Noxa indicators for Neuroshell

[Noxa]

Hi,

English is not my mother language but i understand that 'artifact' means product.
Yes I agree, when the noise goes up, PF was going up i also noticed this.

My explanation was that CSSA 'curve fits' or learn noise and PF goes up because of higher amplitude and mean reversion. The best test would be test on noise with Hurst
exponent=0.8 i think because this mean reverting is misleading.

Trading Strategies Based On Digital Filters - Page 41 - Forex Trading

post 403

You can generate those things with MATLAB WAVELET toolbox.

Krzysztof

I am sorry you don’t feel that CSSA works. But CSSA does what it does; it identifies cycles that can be exploited; short term coherence in price is detected through variance. If there is coherence in price, we can expect that the cycle will continue, at least for a short while.

Let me quote Pr. CodeBreaker on the ForexFactory forum”

Forex Factory - View Single Post - Optimized Trend Trading
“ On the SSA subject, I rather like SSA, particularly because it is DATA driven whereas nearly all other spectral algorithms (like FFT, Goertzel, MESA etc) are MODEL driven. There are no assumptions in SSA, it just decomposes time series into components and is very good at it. “

Wish you well with your experiments with Goertzel.

Noxa

 

[Noxa]

Hello everybody,

In order to conclude my research about CSSA I would like to point one very important thing which is no clear for me. Optimization of CSSA. Here I compiled answers from NOXA about the subject from different posts

- I periodically halted the optimization (at each new epoch) to avoid over-fitting.

Over-fitting occurs when an optimizer learns the noise in the data. One simple way to avoid it is to periodically halt the optimization. The optimization is then continued from the point at which it was halted. Nothing fancy here! Just a way to give you an access to interim results so that you can make an educated guess on when to stop the optimization!


>> So your procedure of optimization for market conditions is simple adjust parameters, optimize, break optimization from time to time to see that model is making money out of sample, if equity curve is rising than model is OK ?? Is that correct ??

It is not fully correct. The interim results can be many things that can help you assess the validity of your model and find market states. There are many things you can do; for example we look for multiple up and down trending segments in the equity lines and some willingness to reproduce that behavior out-of-sample...

post 127

> How you know when epoch is finished in NS ??

This is one of them; simply watch the fitness clusters for the training and out-of-sample data sets. They will move in unison until they diverge.

Can you make a screenschots of it ??

no answer here

post 128 - the same results which proves how easy is to overfit CSSA

post 133

It seems that you are pretty selective in your answers, you didn't comment post 127 and 128 and answer the question if you connect special tool during optimization.

no answer as well


My opinion is that using just Neuroshell is very easy to overfit CSSA optimization, it happened a lot of times for me see post 128. Optimization envirnoment of Neuroshlell is very simple: off, on, pause and paper trading. So no control of other parameters like "look for multiple up and down trending segments in the equity lines and some willingness to reproduce that behavior out-of-sample... or

"simply watch the fitness clusters for the training and out-of-sample data sets. They will move in unison until they diverge."

So my conclusion is that NOXA is using external optimizer to optimize properly CSSA and avoid overfittig. However I couldn't find this tool as an addon for NS but I found as an addon for Tradestation See link and screenshoots. I think NOXA will never confirm using such tool perhaps due to bussiness relations with Ward or not to show to the people that GA from NS is not advanced enough to optimize CSSA properly.

Power Walk Forward Optimizer

If anybody knows about possibility of using external optimizers with NS let me know.

Krzysztof

We use NeuroShell Trader only. We don't know the product you just mentioned. NeuroShell provides all the settings it tried along with their fitness. We use a proprietary tool that looks at this data and tells us when there might be overfitting. CSSA is not more or less prone to overfitting than any other indicator; it simply contributes to very quick convergence so we have to stop the backtesting early. It can be done manually very easily as I described in one previous post.

Noxa

 

[Noxa]

hi guys!

A question to Noxa, is it possible to give a sample setting for forex?
Just to understand...

Thanks a lot for all this infos!

Bye, AT

I can give you the general lines of a system that happens to work beautifully on the forex. We don't trade forex ourselves but we developed specific indicators for it (available in v1.1). You can contact us at : info at noxapredict dot com

As for precise settings, they mostly depends on the trading strategy. The example that I mentioned above trades congestion ranges which produces very different settings than settings from a contrarian approach as we discussed so far.

Noxa

 

[alby1714]

Farewell

I am sorry you don’t feel that CSSA works. But CSSA does what it does; it identifies cycles that can be exploited; short term coherence in price is detected through variance. If there is coherence in price, we can expect that the cycle will continue, at least for a short while.

Let me quote Pr. CodeBreaker on the ForexFactory forum”

Forex Factory - View Single Post - Optimized Trend Trading
“ On the SSA subject, I rather like SSA, particularly because it is DATA driven whereas nearly all other spectral algorithms (like FFT, Goertzel, MESA etc) are MODEL driven. There are no assumptions in SSA, it just decomposes time series into components and is very good at it. “

Wish you well with your experiments with Goertzel.

Noxa

Hi Noxa,
needless to say your dialogue with Krzysiaczek was extremely interesting . This last post of yours sounds like a farewell to Krzysiaczek and that would be a pity as it would leave many points unanswered, (e.g. the queries made by Krzysiaczek on optimisation) that are of general interest.
Your reply would have provided support and confidence to current and would-be users of your product alike.

Finally with due respect I feel that your quote of Pr. CodeBreaker from the ForexFactory is incomplete as it does not include the whole sentence where CB states that:
"....SSA is non-causal (all the points are calculated from all the other points) so cannot be used directly for trading. "
Sincerely,
alby1714

 

[Krzysiaczek99]

overfit control ??

We use NeuroShell Trader only. We don't know the product you just mentioned. NeuroShell provides all the settings it tried along with their fitness. We use a proprietary tool that looks at this data and tells us when there might be overfitting. CSSA is not more or less prone to overfitting than any other indicator; it simply contributes to very quick convergence so we have to stop the backtesting early. It can be done manually very easily as I described in one previous post.

Noxa

Hi,

At least we know that there is a tool, that's good. Why you don't sell it than ?? For sure it would improve the trading ??

Krzysztof

 

[Krzysiaczek99]

bussines talk ?? fuzzy talk ?? making more smog ??

I am sorry you don’t feel that CSSA works. But CSSA does what it does; it identifies cycles that can be exploited; short term coherence in price is detected through variance. If there is coherence in price, we can expect that the cycle will continue, at least for a short while.

Let me quote Pr. CodeBreaker on the ForexFactory forum”

Forex Factory - View Single Post - Optimized Trend Trading
“ On the SSA subject, I rather like SSA, particularly because it is DATA driven whereas nearly all other spectral algorithms (like FFT, Goertzel, MESA etc) are MODEL driven. There are no assumptions in SSA, it just decomposes time series into components and is very good at it. “

Wish you well with your experiments with Goertzel.

Noxa

Well I don't know what to say it. In all tests we had just one cycle (sin+cos) as a clean signal or with noise. Than I was locking to this cycle and changing noise conditions
but we suppouse still trade with underlying cycle. Unfortunatelly underlying cycle was losing control and we were getting 'crazy' trades than you were trying to adjust parameters but it was simple 'backward curve fitting'. You can do it with any indicator.

So there was no new cycles to exploit, you know it very well, so this answer i consider
as bussines talk or fuzzy talk or making more smog talk as you know that a lot of readers don't understand 100% what is going on and using advanced language confuses them even more.

Anyway it works how it works, if it would be perfect than you would not talk to us and
be zillionare already.....you sell it than it is not perfect, only question is how much ****ed up it is...no test instruction document, no test report, I only found what I found.

Is all the time the same person under NOXA alias ?? How many of you is in NOXA full time ??

Krzysztof

 

[Krzysiaczek99]

first 24h out of sample

first 24h out of sample EURUSD30

Don't ask me why it is like this. Out of sample for Monday/Tuesday done by NS got a hit
80%, today the same settings like in post 180 but real out of sample hit 25%. See screenshots. I will keep those setting till end of the week and let's see.

Overfitting ??? Cheat out of sample test done by NS ??? Cycle died ??

Krzysztof

 

[Noxa]

"C" in CSSA for Causal SSA

Hi Noxa,
needless to say your dialogue with Krzysiaczek was extremely interesting . This last post of yours sounds like a farewell to Krzysiaczek and that would be a pity as it would leave many points unanswered, (e.g. the queries made by Krzysiaczek on optimisation) that are of general interest.
Your reply would have provided support and confidence to current and would-be users of your product alike.

Finally with due respect I feel that your quote of Pr. CodeBreaker from the ForexFactory is incomplete as it does not include the whole sentence where CB states that:
"....SSA is non-causal (all the points are calculated from all the other points) so cannot be used directly for trading. "
Sincerely,
alby1714

Needless to say that CSSA is a causal version of SSA so it can be used directly for trading.

Noxa

 

[Krzysiaczek99]

trade repaint

I wanted to check CSSA performance after next 24 hours and found trade repaint. See screenshots, some trades disappeared on 25th on Feb, strategy the same. It just enough to reformat the chart excluding 26of Feb and missing trades are showing up again. Did anybody seen something like this. I already found trade repaint once see link below.

Krzysztof

Neural Network Trading, Serious people only! - Page 6

post88

 

[Noxa]

I wanted to check CSSA performance after next 24 hours and found trade repaint. See screenshots, some trades disappeared on 25th on Feb, strategy the same. It just enough to reformat the chart excluding 26of Feb and missing trades are showing up again. Did anybody seen something like this. I already found trade repaint once see link below.

Krzysztof

Neural Network Trading, Serious people only! - Page 6

post88

You have to make sure that CSSA's Training Range remains static as new bars come in. The condition is: TrainStart + TrainBars < number of bars in you chart.
I refer you to the section on "Setting the Range" in the Helpfile (see screenshot attached).

Noxa

 

[Noxa]

I wanted to check CSSA performance after next 24 hours and found trade repaint. See screenshots, some trades disappeared on 25th on Feb, strategy the same. It just enough to reformat the chart excluding 26of Feb and missing trades are showing up again. Did anybody seen something like this. I already found trade repaint once see link below.

Krzysztof

Neural Network Trading, Serious people only! - Page 6

post88

It looks like your Training Range for CSSA is correct:
Training Range = 1000 bars
Number of bars in your chart = 25 * 48 = 1200 bars

So the condition 1000 < 1200 is verified.

I thought that something went wrong with your training period in NSDT but after rechecking everything looks OK.

Ideas:
changes in your historical data?
changes in your indicator settings?

Noxa

 

[Krzysiaczek99]

repaint continuation

Hi,

Everything seems to be fine, strategy and parameters the same, just a new data. I attach 2 charts e30_25 with trades on 25th of Feb and e30_2526 trades on
25th and 26th of Feb - clear difference. Data is saved, if you want csv separatelly
let me know.

Only what i found is that right click/format chart and than change date to cut the last day don't work well in NS, on one PC where i have RT data feed it works fine, but when i moved chart with data saved to another i could't cut last day by reformatting - it was always aligning to the end of data.

Anyway I dedected simply by comparing the chart which i posted and current chart.

Krzysztof

 

[Krzysiaczek99]

MACD no repaint

here is MACD screenshot on the same data and no repaint - trades and equity curve are matching. So NS fault we can exclude here.

Krzysztof

 

[Noxa]

Hi,

Everything seems to be fine, strategy and parameters the same, just a new data. I attach 2 charts e30_25 with trades on 25th of Feb and e30_2526 trades on
25th and 26th of Feb - clear difference. Data is saved, if you want csv separatelly
let me know.

Only what i found is that right click/format chart and than change date to cut the last day don't work well in NS, on one PC where i have RT data feed it works fine, but when i moved chart with data saved to another i could't cut last day by reformatting - it was always aligning to the end of data.

Anyway I dedected simply by comparing the chart which i posted and current chart.

Krzysztof

Can you send the csv files?

Try to set the training ranges to low values and see how it goes.

Noxa

 

[Krzysiaczek99]

csv

Is here. I will try, have to go somewhere now, will be back in a few hours.

Krzysztof

 

[Krzysiaczek99]

600 bars and disappeared

Is here. I will try, have to go somewhere now, will be back in a few hours.

Krzysztof

for 600 bars repaint disappeared.

Krzysztof

 

[Krzysiaczek99]

repaint

It looks like your Training Range for CSSA is correct:
Training Range = 1000 bars
Number of bars in your chart = 25 * 48 = 1200 bars

So the condition 1000 < 1200 is verified.

I thought that something went wrong with your training period in NSDT but after rechecking everything looks OK.

Ideas:
changes in your historical data?
changes in your indicator settings?

Noxa

I think it's this problem. Chart had only 906 bars - weekends, not 1200.

Additionally when you increase out of sample period and because of this out of sample + TrainBars > number of bars in your chart ==> future leak occurs ==> repaint.

So training window does not readjust to those changes it seems, when it is not enough bars in the chart remains static, it's necessary to be carefull. Anyway, it's good that's only this.

Krzysztof

 

[Noxa]

I think it's this problem. Chart had only 906 bars - weekends, not 1200.

Additionally when you increase out of sample period and because of this out of sample + TrainBars > number of bars in your chart ==> future leak occurs ==> repaint.

So training window does not readjust to those changes it seems, when it is not enough bars in the chart remains static, it's necessary to be carefull. Anyway, it's good that's only this.

Krzysztof

Good!

Now I want to draw your attention to Future Leaks. They are something different; they do not induce repainting but occur when the Training Range for CSSA is set larger than the Training data used by NSDT; you have to make sure that this does not happen as well; you can use ShowRange for that.

Noxa

 

[Noxa]

Is here. I will try, have to go somewhere now, will be back in a few hours.

Krzysztof

I looked at your data. These are just raw signals but I think they could be used to produce a trading strategy . I was able to produce similar signals with two different Training ranges (1~435, 1~600). This is encouraging. I am just a bit concerned by the amount of data you have. The cycles are pretty slow and it seems to me that you would gain in robustness if you had more data.

Noxa

 

[arryex]

Trading strategy procedure discovered???

Hi All,

After verifying the data, I found that my forex data is different with the one given by Krzystof, I use ITFX (1 spread). The trading strategy optimization result will be different due to different input data.

I compare the result using original setting that noxa give on above post#199. I used full optimization, using larger range of m-history, gstart and gdepth keep linked for all inputs, Trainstart=1, trainbars=600 (less than the bar no for forward test), all other parameters as default and free for optimization respectively.

During the optimization, I stopped sometime to check if the forward test result provide the best ouputs (i.e.at least, positive Profit ≥ $1000, Return on Trades ≥1000%, Ret on Account ≥ 200%), Ratio Gross Profit/Loss≥ 2.2, %Profitable Trades ≥ 70% at forward test, Required Account Size (as suitable with my account), Max Drawdown and Max Open Trade Drawdown, equity/profit grows up during optimization and forward test.)

I put the Cost data in Trading Strategy wizard based on my trading account type, mini lot (0.1 lot), spread=1 pip, entry/exit commission ($2/per trade), 4 digit pair EUR/USD (point value = 10,000), margin $100 for any position made. Any brokers have different spread, commission, etc need a different cost parameters. Rebacktest trading strategy as necessary if the objective is not achieved. Optimization more than 2 hrs (depend on the CPU speed and memory).

Hope useful for anyone who has been trying to find a trading strategy using noxa, especially CSSA-cycle. I will continue evaluate this trading strategy whether resulting a good profit with upcoming new data.

Any advice is appreciated.

Good luck,
Arry