Noxa indicators for Neuroshell

This is a discussion on Noxa indicators for Neuroshell within the Trading Software forums, part of the Commercial category; Guys, Noxa CSSA can also be used to smooth oscillators with no lag! This should be quite appealing to technical ...

Reply
 
LinkBack Thread Tools Search this Thread
Old Nov 21, 2008, 6:29pm   #61
Joined Apr 2008
Lag free smoother with CSSA!!

Guys,

Noxa CSSA can also be used to smooth oscillators with no lag! This should be quite appealing to technical analysts.

In the chart posted we use CSSA Cycle to smooth a 5bar RSI (see red line). CSSA has been trained over 4000 bars as shown in the bottom graph (green area) making sure that the indicator remains static (see settings).

Pretty convincing isn’t it?

Noxa
Attached Thumbnails
intc1-1.gif   intc1-2.gif  
Noxa is offline Software vendor   Reply With Quote
Old Jan 29, 2009, 6:17pm   #62
 
8 Posts
Joined Dec 2005
Quote:
Originally Posted by Noxa View Post
I guess I left you speechless with my last post ;-). Here is one tip on using CSSA Flatlines as an on/of switch.

CSSA Flatlines (available in Noxa CSSA v1.1 Beta4.0.3) consists of an upper and lower line, demarcating the + and – thresholds of a congestion range within which you may not want to take any trades. You can use price entry into this range as an off switch so you can get out in time and price exit of either the upper line or lower line from within the range as an on switch to anticipate a breakout. Entries can then be a simple crossover of a shorter m-Histories cycle with a longer one.

BTW, this model is very profitable!

Happy trading.

Noxa
Noxa Analytics Inc :: Your Predictive Edge
Hi Noxa ,

I cant' find any info regarding your CSSA Flatlines in your website.Can you give some additional info and some pictures? Thanks in advance.
mystified is offline   Reply With Quote
Old Jan 29, 2009, 9:09pm   #63
Joined Apr 2008
Quote:
Originally Posted by mystified View Post
Hi Noxa ,

I cant' find any info regarding your CSSA Flatlines in your website.Can you give some additional info and some pictures? Thanks in advance.


See my post 08-09-2008
http://www.trade2win.com/boards/atta...latlines-1.gif
http://www.trade2win.com/boards/atta...latlines-2.gif

I can send you this update on request.

Noxa
Noxa is offline Software vendor   Reply With Quote
Old Feb 6, 2009, 12:37am   #64
Joined Apr 2008
53% yearly return on the DIA

You'll be surprised just how many CSSA cycles are capable of surviving changing market conditions.

This system is simple but still manages to generate 107% return out-of-sample (or 90.3% with $20 round trip commissions) with 65.5% profitable trades over the last two years.

As you can see in graph#1 the equity line (blue curve) is strong and uptrending despite the current meltdown.

Instrument: DIAMONDS TRUST SER 1 (PCX: DIA), daily bars
Trading rules: see Graph#2
Statistics: see Graph#3

Training was done over the first 2500 bars as shown in Graph#2 (last parameter)
Testing starts at bar#2300 ensuring that the results is out-of-sample (see bottom window in Graph#1)

Noxa
Attached Thumbnails
graph-1.gif   graph-2.gif   graph-3.gif  

Noxa is offline Software vendor   Reply With Quote
Old Feb 7, 2009, 3:40pm   #65
Joined Feb 2008
Discovered problems

Hello everyone, Sorry so long not posting here due to my unavailability.

Here I attached some founding while using noxa. Hope anyone can provide some advice in order getting better result.

Any comments are appreciated.

Cheers,
Arry
Note: One in English and one in Indonesian version (may helpful for some of my friends)
arryex is offline   Reply With Quote
Old Feb 8, 2009, 7:08pm   #66
Joined Jan 2009
optimzation time

Hi,

This is a JMACORSSOVER strategy results. Strategy is posted here. It does not use
NOXA indicators but I think optimization problem is common.

Neural Network Trading, Serious people only! - Page 6

This strategy uses filter advance technique (LEAD function) so buy and sell
were printed by NS with 10 bars delay. In order to obtain real time results i simply moved delayed signals by 10 bars (LAG function). I optimized this strategy for 1min and 5min charts with different optimization periods from 1 till 5 weeks and 1 day out of sample trading.

Here I post statistic summary to spark a discussion about optimization periods and methods as it seems to be crucial to obtain good results. Maybe someone knows a procedure how to do it in the most efficient way. Statistic below has following meaning w1,w2...- weeks of optimization, first result- 1 year account return, second result - number of profitable trades.

It is visible here that for 1min this strategy in inefficient due to fixed JMA periods perhaps adjusted to 5 min TF

1 Minute 10B delay
w1 1201% 73,3%
w2 1423% 77%
w3 1231% 73,3%
w4 1231% 73,3%
w5 1231% 73,3%

5 Minute 10B delay
w1 260% 60%
w2 310% 60%
w3 260% 60%
w4 310% 60%
w5 310% 60%

1 Minute RT trading
w1 -335% 25%
w2 -317% 16,7%
w3 26,1% 42,9%
w4 -329% 32%
w5 -301% 41.2%

5 Minute RT trading
w1 165% 76,9%
w2 74,5% 58,3%
w3 85,6% 75%
w4 137,7% 64,3%
w5 81,9% 61,5%

10 Bars delay results are not so important here, I just printed them for information.

It looks that the best results out of sample ocured for 1min for 3 weeks optimization
and for 5 min for 1 than 4 weeks optimization. Optimization time was fixed for 5 min
in both cases. Any idea how to approach this problem ??

Krzysztof
Attached Files
File Type: zip JMA stat.zip (26.8 KB, 406 views)
Krzysiaczek99 is offline   Reply With Quote
Old Feb 9, 2009, 10:12pm   #67
Joined Apr 2008
Setting the Training Range in CSSA

Quote:
Originally Posted by arryex View Post
Hello everyone, Sorry so long not posting here due to my unavailability.

Here I attached some founding while using noxa. Hope anyone can provide some advice in order getting better result.

Any comments are appreciated.

Cheers,
Arry
Note: One in English and one in Indonesian version (may helpful for some of my friends)


Arry,

Setting the range depends on your intent: you can select a portion of data that emphasizes a feature you are interested in. If you trade breakouts for instance, you would want to select a range of past data which exhibits breakouts to learn from. If you trade peaks and troughs, you would select a range with cyclical content. Same for trends and volatility… It has less to do with selecting the right length of data than selecting the proper data.

You can also select all the data available in your training set. It works just fine but with the difference that cycles are now inferred from the average variance in the data. You can then use the fact that some cycles can become irrelevant when they explain less of the variance to identify market regimes. Let me explain.

In our example post#64, we have set the following training range:
TrainStart = 1
TrainBars = 2500

BTW, it is mandatory that TrainStart + TrainBars < number of bars in the training period to avoid future leaks. Moreover, it is mandatory that TrainStart + TrainBars < number of bars in your chart to prevent the training range from changing as new bars arrive; if the training range were to change, the indicator would be recalculated at each new bar then triggering unwanted changes in signals.

Early 2007, our Shannon Entropy indicator (from the NEI set, see red curve on graph attached) exhibited an unprecedented drop giving us a strong indication that the market was about to change; and the current crisis situation proves us right! We then looked for a cycle that was not working well during this resuming period. There is no reason to believe that a cycle should keep performing well as the market conditions change. A corollary to this is that a cycle that was not performing well during a market regime is more likely to perform well as a new regime settles down. As you can see in the third chart from the top, we chose a cycle whose equity line kept dropping from end 2003 to early 2007. You can also see a correspondence in the volatility of the cycle (top graph). Fortunately to us, it keeps performing well since then.

I got to run… I hope I addressed some of your concerns on how to set the training range. You also have a glimpse of what I’ll be talking about when I am back; the assumption that something that worked well in the past should keep working well in the future is more than questionable and I'll describe what we can do about it with CSSA…

Noxa
Attached Thumbnails
fitting-market-regimes-cssa-cycles.gif  
Noxa is offline Software vendor   Reply With Quote
Old Feb 10, 2009, 2:26am   #68
Joined Jan 2009
NOXA indicators performance

Hello everybody,

I read this thread and NOXA indicators sounds pretty interesting. First thought which came to my mind was how do they really perform ?? Than I made a test. I created three strategies, 2 based on NOXA entries 0 and 2 and third one based on MACD as a reference point. All strategies were applied to the same time period of optimization with GA (1 week), 1 day of paper trading and 1 one day out of sample. Here are the results
Attached Thumbnails
n1.jpg   n2.jpg   n3.jpg  

n4.jpg   n5.jpg  
Krzysiaczek99 is offline   Reply With Quote
Old Feb 10, 2009, 2:36am   #69
Joined Jan 2009
NOXA indicators performance

The results are in order: Entries 0, Entries 2 and MACD. Last picture shows MACD settings.
All strategies are attached.

So conclusion is that NOXA buy/sell signlas from Long/Short Entries 0 & 2 perform much worse than simple MACD strategy. All indicators parameters for NOXA and MACD were chosen by GA, maybe manual setting would be more efficient.

Krzysztof
Attached Files
File Type: zip NOXA E0.zip (28.1 KB, 468 views)
Krzysiaczek99 is offline   Reply With Quote
Old Feb 10, 2009, 2:37am   #70
Joined Feb 2008
To compare performace of Trading Strategy

Hi Kryzs,

If you would like to make a comparison of equity for several Trading Strategy, simply by insert new indicator, select Trading Strategy:system information, All Trades (SystemEquity), select all trading strategy that you wish to compare, modify the type of line if necessary.

Cheers,
Arry
arryex is offline   Reply With Quote
Old Feb 10, 2009, 4:37am   #71
Joined Feb 2008
Readjust the Gstart and GDepth

Hi Krzys,

I am lucky both of us working in the same forex data.

I found that the Trading Strategy can be improved by adjusting noxa parameter, I get different result.
On my new TS Noxa E0a, the GDepth adjusted to the end period of paper trade, optimization changed to Maximize ROA*ECC, all other parameters kept.
On my new TS Noxa-E2a, adjusting GDepth not succeed then I try to enlarge m-history, optimization kept maximize ROA, all other parameters kept.

You can see till the walk forward 1 day test both Noxa E0a and E2a resulting better even compare to MACD (see on the snapshot view of system equity , let see if further real test will giving better result.

regards,
Arry
Attached Thumbnails
better-setting-parameter-result.png  
arryex is offline   Reply With Quote
Old Feb 10, 2009, 5:08am   #72
Joined Feb 2008
Which training range to be selected?

Hi Noxa,

Refer to your chart, should we use Noxa Entrophy to define the training and walk forward test range?

Is there any other Noxa CSSA indicators can help to define these ranges?

Thank you,
Arry

Quote:
Originally Posted by Noxa View Post
Arry,

Setting the range depends on your intent: you can select a portion of data that emphasizes a feature you are interested in. If you trade breakouts for instance, you would want to select a range of past data which exhibits breakouts to learn from. If you trade peaks and troughs, you would select a range with cyclical content. Same for trends and volatility… It has less to do with selecting the right length of data than selecting the proper data.

You can also select all the data available in your training set. It works just fine but with the difference that cycles are now inferred from the average variance in the data. You can then use the fact that some cycles can become irrelevant when they explain less of the variance to identify market regimes. Let me explain.

In our example post#64, we have set the following training range:
TrainStart = 1
TrainBars = 2500

BTW, it is mandatory that TrainStart + TrainBars < number of bars in the training period to avoid future leaks. Moreover, it is mandatory that TrainStart + TrainBars < number of bars in your chart to prevent the training range from changing as new bars arrive; if the training range were to change, the indicator would be recalculated at each new bar then triggering unwanted changes in signals.

Early 2007, our Shannon Entropy indicator (from the NEI set, see red curve on graph attached) exhibited an unprecedented drop giving us a strong indication that the market was about to change; and the current crisis situation proves us right! We then looked for a cycle that was not working well during this resuming period. There is no reason to believe that a cycle should keep performing well as the market conditions change. A corollary to this is that a cycle that was not performing well during a market regime is more likely to perform well as a new regime settles down. As you can see in the third chart from the top, we chose a cycle whose equity line kept dropping from end 2003 to early 2007. You can also see a correspondence in the volatility of the cycle (top graph). Fortunately to us, it keeps performing well since then.

I got to run… I hope I addressed some of your concerns on how to set the training range. You also have a glimpse of what I’ll be talking about when I am back; the assumption that something that worked well in the past should keep working well in the future is more than questionable and I'll describe what we can do about it with CSSA…

Noxa
arryex is offline   Reply With Quote
Old Feb 10, 2009, 11:00pm   #73
Joined Apr 2008
Quote:
Originally Posted by Krzysiaczek99 View Post
The results are in order: Entries 0, Entries 2 and MACD. Last picture shows MACD settings.
All strategies are attached.

So conclusion is that NOXA buy/sell signlas from Long/Short Entries 0 & 2 perform much worse than simple MACD strategy. All indicators parameters for NOXA and MACD were chosen by GA, maybe manual setting would be more efficient.

Krzysztof
Hello Krzysztof,

Don’t take it personally but comparative examples such as yours tell me more about the inability of a user to do auto-optimization than the predictive skills of an indicator. If not done properly, the odds favoring good auto-optimization are vanishingly small.

Check out these intraday examples of what you should expect from CSSA: Noxa Indicators

If there is one piece of advice I could give to non experts, it would be to try to optimize the figure manually first as you suggest. We found it already very useful just to look at the cycles. In fact, if you are an experienced trader, you can just couple them with resistance and support lines and you should be ready for manual trading.

- Try identifying one or two parameters that, when taken in sequence, induce an incremental improvement of your equity line. Identifying such hill in the searching landscape gives a better defined model.

Another suggestion from a fellow user of CSSA:

“ The best way to describe what I do is "time relevance," meaning, if you have a 1-minute bar and you want to see the 1-hour wave, then set the m-Histories to 60. If you want the TrainBars on 1 day of data, then set it to 1439 (I think Phil's first bar is labeled zero) in the 1-minute and set the equivalent in 5 mins, and so forth (but don't forget that currencies have 24-hour data, so do something comparable with other financial issues). There are many variations you can do, but usually, if you stick to making things time-relevant, the appropriate configuration will pop-out for decision-making in that moment or the short time period when you need to make a decision (I usually use this set-up for closing trades manually, especially those caught within news reports). If you do auto-optimization to create trading strategies, I suggest you set all the TrainBars to the same date, time, and equivalent length ”


On another note, the strategies in the package might not work for you on finer granularity timeframes; they were crafted for daily charts. But if you get creative, you will find excellent results:

- Trendline has a heavy tendency to trade by countertrend in 1~5-min bars meaning that it crosses below first then crosses above to give a long signal.

- You can also experiment with Flatlines to identify congestive ranges. In your chart example, the market is moving sideways with very little upward and downward movement; these sideways coils are very prone to fakeouts and we all know that escaping a position against a breakout can be expensive.

Check out this other example which uses Bollinger Bands with the Flatlines indicator:
http://www.trade2win.com/boards/atta...latlines-2.gif.

You can use price entry into this range to give up the trade and price exit of either the upper line or lower line from within the range as an on switch to anticipate the breakout.

Related to using Flatlines let me quote a user who posted on NeuroShell’s forum:

“ Remember the old adage about trading breakouts/breakdowns? Never trade the first bar of the breakout or breakdown. Same with cycles, never trade the first cycle that cycles out of the flatline indicator (No matter how huge and tempting the move is. Let's just say I've learned this lesson one too many times!). Wait for the next trough or peak and buy or sell there (or wait for zero if your quadrature phase is shifted). ”

Another piece of trading advice that has been suggested to us by the same author and happens to be consistently true about CSSA: “ look for direction in higher time frames, look for entry/exit in lower time frames. This is an old trading lesson, but you can make money with it using Noxa-CSSA alone! ”

“ It is based on the very old but reliable Alexander Elder's trading system with two or three time frame windows. It's trade-able, if you know the character of the price action of the instrument you are trading. Plot a longer time frame, say a 1-hour bar in a cycle indicator. Plot a shorter time frame, say, 5-min bars with the same cycle indicator. When the 1-hour bar cycle is turned up, trade the 5 min cycles that are turning up. You can try this with Mesa or CSSA (You can also opt to shift the quadrature phase so that the beginning of the up/down cycles is zero). This way, you are actually predicting from the slower movement of the 1-hour bar (the waves) what the momentum of the ripple will be (5-min bars). It is useful to have a trendline of the hourly bars so you can also trade only up 1-hour cycles when the trend is up (the tide, a-la John Ehlers). Although range traders like me work with a lot of cycles, it always pays to take a peek at your trendlines and your retracements. For example, take a look at the very sharp drop (almost 500 pips in 1 hour and 15 mins) in EURJPY at 10AM today (10/21). Although you can't trade that with cycles, you can trade the retracement with one, easily picking the bottoms and tops of the retracement with a smaller time period starting at 10:15 AM, where the conditions are once more amenable to cycle analysis. ”

Hope this helps.

Noxa
Noxa is offline Software vendor   Reply With Quote
Old Feb 11, 2009, 1:26am   #74
Joined Jan 2009
Exact settings

Quote:
Originally Posted by Noxa View Post
Hello Krzysztof,

Don’t take it personally but comparative examples such as yours tell me more about the inability of a user to do auto-optimization than the predictive skills of an indicator. If not done properly, the odds favoring good auto-optimization are vanishingly small.

Check out these intraday examples of what you should expect from CSSA: Noxa Indicators

If there is one piece of advice I could give to non experts, it would be to try to optimize the figure manually first as you suggest. We found it already very useful just to look at the cycles. In fact, if you are an experienced trader, you can just couple them with resistance and support lines and you should be ready for manual trading.

- Try identifying one or two parameters that, when taken in sequence, induce an incremental improvement of your equity line. Identifying such hill in the searching landscape gives a better defined model.

Another suggestion from a fellow user of CSSA:

“ The best way to describe what I do is "time relevance," meaning, if you have a 1-minute bar and you want to see the 1-hour wave, then set the m-Histories to 60. If you want the TrainBars on 1 day of data, then set it to 1439 (I think Phil's first bar is labeled zero) in the 1-minute and set the equivalent in 5 mins, and so forth (but don't forget that currencies have 24-hour data, so do something comparable with other financial issues). There are many variations you can do, but usually, if you stick to making things time-relevant, the appropriate configuration will pop-out for decision-making in that moment or the short time period when you need to make a decision (I usually use this set-up for closing trades manually, especially those caught within news reports). If you do auto-optimization to create trading strategies, I suggest you set all the TrainBars to the same date, time, and equivalent length ”


On another note, the strategies in the package might not work for you on finer granularity timeframes; they were crafted for daily charts. But if you get creative, you will find excellent results:

- Trendline has a heavy tendency to trade by countertrend in 1~5-min bars meaning that it crosses below first then crosses above to give a long signal.

- You can also experiment with Flatlines to identify congestive ranges. In your chart example, the market is moving sideways with very little upward and downward movement; these sideways coils are very prone to fakeouts and we all know that escaping a position against a breakout can be expensive.

Check out this other example which uses Bollinger Bands with the Flatlines indicator:
http://www.trade2win.com/boards/atta...latlines-2.gif.

You can use price entry into this range to give up the trade and price exit of either the upper line or lower line from within the range as an on switch to anticipate the breakout.

Related to using Flatlines let me quote a user who posted on NeuroShell’s forum:

“ Remember the old adage about trading breakouts/breakdowns? Never trade the first bar of the breakout or breakdown. Same with cycles, never trade the first cycle that cycles out of the flatline indicator (No matter how huge and tempting the move is. Let's just say I've learned this lesson one too many times!). Wait for the next trough or peak and buy or sell there (or wait for zero if your quadrature phase is shifted). ”

Another piece of trading advice that has been suggested to us by the same author and happens to be consistently true about CSSA: “ look for direction in higher time frames, look for entry/exit in lower time frames. This is an old trading lesson, but you can make money with it using Noxa-CSSA alone! ”

“ It is based on the very old but reliable Alexander Elder's trading system with two or three time frame windows. It's trade-able, if you know the character of the price action of the instrument you are trading. Plot a longer time frame, say a 1-hour bar in a cycle indicator. Plot a shorter time frame, say, 5-min bars with the same cycle indicator. When the 1-hour bar cycle is turned up, trade the 5 min cycles that are turning up. You can try this with Mesa or CSSA (You can also opt to shift the quadrature phase so that the beginning of the up/down cycles is zero). This way, you are actually predicting from the slower movement of the 1-hour bar (the waves) what the momentum of the ripple will be (5-min bars). It is useful to have a trendline of the hourly bars so you can also trade only up 1-hour cycles when the trend is up (the tide, a-la John Ehlers). Although range traders like me work with a lot of cycles, it always pays to take a peek at your trendlines and your retracements. For example, take a look at the very sharp drop (almost 500 pips in 1 hour and 15 mins) in EURJPY at 10AM today (10/21). Although you can't trade that with cycles, you can trade the retracement with one, easily picking the bottoms and tops of the retracement with a smaller time period starting at 10:15 AM, where the conditions are once more amenable to cycle analysis. ”

Hope this helps.

Noxa
Thanks for reply. Obviously to set the parameters in your indicators is not easy. As you are the main expert can you post exact settings for certain day like e.g. this thursday
for EURUSD 5 min than from the moment when those setting are posted we can start out of sample test till end of the week.

Krzysztof
Krzysiaczek99 is offline   Reply With Quote
Old Feb 11, 2009, 1:45am   #75
Joined Jan 2009
NOXA indicators signal test

I made a following test of NOXA cycle indicator. I applied 4 different types of signal to it.

1) combiend cycle (sin 0.5HZ + cos 0.1HZ)
2) gauss noise
3) gauss noise smoothed with SMA
4) Cmbined signal 1+2 smoothed

Purpouse of the test was to find out if cycle indicator is acting properly for those signal
as forex and stock market data is pretty noisy sometimes..CSV files in NS format and all charts are attached. See ref file for info, you can use them by adding to NS from data source. Here are the results.

1) There are two cycles in this signal 0.5HZ and 0.1HZ. Using strategy Entries 0
the peaks of fast cycle were found properly. However I was not able to find any proper
settings of cycle indicator to extract the slower cycle, I tried I think all combinations
within range of 5 of GroupStart and GroupDepth. I would be very interested to see the setting which extract slow cycle.
Attached Thumbnails
g15.jpg  
Krzysiaczek99 is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search

Similar Threads
Thread Thread Starter Forum Replies Last Post
Neuroshell Trader Burney Technical Analysis 51 Jul 29, 2012 7:05pm
neuroshell adrianallen99 Technical Analysis 2 Mar 3, 2008 4:08pm
Neuroshell rizwanuk Trading Software 6 May 6, 2006 11:40pm
Neuroshell rizwanuk General Trading Chat 0 Sep 23, 2005 10:29am
Neuroshell training and testing results rasty1611 Trading Software 1 Mar 14, 2004 6:24pm

Currently Active Users Viewing This Thread: 2 (0 members and 2 guests)