Neural Network system in action - Live

[Intel_Bvc]

So finally we're seeing a sustained pull-back. Currently account equity is $132k up $12k (10%).

Interestingly enough one of the big losers (long EUR/AUD) had spent the previous 7 days being profitable. However the optimum hold time is exactly that... the optimum. Not necessarily the best hold time every time. There are times when the trades should be taken off early and times when they should stay on for longer. However the optimum means that over the course of a year if you stick to it, that hold time will make you the most money.

I could also actually see from the predictions that there was a dip on the horizon for GBP/USD, but I resisted the urge to tinker.

For the first time we're seeing a negative floating P/L although the closed balance is still up $15k .

Another interesting thing for me is that I've noticed a difference in the way I've been calculating drawdown. Up until now its been calculated in the excel spreadsheet of all my trades for the year meaning its based on the closed balance, however really the way I should be measuring it is based on the hourly account equity that truly reflects the peak to trough move occurring on an account. In that respect so far max drawdown is looking like approx $18k or 15%.

Please find aggregated positions attached. Didn't get the chance to pick up the equity chart before I left home this morning. Will add it later.

 

[c6ackp]

Hi Intel,

Where did you get your back-test data. And your real-time market data?

Also, which API do you use to execute your orders?

Thanks,
c6

 

[Intel_Bvc]

The system is so computationally intensive that I cannot go back and retrospectively generate data. It would take me 1 day to generate 1 day of data/predictions etc. So I only have back testing data for as long as I've been running the system.

The system stores every single prediction that it makes. If you consider that there's 9 currency pairs that I trade and that each generates one 5-day ahead prediction, every hour, that's 1080 predictions a week.

All this info is stored in a DB and I can go back at any time and devise a strategy based on the predictions and then back-test that strategy across each currency pair to see if it can remain profitable over a 1 or 2 year period.

Some of the strategies running on this demo have 2 years worth of data behind them, overs have 6 months.

I use MetaTrader 4 Expert Advisors to open and close all the trades based on the strategy signals/triggers generated every hour by my system.

Its not a real time system so I can get the last hour of data from MetaTrader.

 

[Intel_Bvc]

Backtesting Example

Thought I'd share the type of back-testing analysis I do on my trading signals.

Below there are 3 screenshots.

Screenshot 1: Shows the spreadsheet that contains every single signal for Threshold strategy on EUR/AUD from July-2012 to April-2013. This equates to 892 trade signals.

Screenshot 2: Shows the table which breaks down the strategy stats by hold time. The left-most column is T1 meaning 1 day hold time, all the way up to the right-most column of T10 which is hold time of 10 days. Important to note that as I commented on earlier today, these stats are based on closed trades and 24-hr intervals. So in actuality on a real world account the Max Draw-down and STD Dev are going to be slightly larger. However the stats still offer a good indication of the performance of the strategy especially when compared to other strategies I have. Also notice the Stop-Loss and Profit Take cells highlighted yellow. These constants are referenced by calculations in the spreadsheet and update the stats depending on the level that's set. 100% effectively says no stop-loss or take profit. 1% would obviously mean stop loss or take profit 1% away from starting price.

Screenshot 3: Shows the total cumulative returns by hold time. As you can see in this instance the optimum hold time is 10 days.

Threshold strategy on EUR/AUD (ignore name of file in screenshot) is definitely the best performing strategy I have. I also actually have 2 years worth of signals and it has consistently performed well over the last 2 years.

Edit - NOTE: Figures are in Cumulative Returns (http://www.investopedia.com/terms/c/cumulativereturn.asp), that means they're absolutes rather than relative to account size.

 

[Intel_Bvc]

First 2 weeks round-up

Yet again the system reacted favourably to the data releases on Friday to finish its first two weeks of trading at $142k, up $22k (18.3%).

The last two weeks have shown that despite this being a completely technically driven system, fundamental events like data releases and press conferences etc can be your friends rather than your enemies. They provide the volatility and catalysts to move price in the direction the Nets anticipated. What you will find is that the data releases always seem to magically be just what the nets were positioned to take advantage of. Yes, of course sometimes they are wrong, but not often.

One mistake I've managed to learn from is that I cannot pick and choose when to trade. On this demo I've managed to let the system run and put on all the trades it chooses to put on and take them off when its the optimum time to do so (determined by statistical analysis).

Over the last 2 weeks the system has shown little regard for the perceived risk of: AUD / GBP / EUR Interest Rate Decisions, USD Non-Farm Payrolls, AUD / GBP / EUR GDP data, GBP and AUD Employment data etc. These are just those given high importance rating, there were tonnes of other releases and meetings.

So here we have a fully automated system, that performs in-line with back-tested stats, that doesn't appear to be overly concerned with data releases, that has approx 65% win rate, average winning trade amount almost double losing trade amount, relatively low max draw-down (under 20%), triple digit projected annual return (which is now easier for some doubters to imagine, considering it just achieved nearly 20% in two weeks).

I've attached the latest equity chart below and wish you all a happy week of trading ahead come Monday.

 

[Vasco da Banana]

Would it be worth running this on several accounts starting a week or two weeks apart?

 

[Intel_Bvc]

What would you be hoping to see from that setup? I'm about to hook it up to a real money account in the next week or so anyway, so I will try to keep those who are interested up to date every now and again.

But this is not the first time I've seen the system perform, its done it on demo's and its done it on real money accounts, hence my total confidence in setting up this thread.

I've attached every single trade from a demo I ran for five weeks back in Jan/Feb this year. I turned a $50k account into +$75k in 5 weeks.

Bear in mind that there were several bugs in the system at this time and several of the strategies have since been replaced or improved. Also the account wasn't really well sized so I thought I'd be less aggressive this time round.

That system still generated approx $5k a week ($50k start)... this latest system can generate approx $9k a week ($120k start).

Everything I said it would do its done so far.

The point of this thread is simple... I want to show you that Neural Nets can and do work. So far, so good.

 

[Intel_Bvc]

Crazy Crazy Crazy last few days!

I actually don't know where to start...

First off I thought I'd stop posting updates every day because I got the impression that me going on about how well the nets are performing doesn't quite seem to be what people want to hear on this forum.

No-one seems to actually openly share positive experiences in this way, just post after post of people complaining and pointlessly debating pointlessness, and the usual "experienced" members of the forum patrolling new threads to post some negative comments as if they actually gain an extra degree of intelligence after each unconstructive & bitchy comment they post.

Anyway... that's my little rant out the way.

So after a promising end to last week and a promising start to this week, the Equity failed to push past the $148k barrier. We then saw major draw-down over the last two days with equity touching $123k at one point.

There has been an awful amount of volatility recently and some big announcements over last two nights in particular seeing the large long GBP/USD position getting smashed back down.

I'm starting to think that I may need to get a little bit smarter with my hold times. Too many instances on this demonstration where I've seen good trades stay open too long and swing back round to ultimately make me a loss. Of course I've said all along these are optimum hold times, but I think I need to look into the possibility of having a variable hold time that obviously has some consistent rule to govern its variance.

One of the reasons why I created this live Demo on this forum was to try and at least partially re-create that feeling of fear similar to if I had been on my real money account. Obviously money isn't involved. But pride is. So instead my fear was of starting up a thread, banging on about my nets and been made to look a fool when they flop. Then all those people going on about how useless Neural Networks are could have a field day dancing all over them and my failure.

So when the account was getting absolutely smashed two days ago, I had that usual dilemma... I had already just pushed past my previous max drawdown, more trades were going on that just looked like throwing bad trade after bad trade, I still had a large volume on that could have easily taken me back to my start balance. I was already imagining the abuse that I would get. But I just did the one thing that I've found so hard to do up till now... I just left it alone.

Before I knew it the losses stabilised and I was able to relax and think about how many times I had been in that exact same position on a real money account and panicked and closed off every trade (at their peak loss!) only for the system to rebound and usually go on a massive rally that would have been in my favour!

So anyway, encouraged by how the system had previously responded to adversity I was quietly confident that going into the major USD announcements yestrdy evening that by hook or by crook the Nets would do what they've always done and come out fighting.

My long GBP/USD again took another hammering, but this time the system seemed well enough hedged (i.e Long USD/CHF) to absorb the worst that could be thrown at it.

After a fairly stable asia-pac session the europe open has unleashed a massive move... and you guessed it, the nets were perfectly positioned to benefit.

Account Equity is currently $138+ up $18k (15%) after nearly 3 weeks.

My real money account will be ready for Monday so I'll prob be cutting this demo soon.

If anyone wants to keep an eye on the system then they can check out the simplified version on collective2 - Neural Net FX - a forex trading system

I hope I've shared some useful material and insight on this thread.

On a personal level, I'm still not where I need to be as a trader and the system is not exactly where it needs to be, but for sure I feel like in all aspects I've taken a few steps forward over the last few weeks.

 

[Intel_Bvc]

As promised I thought I'd post an update on how the system is doing now that I'm trading a real money account.

I had £12k capital but only really needed £6k to service the account/system, so I opened the account and deposited £6k with £6 held back but available if needed further down the line.

So, in a nutshell, over the last 2 months of trading the account itself is up 45%. Funnily enough these sorts of numbers scare people, but end of the day you can sit £18k on that account (although its completely unnecessary) and say it made 15% and it seems people would actually be more comfortable with that...

The current Equity stands at £8800.

I've attached the detailed statement from Metatrader 4.

Over time, and as I cut out my manual trades and interventions (which have hurt my profits rather than helped them), I'd expect the system to settle on its statistically proven win rate of 60-65%. Which as you could imagine would boost the profits further.

Just as I declared when I first joined this forum, the nets will make more money on winning trades than it loses on losing trades. Max Drawdown will be less than 20%. The other piece of the equation now is that win rate improving and moving inline with my expectations.

10 months from now I fully expect to hit that 300% annual return that I claimed (to much scepticism) that the system is capable of delivering. 45% in two months with more room for improvement puts the system firmly in line to achieve this.

 

[Les Moore]

If this works out and you are as successful as this guy...

QIM's Jaffray Woodriff: The monk in managed futures | Institutional Investor

...please come back and tell us the story.

Interesting stuff with the NNs.

Re the above, just so people know, Jaffray averaged about zero return over the next 4 years following that interview.

Kind of almost wonder if Jaffray didn't give away too much in the interview.

 

[Crunchy1]

Oh, I actually own Neuroshell Daytrader Pro

 

[c6ackp]

That's interesting Les - looks like since the recovery started he couldn't trade his way out of a paper bag - maybe his trades were getting too big for the market to stand, who knows... I guess that suggests that you have to keep evolving, even if you have something that produces stellar returns for several years...

Interesting stuff with the NNs.

Re the above, just so people know, Jaffray averaged about zero return over the next 4 years following that interview.

Kind of almost wonder if Jaffray didn't give away too much in the interview.

 

[c6ackp]

BTW: congratulations to Intel

Assuming, these numbers are genuine/rigorous, Intel is the first ever "trader" on trade2win to demonstrate an ability to make money (consistently) from trading forex.

That's a considerable achievement since there are, supposedly, more than 200,000 members