Extreme scalping using neural networks - EA/Manual | Page 30 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Any success with neural networks in trading 29 replies

Neural Networks: Myths And Reality 19 replies

Cognitive indicators and neural-networks 19 replies

Neural Networks in Trading (Java, Algodeal, Technical Analysis) 3 replies

Price Analysis with Neural Networks 41 replies

Attachments: Extreme scalping using neural networks - EA/Manual
Exit Attachments

Extreme scalping using neural networks - EA/Manual

  • #581
  • Jul 7, 2010 2:43am Jul 7, 2010 2:43am
Quoting Kosomolate
Hi all!

see this result, according to scooby's settings, and version 155, alpari uk.

It was placing trades many on one candle, I thought it was supposed to predict the close of the bar and so the trade should be one per candle. Anyway I am not complaining with what it did, if it continues this way...

kos
Yup. I didn´t catch on all of this system too so far. Some things are still strange to me. Well, at least its seems to work now. Had a lot of trouble attaching it to a chart until it finally triggered any trades.

I would like to know more about the backgrounds of this system like

Explaining how the neural network works with some more detailed informations on it (it´s quite new to me) ?

Explaining what timeframe is best for trading it. ?

How long got this system tested? Backtests/Forward Tests?

If anyone´s trading this live allready i would be interested in for how long and what you got in performance.
  • #582
  • Jul 7, 2010 3:45am Jul 7, 2010 3:45am
And how you avoid "overtraining" the neural network @neuro?. Even if you have contant profits, market conditions will change and signals instantly triggered then are start generating constant loss :/
  • #583
  • Jul 7, 2010 3:57am Jul 7, 2010 3:57am
Quoting lddd
And how you avoid "overtraining" the neural network @neuro?. Even if you have contant profits, market conditions will change and signals instantly triggered then are start generating constant loss :/
You should read this article and then put NEXTCLOSE in the bin.

http://www.forexfactory.com/showthread.php?t=9805
  • #584
  • Jul 7, 2010 4:24am Jul 7, 2010 4:24am
Quoting scooby-doo
You should read this article and then put NEXTCLOSE in the bin.

http://www.forexfactory.com/showthread.php?t=9805
what you mean with putting it in the bin? Aren´t you trading NextClose by your self? Whats your conclusion?
  • #585
  • Jul 7, 2010 4:34am Jul 7, 2010 4:34am
Quoting lddd
what you mean with putting it in the bin? Aren´t you trading NextClose by your self? Whats your conclusion?
My conclusion is that it doesnt work and certainly wont work on a LIVE account.
  • #586
  • Jul 7, 2010 4:56am Jul 7, 2010 4:56am
Quoting scooby-doo
My conclusion is that it doesnt work and certainly wont work on a LIVE account.
well, why you pointed to this thread then? did you read it your sellf? i think there is no automated system that will generate constant profits when you don´t have any idea of howto manage your risk/money.
  • #587
  • Jul 7, 2010 5:20am Jul 7, 2010 5:20am
for installer this system i put file nextclose 156.ex4 in my folder expert on metatrade ??? It's ok?
  • #588
  • Jul 7, 2010 5:32am Jul 7, 2010 5:32am
Quoting Yu1976
for installer this system i put file nextclose 156.ex4 in my folder expert on metatrade ??? It's ok?
no you have to put it in the bin, ask scooby
  • #589
  • Jul 7, 2010 6:23am Jul 7, 2010 6:23am
Quoting scooby-doo
You should read this article and then put NEXTCLOSE in the bin.

http://www.forexfactory.com/showthread.php?t=9805
The article doesn't say NNs and AI don't work. It says that they're not magic, not absolutely precise and not a replacement for TA.

Here's my take on the article and NNs in general. If patterns exist that have a statistical edge, the NNs are good at finding them. That's all. Much depends on the input to the NN. Don't forget the old computer science acronym GIGO.

Regarding the NextClose EA, I have my doubts that it has an edge. I'm planning on doing some science experiments in the next few weeks to see if it actually does. The core concept of NextClose is to feed the last 16 normalized bar close prices into the NN and out comes a normalized prediction of the next close price. I'm not sure this is a reasonable NN approach, hence the science experiments.

A NN would be very good at finding candlestick patterns with an edge. These 2 links have very good info on the probabilities for various candlestick patterns:
http://thepatternsite.com/CandleContinue.html
http://thepatternsite.com/CandleReverse.html
These "top 10" candlestick patterns have high probabilities for daily stock prices. Do these patterns apply to M15 bars on EURUSD? Maybe, maybe not. But *some* patterns probably *do* apply.

You could do normal statistical studies to find these patterns, or you could use a NN to find them. It's straightforward to program a NN to find these patterns and give a useful output. The beauty of a NN is it's constantly adapting to changing conditions. So as some patterns "stop working", others may become more statistically significant. The NN can adapt to this.

Could you just do normal statistical studies periodically to adapt your system to changing conditions? Of course. But NNs provide another way to accomplish the same thing, that's all.
You are in a maze of twisty little passages, all alike.
  • #590
  • Jul 7, 2010 6:38am Jul 7, 2010 6:38am
Quoting stevegee58
The article doesn't say NNs and AI don't work....
well thats exactly what i think. a NN driven system is no money making mashine like every other system isn´t too. Otherwise we all would be rich and banks doesn´t had to employ traders anymore. I also think it still needs adapt changing market conditions. Even if NextClose would make a contant profit it could fail some day, so you have to watch it and work on it like you have to struggle with all in life, nothing is for free. you have to use your brain too, at least a little bit
  • #591
  • Jul 7, 2010 6:59am Jul 7, 2010 6:59am
Here's something else to consider regarding the NextClose EA: The initial SL is 140 pips away. Everyone's been fixated on the NN implementation of the EA, but this 140 pip SL always bothered me. Most of the time, the profit is taken about 10 or 20 pips from the open. Much of this EA's high probability comes from the simple fact that the risk:reward is about 10:1. I'll bet if you took random entries with a 10:1 SL:TP you'd have a 90% system too. Of course, you'll occasionally hit that 140 pip loss. If the system doesn't have an edge, you'll gradually lose money over time even though the system has a 90% hit rate.
You are in a maze of twisty little passages, all alike.
  • #592
  • Jul 7, 2010 7:03am Jul 7, 2010 7:03am
stevegee are you using the defaults on sl/tp ?
  • #593
  • Jul 7, 2010 7:48am Jul 7, 2010 7:48am
Quoting lddd
stevegee are you using the defaults on sl/tp ?
I'm using default settings on an IBFX 5 digit demo account.
You are in a maze of twisty little passages, all alike.
  • #594
  • Jul 7, 2010 8:10am Jul 7, 2010 8:10am
Quoting stevegee58
I'm using default settings on an IBFX 5 digit demo account.
kewl, same as here. for how long you trading it allready?
  • #595
  • Jul 7, 2010 8:40am Jul 7, 2010 8:40am
Quoting lddd
kewl, same as here. for how long you trading it allready?
First trade was 2010.07.01, so it's been 1 week.

Here are the basic stats:

14 trades
12 winners, 2 losers (86% win)
Average win: $8.66
Average loss: -$11.49
Largest win: $17.00
Largest loss: -$19.00
Net profit: $80.96
You are in a maze of twisty little passages, all alike.
  • #596
  • Jul 7, 2010 9:15am Jul 7, 2010 9:15am
Quoting stevegee58
First trade was 2010.07.01, so it's been 1 week.

Here are the basic stats:

14 trades
12 winners, 2 losers (86% win)
Average win: $8.66
Average loss: -$11.49
Largest win: $17.00
Largest loss: -$19.00
Net profit: $80.96
loel, what pair/timeframe you trading? I am trading it on EUR/USD (M30) for not even 8 hours, so 1 day and allready had 42 trades triggered...
  • #597
  • Jul 7, 2010 9:23am Jul 7, 2010 9:23am
Quoting lddd
loel, what pair/timeframe you trading? I am trading it on EUR/USD (M30) for not even 8 hours, so 1 day and allready had 42 trades triggered...
Timeframe H1, EURUSD.

One quirk about neural networks is that they don't all train the same, even with the same training data set. The reason is that the neurons have to be initialized to randomized values. So the initial state of the NN isn't going to be the same for you and me. It's not deterministic and *will* affect performance.
You are in a maze of twisty little passages, all alike.
  • #598
  • Jul 7, 2010 9:38am Jul 7, 2010 9:38am
Quoting stevegee58
Timeframe H1, EURUSD.

One quirk about neural networks is that they don't all train the same, even with the same training data set. The reason is that the neurons have to be initialized to randomized values. So the initial state of the NN isn't going to be the same for you and me. It's not deterministic and *will* affect performance.

oh kay understand. thanks for that information. And why trading H1. Is this cause of your prefared trading style or are you expecting better performance on a longer timeframe, could you explain?
  • #599
  • Jul 7, 2010 9:52am Jul 7, 2010 9:52am
Quoting lddd
oh kay understand. thanks for that information. And why trading H1. Is this cause of your prefared trading style or are you expecting better performance on a longer timeframe, could you explain?
My backtesting had better results on H1 than on M30.
You are in a maze of twisty little passages, all alike.
  • #600
  • Jul 7, 2010 9:54am Jul 7, 2010 9:54am
Quoting stevegee58
My backtesting had better results on H1 than on M30.
ok, thanks.
0 traders viewing now