I'm definitely a fan of combining multiple signals and continuous forecasts. I used it heavily in other systematic models.
I'm getting a bit concerned with my US 2y bond (ZT) position.. The system is currently long 6 contracts, and it's mostly because volatility got extremely low. I'm using volatility flooring, but it's still very low.
The carry signals are close to zero, but trend and breakout permutations are somewhat positive, which makes the system go long. I have a graph for each contract which shows approximate position with the average forecast and current vol levels, and it's about 6 contracts, which is what's happening right now..
Don't know, it's a small contract, but still holding 6 of them on an average signal seems a bit too much., Also, what is my upside here? I mean the interest rates are almost at zero already, so I'm just not gonna make any decent profits on it even in the best case.. So Maybe it's a similar situation as with the German bond several years ago - not worth holding it until it's volatility increases ?..
Theoretical current position with average forecast:
current signal levels (red is the final combined):
Volatility (red is long-term floor, blue - actual vol):
my positions:
The carry signals are close to zero, but trend and breakout permutations are somewhat positive, which makes the system go long. I have a graph for each contract which shows approximate position with the average forecast and current vol levels, and it's about 6 contracts, which is what's happening right now..
Don't know, it's a small contract, but still holding 6 of them on an average signal seems a bit too much., Also, what is my upside here? I mean the interest rates are almost at zero already, so I'm just not gonna make any decent profits on it even in the best case.. So Maybe it's a similar situation as with the German bond several years ago - not worth holding it until it's volatility increases ?..
Theoretical current position with average forecast:
current signal levels (red is the final combined):
Volatility (red is long-term floor, blue - actual vol):
my positions:
I mean, here's the 2Y interest-rate plot from cme - the thing is practically at zero and almost a straight line, so what exactly am I(my system) trying to achieve here by holding a long position? Like am I betting on negative interest-rates?
https://www.cmegroup.com/trading/interest-rates/us-treasury/2-year-us-treasury-note.html
https://www.cmegroup.com/trading/interest-rates/us-treasury/2-year-us-treasury-note.html