Noxa indicators for Neuroshell

[Krzysiaczek99]

SSA and CSSA

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

Noxa

I think i miss this as i was busy with 'repaint issue' but just to clarify to the readers what is going on here SSA, CSSA and references to statements of pr. Codebreaker about SSA

NOXA in order to achieve causality and good performance made a hybrid system so SSA + Neural Nets + unclear for me smooth and lead. Specjally LEAD as it introduces phase advance is suspicious - see help file of LEAD indicator used in Neuroshell - it says 'dont use for trading'.

Anyway this is hybrid system which one part is SSA wiht completelly new BASIC FUNCIONALITY which most likely was never properly tested as is should, it was only tested on market data. I hope it clarifies a bit.

So references to SSA are quite out of topic because we are dealing with hybrid here.

Krzysztof

 

[Krzysiaczek99]

non stationary CHIRP, groups and crazy trades

From the post 174, 175 and 176 it was becoming obvious that after market conditions change current settings are becoming not compatibile with a new market conditions and CSSA starts to make 'crazy trades', number of trades grows immediatelly (example post 175 from 49 to 443) and PF falls. Than manual adjustments of settings was necessary. But where is the limit of CSSA tracking capabilities and how to dedect border conditions. This is crucial question when to stop trade using CSSA.

Than I made a test, I generated non stationary CHIRP signal with exponentially modulated amplitude. It simulates well known 'triangle'. Additionally I added noise to it. See results. For clean signal 10 perfect trades, after adding the noise - 50 trades, some of them really 'crazy, immediate buy and sell and making groups (marked on the chart screenshot). Any idea how to get rid of groups and 'crazy trades'

I tried to find better settings of CSSA but no way in this case I think.

Chart ad CHIRP csv attached. Noise was added from NS with standard method with amplitude 1 and 2. In this case S/N ratio is falling all the time because signal amplitude is decreasing, Phenomenon of raising PF with noise amplitude also occurs here.


Krzysztof

 

[alby1714]

From the post 174, 175 and 176 it was becoming obvious that after market conditions change current settings are becoming not compatibile with a new market conditions and CSSA starts to make 'crazy trades', number of trades grows immediatelly (example post 175 from 49 to 443) and PF falls. Than manual adjustments of settings was necessary. But where is the limit of CSSA tracking capabilities and how to dedect border conditions. This is crucial question when to stop trade using CSSA.

Than I made a test, I generated non stationary CHIRP signal with exponentially modulated amplitude. It simulates well known 'triangle'. Additionally I added noise to it. See results. For clean signal 10 perfect trades, after adding the noise - 50 trades, some of them really 'crazy, immediate buy and sell and making groups (marked on the chart screenshot). Any idea how to get rid of groups and 'crazy trades'

I tried to find better settings of CSSA but no way in this case I think.

Chart ad CHIRP csv attached. Noise was added from NS with standard method with amplitude 1 and 2. In this case S/N ratio is falling all the time because signal amplitude is decreasing, Phenomenon of raising PF with noise amplitude also occurs here.


Krzysztof

Hi Krzysztof,
Perhaps this question is a bit out of the thread topic, however, have you tried the same experiments by using NS only ?
if yes what results did you get ?
Thanks
alby1714

 

[Krzysiaczek99]

Hi Krzysztof,
Perhaps this question is a bit out of the thread topic, however, have you tried the same experiments by using NS only ?
if yes what results did you get ?
Thanks
alby1714

Hmmmm. I'm using only NS here. Only instead of time series from financial market
there is a serie generated arteficialy with known content in order to know what we are looking for and to see reaction of CSSA and to know what are it's limits.

If you are talking about how CSSA performs with real market than arryex in post 200
found the settings and i think will post results for out of sample test. so we will know it soon. I will also find some settings based on last week data and post so we will see.

Krzysztof

 

[alby1714]

NS without CSSA

Hi Krzysztof,
I guess my question was not clear. I understand the nature of your tests and that you are not using market time series. What I am saying is that CSSA, if I am not mistaken, is an add on to NS, for instance CSSA is also offered by Noxa as an add on to Metastock. Then one could run your exact test on NS only (without CSSA) just to see whether NS on its own can "learn" your artificial time series and produce better results than when NS is used in association with CSSA. Does it not make sense ?

alby

 

[Krzysiaczek99]

Hi Krzysztof,
I guess my question was not clear. I understand the nature of your tests and that you are not using market time series. What I am saying is that CSSA, if I am not mistaken, is an add on to NS, for instance CSSA is also offered by Noxa as an add on to Metastock. Then one could run your exact test on NS only (without CSSA) just to see whether NS on its own can "learn" your artificial time series and produce better results than when NS is used in association with CSSA. Does it not make sense ?

alby

Hi,

I was doing this, comparing to MACD strategy so not NOXA related. Results were mixed,
the chart which I posted with CHIRP has MACD strategy hidden so you can see how it performs.

Krzysztof

 

[Krzysiaczek99]

EURUSD15m out of sample

Here are the settings which I found for EURUSD 15m based on last week data. Let's check how they will perform today and tomorrow

csv is attached

Krzysztof

 

[Krzysiaczek99]

EURUSD 15m 1st 24h

NOXA -64pips, MACD - 120 pips excluding commisions Cycles from last week and settings of MACD seems not to be valid at all. Market changed characteristic and was very flat today comparing to last week. Should I reoptimize settings ??


Krzysztof

 

[Krzysiaczek99]

EURUSD 5m

I was not able to find any profitable settings for yesterday for 12h paper trading and 12h out of sample for EURUSD5m for E0 entries but i found for MACD. See comparision and lets check at the end of today how those settings were performing.

Krzysztof

 

[Krzysiaczek99]

CSSA settings for EURUSD 5m above

see screenshot, initially it was performing OK but at the end of day performance collapsed as market started to fall a bit. MACD survived change of conditions

 

[Krzysiaczek99]

EURUSD 1m settings

settings based on yesterday 12h paper trading + 12h out of sample for 1m EURUSD, seems to be profitable. MACD didi't want to optimize somehow.

Krzysztof

 

[Krzysiaczek99]

EURUSD 15m 2nd 24h

NOXA -64pips, MACD - 120 pips excluding commisions Cycles from last week and settings of MACD seems not to be valid at all. Market changed characteristic and was very flat today comparing to last week. Should I reoptimize settings ??


Krzysztof

DIsaster. Settings from Friday completelly invalid for both CSSA ans MACD on Tuesday exactly the same situation like on Monday. CSSA - 130 pips MACD - 175 pips

 

[Krzysiaczek99]

EURUSD 5m 1st 24h

I was not able to find any profitable settings for yesterday for 12h paper trading and 12h out of sample for EURUSD5m for E0 entries but i found for MACD. See comparision and lets check at the end of today how those settings were performing.

Krzysztof

Here are the results. CSSA 4 pips, MACD which was performig so well last 12 h on Monday
-136 pips. Setting of MACD was performing reasonable till 7am my time than collapsed. Settings of CSSA were performing 'up' and 'down' but is visible that they prefer 'flat market'
like on Monday.

 

[Krzysiaczek99]

EURUSD 1m 1st 24h

settings based on yesterday 12h paper trading + 12h out of sample for 1m EURUSD, seems to be profitable. MACD didi't want to optimize somehow.

Krzysztof

CSSA Setting which made a big profit for 12h on Monday made a big loss on Tuesday - 150 pips

 

[Krzysiaczek99]

CONCLUSIONs

From all those test is very simple conclusions

There is no advantage/disadvantage of using CSSA against GA optimized MACD.

Both of them are very UNADAPTIVE to market condition change, market tests and signal tests just confirmed this, change of noise was causing immedialely that settings were becoming invalid and 'crazy trades' were occuring. Additionally using of CSSA has a risk of overfit, mysterious properitary tool from NOXA maybe would help.

If somebody has different opinion instead of many words just prove it with exact settings and out of sample test

Krzysztof

 

[at120]

interessting tests....
Noxa any comments?
Thanks a lot and

Bye, AT

 

[Noxa]

From the post 174, 175 and 176 it was becoming obvious that after market conditions change current settings are becoming not compatibile with a new market conditions and CSSA starts to make 'crazy trades', number of trades grows immediatelly (example post 175 from 49 to 443) and PF falls. Than manual adjustments of settings was necessary. But where is the limit of CSSA tracking capabilities and how to dedect border conditions. This is crucial question when to stop trade using CSSA.

Than I made a test, I generated non stationary CHIRP signal with exponentially modulated amplitude. It simulates well known 'triangle'. Additionally I added noise to it. See results. For clean signal 10 perfect trades, after adding the noise - 50 trades, some of them really 'crazy, immediate buy and sell and making groups (marked on the chart screenshot). Any idea how to get rid of groups and 'crazy trades'

I tried to find better settings of CSSA but no way in this case I think.

Chart ad CHIRP csv attached. Noise was added from NS with standard method with amplitude 1 and 2. In this case S/N ratio is falling all the time because signal amplitude is decreasing, Phenomenon of raising PF with noise amplitude also occurs here.


Krzysztof

Use Entries#2 or QPhase crossovers instead of Entries#0. They both filter out false signals that Entries#0 might generate on successive shallow extrema.

The system was optimized on the clean signal. I show the result when noise is added (Gaussian noises at two different levels: 6dB difference). As you can see the entry signals barely change as I add the noise.

Noxa

 

[Noxa]

interessting tests....
Noxa any comments?
Thanks a lot and

Bye, AT

As we all know, it is way easier to show bad results than good ones in trading; unfortunately the laws of entropy prevent most traders from becoming millionaires. But despite being new to NeuroShell Trader and probably new to trading as well, Krzysztof managed to generate equity with CSSA (his posts #180, #211) which is quite remarkable by itself and confirms the potential of these indicators. Given that they are put to good use in a trading strategy (see my tips post #73) CSSA can really shine making it one of the nicest cycle-based indicator set.

Noxa

 

[Krzysiaczek99]

solutions ??

As we all know, it is way easier to show bad results than good ones in trading; unfortunately the laws of entropy prevent most traders from becoming millionaires. But despite being new to NeuroShell Trader and probably new to trading as well, Krzysztof managed to generate equity with CSSA (his posts #180, #211) which is quite remarkable by itself and confirms the potential of these indicators. Given that they are put to good use in a trading strategy (see my tips post #73) CSSA can really shine making it one of the nicest cycle-based indicator set.

Noxa

Hi,

Post 180 - training range was set incorrectly so future leak and repaint occured later
so those results are not valid unfortunatelly

Post 211 - all settings were correct ===>very profitable for 1st 12h

So any solution (except falling or flattering equity curve)to dedect that conditions of market are changing and it's time to reoptimize CSSA or simple to stop trade with it ??

Regarding your optimization tool. Any special reason that you keep it secret ??

Krzysztof

 

[Krzysiaczek99]

Use Entries#2 or QPhase crossovers instead of Entries#0. They both filter out false signals that Entries#0 might generate on successive shallow extrema.

The system was optimized on the clean signal. I show the result when noise is added (Gaussian noises at two different levels: 6dB difference). As you can see the entry signals barely change as I add the noise.

Noxa

Hi,

IF it can make on CHIRP with noise than it is very powerfull. All other systems like MESA/Fourier/Goertzel will have a big problem here because of lack of spectrum.

Krzysztof