windsurfing_stew
Member
- Messages
- 77
- Likes
- 1
Make sure you keep the start and end dates the same too.
Noxa indicators for Neuroshell
- Thread starter val2004
- Start date
- Watchers 71
supremegizmo
Active member
- Messages
- 219
- Likes
- 86
thanks i tried its improved the results ... i first had match chart selection but now i have used specific dates .. one other thing may be neuroshell loaded from a demo mt4 account to real mt4 account may have caused a bit change in data when i loaded the charts.. this nets are some data sensitive stuff .. cheers wind and high freq
windsurfing_stew
Member
- Messages
- 77
- Likes
- 1
Also worth noting that running the same test on two different machines may produce different solutions because -
1) Often computationally expensive operations are optimised for different chipsets - eg AMD vs Intel.
2) The seed may be different each run.
1) Often computationally expensive operations are optimised for different chipsets - eg AMD vs Intel.
2) The seed may be different each run.
Hi,
If you are using prediction strategy then you will use predicted values as input of your strategy.
Example if output prediction is percent change of open (5 bar in future), then your prediction strategy will generate buy signal if predicted value > actual value and sell signal if predicted value< actual value. Actual value means current percent change of open.
If you are using trading strategy, then Neuroshell trader will optimize input parameter and set point to get higher profit or other strategy output (minimize drwdown,etc). Example if you make a trading strategy based on MACD(Close, ema1, ema2, ema3) cross over zero. Then Neuroshell will optimize the value of ema1, ema2, ema3 as well as threshold value (not always zero).
One thing you should know that you can use predicted values as your trading strategy inputs. You can get predicted values by making a prediction and select No Trading Position on Predictions Parameters- Position.
Good luck
If you are using prediction strategy then you will use predicted values as input of your strategy.
Example if output prediction is percent change of open (5 bar in future), then your prediction strategy will generate buy signal if predicted value > actual value and sell signal if predicted value< actual value. Actual value means current percent change of open.
If you are using trading strategy, then Neuroshell trader will optimize input parameter and set point to get higher profit or other strategy output (minimize drwdown,etc). Example if you make a trading strategy based on MACD(Close, ema1, ema2, ema3) cross over zero. Then Neuroshell will optimize the value of ema1, ema2, ema3 as well as threshold value (not always zero).
One thing you should know that you can use predicted values as your trading strategy inputs. You can get predicted values by making a prediction and select No Trading Position on Predictions Parameters- Position.
Good luck
I'm new to Neuroshell and I thought in the prediction wizard you could add what indicators you wanted to use in a trading system and it would figure out of those indicators what to use and the entries and exits? Basically you throw in the indicators you want to use and prediction wizard will formulate the best trading system?
I feel like of barley scratched the surface of Neuroshell, but I'm really liking it so far.
You are right..you can fell free to put whatever prediction you want and inputs used to predict, Neuroshell can find for you the best input and its parameter, if input not correlated or have no impact on prediction then it will be removed. As well as other neural network software, Neuroshell will predict output based on learning parameter you gave (prediction output, inputs, training range, out of sample, etc).
Simple analogy, you train a network to predict/classify apple or banana and your inputs are width, length and color. If your input during the training period you gave are only red for apple and yellow for banana, then your prediction may get false if during actual test there are some other colors come such as green apple, green banana or red banana. Your prediction may get wrong if actual width and length during the test are over than the range you gave during training.
In this case, you should improve your inputs, example width, length, width/length, all colors available for apple and banana. This will improve your prediction/classification.
Consider also the training range, if we train our net with all available size of apple (e.q. Washington's apple, Indonesia's apple, China's apple) and banana (brazil's banana or South asia's banana) then your net prediction will be better than restricted size (only Washington apple or brazil's banana).
On real financial prediction, we will have similar case, you need to collect all possible inputs that influence the output prediction. Otherwise your net will not responded or give false prediction if your input are out of the training range or not have similar characteristic as during the training.
For me, neural network is not a technical thing ...but also an art...
Simple analogy, you train a network to predict/classify apple or banana and your inputs are width, length and color. If your input during the training period you gave are only red for apple and yellow for banana, then your prediction may get false if during actual test there are some other colors come such as green apple, green banana or red banana. Your prediction may get wrong if actual width and length during the test are over than the range you gave during training.
In this case, you should improve your inputs, example width, length, width/length, all colors available for apple and banana. This will improve your prediction/classification.
Consider also the training range, if we train our net with all available size of apple (e.q. Washington's apple, Indonesia's apple, China's apple) and banana (brazil's banana or South asia's banana) then your net prediction will be better than restricted size (only Washington apple or brazil's banana).
On real financial prediction, we will have similar case, you need to collect all possible inputs that influence the output prediction. Otherwise your net will not responded or give false prediction if your input are out of the training range or not have similar characteristic as during the training.
For me, neural network is not a technical thing ...but also an art...
I been using EOD for trading stocks and I've been using 3-4 years for optimization and 1 year paper trading. It seems if I go longer I don't get enough trades and if I go shorter I get too many trades.
On another note 6 months ago I set up some Charts and stopped using Neuroshell and I went back this week and opened them and my optimization to my surprise produced good results. I though it would use too much curve fitting but I was wrong.
Just ensure that training data:
- already cover the all possible market trend (up, down, side ways), even shorter data range should sufficient if consist all possible market moves.
- curve fitting on larger range should be better than in a shorter range.
we can proof how good our prediction only by OOS data in all possible case. One model may good work on an strong bullish market but worst during side way or down trend. Some new user was happy at the beginning but bad feeling after market crash..so probably need more time to do OOS test...
Happy trading
- already cover the all possible market trend (up, down, side ways), even shorter data range should sufficient if consist all possible market moves.
- curve fitting on larger range should be better than in a shorter range.
we can proof how good our prediction only by OOS data in all possible case. One model may good work on an strong bullish market but worst during side way or down trend. Some new user was happy at the beginning but bad feeling after market crash..so probably need more time to do OOS test...
Happy trading
Your right, I don't trade below the 200 MA. I try to do what works and not get fancy.
bismillahwd
Newbie
- Messages
- 7
- Likes
- 0
why all of this great topic stopped 
donkingdon
Newbie
- Messages
- 9
- Likes
- 0
I am very grateful to the Admin here, I was greatly helped at all ... and especially on this forum ... because I want to invest in the world with the help of NN Treding and I'm still ....
donkingdon
Newbie
- Messages
- 9
- Likes
- 0
I still find it difficult to NOXA if anyone could help me because I might see noxsa excellent for a variety of characters for Treding market, so I want to deepen about Noxa if anyone could help me ....
Hi All,
Long time not to see this thread..
NOXA is using the similar algorithm Singular Spectrum Analyzer (SSA), it is one of few other available software. You may found some useful information on the following link:
Time series analysis and forecasting, Caterpillar SSA method
Singular spectrum analysis - Wikipedia, the free encyclopedia
SSA-MTM Toolkit SSA-MTM Toolkit is freeware but in UNIX and MAC OS only
SSAwiki | Main / Software
http://www.pdmi.ras.ru/~theo/autossa/ if you have capability to matlab programming
The idea is to catch the useful data and separate from its noise by decomposing.
Long time not to see this thread..
NOXA is using the similar algorithm Singular Spectrum Analyzer (SSA), it is one of few other available software. You may found some useful information on the following link:
Time series analysis and forecasting, Caterpillar SSA method
Singular spectrum analysis - Wikipedia, the free encyclopedia
SSA-MTM Toolkit SSA-MTM Toolkit is freeware but in UNIX and MAC OS only
SSAwiki | Main / Software
http://www.pdmi.ras.ru/~theo/autossa/ if you have capability to matlab programming
The idea is to catch the useful data and separate from its noise by decomposing.
Hi Vgerman,
I verified that even Noxa CSSA actually is repainted..see attached verification result.
The main reason is because the base of SSA is calculating of variance and covariance of a time series, hence the values are changed anytime the data are changes including new data included.
I found the scilab guide about SSA from here
http://129.199.13.40/IMG/file/DavidPDF/SSA_beginners_guide_v9.pdf
Scilab is an open source which having capabilities like Matlab.
Now I am verifying what is the impact of repainted SSA on trading strategy, I will put the result when available.
I verified that even Noxa CSSA actually is repainted..see attached verification result.
The main reason is because the base of SSA is calculating of variance and covariance of a time series, hence the values are changed anytime the data are changes including new data included.
I found the scilab guide about SSA from here
http://129.199.13.40/IMG/file/DavidPDF/SSA_beginners_guide_v9.pdf
Scilab is an open source which having capabilities like Matlab.
Now I am verifying what is the impact of repainted SSA on trading strategy, I will put the result when available.