Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,386
Thanks Received: 10,206
General iSystems Update
Put in new all time highs early last week but had bad days Thursday and Friday which pulled me down from all time highs to negative !!! WTF !!! Was looking at my systems this weekend and thought, "Well if all but one of them doesn't do better in the rest of the month, I may be switching them off!". Then of course Monday (-5.5ish%) happened and the one system that has been money got smoked. So in about 6 trading days I've gone from account highs to new account lows, a swing of over $20k! WTF Part 2!
Machine Learning / Random Forest Update
Last time I made an update on the progress of this, it may have been apparent that I was concerned that my features and target variables were not aligned properly. Expanding on the conversation below with @dadaraa, things like gross profit aren't very useful in predicting system performance, while things like PF and Sharpe should be better. I did already know this, and wasn't using gross profit, and was using both PF and Sharpe, as variables. The problem is I was using some variables that were a little ambiguous. In the last several weeks I have completely reviewed my original data set, removed certain variables and added significantly more that are all scale-able. This took time because it involved rewriting the analysis functions that I pass the PnL tables to, which was a bigger job than I planned. I had hoped to use these new variables with the RFs this weekend but as it happened I just didn't have time. Maybe next weekend - if I'm not looking for a part time job to make up for all my trading losses In reply to dadaraa
Passion yes, but my PnL implies my knowledge still isn't good enough.
Not necessarily. As I've mentioned before a lot of my thoughts on analyzing automated trading systems come from what I've learnt in particular from @kevinkdog. He teaches WFO analysis which I strongly believe generates much more robust systems. By definition, none of the iSystems systems are WFO and as such are based on single point in time backtests. While I believe there are WFO systems that behave nearly as well outside of backtest, I agree there are very few single point in time backtests that do as well.
There's several essay long answer's to the questions in that paragraph. I would remind you, that everything needs to be considered in relation to risk. Would you rather trade a system that has monthly returns +10 -1 +10 or a system that is +1, +.5, +1? The first because it makes 19 versus the second that only makes 2.5? Well assuming those returns were commission and slippage adjusted (and hence scalable) would you rather be trading +10 -1 +10 or +10 +5 +10? Obviously the latter, but that is just the original second system scaled by a factor of 10. Hence absolute numbers are not always relevant. Things ilke PF and Sharpe are scaled metrics, while gross PnL and drawdown aren't (more to come on this when I get around to describing my most recent Random Forest iSystem Models). Kevin runs a challenge for his students every month, in order to even enter the challenge the system needs to have a Return to Drawdown ration of 2.0. Trust me, that is not easy to without cheating.
Without knowing a lot more about your system it's difficult to give you advice. There are so many things and ways you could be doing that could be distorting things. Even then I'm not sure I'm the best person to ask compared with somebody like Kevin*.
My hypothesis was exactly that. Systems that have a good backtest, and outside of backtest perform similarily, should be robust systems. I think my hypopthesis, at least in relation to isystems, was proven wrong.
I can generate backtested systems that look great easily. Inevitably as soon as I run them on out of sample data they fail. I really struggle (annoyingly so) to generate systems that perform well on a WFO basis. The few I do, tend to perform well out of sample. I would have said I'm still an intermediate system trader, but after the last few days I'm probably back to being a newbee. So again, I'm probably not the best person to ask about this. Now if you want to talk about trading crude oil butterflies, maybe we can talk.
* Full disclosure, I'm a Kevin fan, have his book, attended his course, and two of his seminar's, but am not affiliated in anyway.
Maybe I am an Idiot but i think doing this trading third parties strategies should be done using CDF´s and or Forex. It works the same it is just less risky but real money which is a game changer and You can easily scale up. I am still trying to figure out which strategies to trade and when even after having traded them live on mini accounts after 1,5 years. Portfolios can have 3 months draw down and then take everything back in 1 week.
Hi guys, any more update on iSystems live trading results and experiences? Some of the ML and AI based strategies looks interesting and does that mean with AI involved in constant adaptation, there is no need to re optimize the strategy parameters?
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,386
Thanks Received: 10,206
CFDs aren't legal in the US so this is a moot point for me - but maybe not others.
I'm a full time trader, so this is a part time, after hours hobby for me - that doubles as an R learning exercise - and triples as fulfilling my machine learning interest. I also try and enter Keven Davey's Strategy Factory Club every month as well. The SF club entry has to be in by the 25th. So normally from the 25th to the 15th I'm working on my iSystems project and then from the 15th to 25th my Strategy Factory entry. Starting this month though I'm putting more effort into my Strategy Factory entry, so I'll only be working on iSystems once I get a successful entry for the club done. Saying that...
General iSystems Update
Last update account had just swung to new all time lows in the Feb Equity Melt Down. Since then it's been one step forward one step back. One of the systems continues to perform well, while the others do not. I was looking at the performance charts last night and it looks like all of the poor ones will exceed previous max drawdown this month so I think I'll be switching them off. Question is do I leave the one good one running. For example (and this is NOT an equity index system) ...
(excuse the poor quality graphic but it's just a screenshot of a report I ran last night)
Machine Learning / Random Forest Update
I have my new data set and I quite like it. I've run some real simple initial random forests and the results look interesting. Unfortunately when I ran the simulation of the best and worst system each month the results were almost identical to the old data set. I had hoped to have done enough work on this to get the new systems rolling in March but I don't think that will happen now.
Below is a tutorial of the copier i am using to trade my own portfolios. It is commercial available copier but by me and a friend heavely modified . What i want to show with this is how I control when to trade a portfolio and what tools i use to control risk.
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,386
Thanks Received: 10,206
iSystems Update
The systems I am trading continue to do poorly putting in new account lows. I'll post the graphs latter, but for now I have deactivated everything prior to month end.
Hi SMCJB, sorry to hear about the unsatisfactory results you experienced, if even these systems with nice equity curves cannot work, what chance are there left for systematic traders who are starting from scratch? I wonder if it will help if you zero in on a pool of systems with consistent performance and activate them only after previous month is losing month, for a 'x' period of months and switch them off regardless after that.
Trading: Primarily Energy but also a little Equities, Fixed Income, Metals and Crypto.
Frequency: Many times daily
Duration: Never
Posts: 5,049 since Dec 2013
Thanks Given: 4,386
Thanks Received: 10,206
If you have a strong robust process, I think there's a lot of possibility. Key to that is Walk Forward Optimization rather than single point in time optimization. Unfortunately the way iSystems works, with code being locked upon submittal the opportunity for people to use WFO is limited.
That's what I'm hoping my machine learning venture will discover. If there are patterns like that the algorithms should find them.
Ive been wondering about using the isystems. Does it actually work and can you make any money off off it. I will only do minis and i have researched the sytem for profitable developers. I was going to go with Gain Direct.
This isystems looks like Fores robots that always lost money. The broker makes all the money. I am ging back to my ETF which makes 20& a year Its safe.
