Statistically testing every indicator with my Genetic Algoritm | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

codification of the market / Genetic Algorithms 36 replies

S/R Area Algoritm 17 replies

What Constitutes a Statistically Significant Sample for Back Testing? 4 replies

Has anyone had luck with high frequency using either neural nets or genetic Pr? 3 replies

Genetic Algorithm for Metatrader 1 reply

Attachments: Statistically testing every indicator with my Genetic Algoritm
Exit Attachments

Statistically testing every indicator with my Genetic Algoritm

  • #21
  • Edited 9:24am Jun 4, 2014 8:59am | Edited 9:24am
Quoting Proximus
11) (NOT)Awesome Oscillator Comments: Total opposite to its name, its not awesome at all, its a bullsh*t instead
Ha ha ha ha ha ha

This is a great analysis Proximus.

I understand the value of this type of work.
Like you I also create and use custom indicators/EAs for trading, and I know that we must test them intensively to assure a positive performance.
I also test my indicators on EUR/USD ( 6M to 1 Year in 1H TFs) and you're right, it's a good benchmark.


Keep up the good work... Maybe you can enlighten the ones that every day say all kind of "BS" about using indicators ...
Don't swim with the sharks ...
  • #22
  • Jun 4, 2014 12:57pm Jun 4, 2014 12:57pm
Quoting bhavik1311
Wow. thanks for the result. Must be tedious to work so much but would encourage your work by saying i really appreciate it. Thanks
Thanks

Quoting merquise
Proximus, excellent work And when you say for example for DeMarker or RSI there Minimum parameters/price cluster: Above 100 at least.. What does that mean exactly a setting for Period of at least 100 candles?
I means that anything below 100 is not profitable in average.But note that doesnt necessarily imply that everything above 100 is profitable, it only means that under 100 it is not.

Now you may find periods where under period 100 may be profitable, but across the 12 year , in average, it is not.So the fact that RSI and those have their default value as 14 or something, is totally wrong and not profitable.

Period 14 can be profitable on H4 or D1 timeframe instead, but not on M30.

But still you have to optimize them, but now atleast you know which indicators are worth of it.

Quoting Szpons
Excellent work, thanks! I mostly trade on 4h timeframe and usually use 9-14 parameter which stands for 72-112 in 30m timeframe. Can you also check Commodity Channel Index I wonder how will it perform
Yea.

Quoting lcast
{quote} Ha ha ha ha ha ha This is a great analysis Proximus. I understand the value of this type of work. Like you I also create and use custom indicators/EAs for trading, and I know that we must test them intensively to assure a positive performance. I also test my indicators on EUR/USD ( 6M to 1 Year in 1H TFs) and you're right, it's a good benchmark. Keep up the good work... Maybe you can enlighten the ones that every day say all kind of "BS" about using indicators ...
Yea it would be fun if they would rename it the Bullsh*t Oscillator
"There's a sucker born every minute" - P.T. Barnum
  • #23
  • Jun 4, 2014 1:33pm Jun 4, 2014 1:33pm
awesome thread, keep up the good work!
Luck is what happens when preparation meets opportunity
  • #24
  • Jun 4, 2014 2:12pm Jun 4, 2014 2:12pm
CCI indicator for some reason is bugged, the 2 extremes would be -100 and + 100, but for some reason it jumps to +/- 300 and 400 sometimes so its impossible to test it.

14) ADX (+DI /-DI Cross)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): -38490
Total trades: 4442
Accuracy rate: 24.58%
Minimum parameters/price cluster: None
Profitable: Nope
Comments: Not good.
"There's a sucker born every minute" - P.T. Barnum
  • #25
  • Jun 4, 2014 9:22pm Jun 4, 2014 9:22pm
15) Parabolic SAR (dot below price buy / dot above price sell)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +20652
Total trades: 3604
Accuracy rate: 39.28%
Minimum parameters/price cluster: Step smaller than 0.02 , Maximum is ok at default
Profitable: Yes
Comments: Average

Ok i finished all indicators which can be tested, only the moving average crosses are left, after than i will test custom indicators
"There's a sucker born every minute" - P.T. Barnum
  • #26
  • Jun 5, 2014 1:15pm Jun 5, 2014 1:15pm
Quoting Proximus
15) Parabolic SAR (dot below price buy / dot above price sell) Pair: EURUSD Timeframe: M30 Years tested: 12 Net pipettes acumulated (win-lose): +20652 Total trades: 3604 Accuracy rate: 39.28% Minimum parameters/price cluster: Step smaller than 0.02 , Maximum is ok at default Profitable: Yes Comments: Average Ok i finished all indicators which can be tested, only the moving average crosses are left, after than i will test custom indicators
Since I use moving average indicators, that one will be especially interesting to me. They are the oldest indicator used, as far as I know, and there are so many different variations of them, and how they are used.

I will be interested to know what you find.

I myself trade more the divergences from the moving averages than the moving averages themselves under the theory that "when market breaks a pattern, it has a "reason" to do so".
  • #27
  • Jun 7, 2014 3:46am Jun 7, 2014 3:46am
Quoting Rap Skallion
{quote} Since I use moving average indicators, that one will be especially interesting to me. They are the oldest indicator used, as far as I know, and there are so many different variations of them, and how they are used. I will be interested to know what you find. I myself trade more the divergences from the moving averages than the moving averages themselves under the theory that "when market breaks a pattern, it has a "reason" to do so".
Ok guys i am testing now the moving averages, but because there are many combinations, this may take a while.

But my bet would be that naked MA systems are not good because the MA in itself is a very bad noise filter, but regardless i will still test it.
"There's a sucker born every minute" - P.T. Barnum
  • #28
  • Edited 8:40pm Jun 7, 2014 5:42pm | Edited 8:40pm
16) Simple Moving average crosses (quick MA > slow MA =BUY ,quick MA < slow MA =SELL)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +43362
Total trades: 1521
Accuracy rate: 45.43
Minimum parameters/price cluster: Both above 100, and the slow MA should be 1.2 times bigger than the fast one
Profitable: Yes
Comments: Contrary to the golden cross/deadly cross, and similar BS, when the slow MA is 2 times bigger than the fast MA, is probably not profitable, they must be closer to eachother, like to a 1.2 ratio
"There's a sucker born every minute" - P.T. Barnum
  • #29
  • Jun 7, 2014 8:26pm Jun 7, 2014 8:26pm
Quoting Proximus
16) Moving average crosses (quick MA > slow MA =BUY ,quick MA < slow MA =SELL) Pair: EURUSD Timeframe: M30 Years tested: 12 Net pipettes acumulated (win-lose): +43362 Total trades: 1521 Accuracy rate: 45.43 Minimum parameters/price cluster: Both above 100, and the slow MA should be 1.2 times bigger than the fast one Profitable: Yes Comments: Contrary to the golden cross/deadly cross, and similar BS, when the slow MA is 2 times bigger than the fast MA, is probably not profitable, they must be closer to eachother, like to a 1.2 ratio

Moved my medium until it was 1.2 times the fast. The superior results were immediately visible. I am quit sure this will improve my useage of this indicator. ( my slowest ma is a trigger signifying I should close at loss).
I should closely look at the recrossing for close at loss. Definitely close in profit.
Thank you for your help!
  • #30
  • Jun 7, 2014 8:38pm Jun 7, 2014 8:38pm
17) Single Simple MA (PRICE> MA = BUY , PRICE < MA =SELL)


Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +19215
Total trades: 3165
Accuracy rate: 15.92
Minimum parameters/price cluster: Above 300 atleast
Profitable: Yes
Comments: Looks like 2 MA crosses are better than 1 MA to cross price, and its logical too, 1 MA for dynamic S/R with a filter 1MA is better than 1 dynamic S/R where the price is the trigger, but the price itself is not filtered.
"There's a sucker born every minute" - P.T. Barnum
  • #31
  • Jun 7, 2014 9:00pm Jun 7, 2014 9:00pm
I`ve tested the exponential versions too, they gave horrific results, all of them negative.Now i know that the simple moving average puts equal weight on every bar which is by default flawed because volatility is not uniform , but the exponential weighting and the other linear or whatever is even worse than the simple.The simple atleast treats every bar with equal chances, while the exponential puts weight, but on the wrong ones

So it seems that the simple average with 2 crosses to filter is the best of the moving averages, however not the moving average is the best indicator , other indicators have a better "feel" on the rythm of the price than this.


Quoting Rap Skallion
{quote} Moved my medium until it was 1.2 times the fast. The superior results were immediately visible. I am quit sure this will improve my useage of this indicator. ( my slowest ma is a trigger signifying I should close at loss). I should closely look at the recrossing for close at loss. Definitely close in profit. Thank you for your help!
I`m happy if my observations helped you, but remember that no constant parameter will hold for long term, you have to switch your parameters when the market changes shape, othewise you will be trading an obsolete edge.
"There's a sucker born every minute" - P.T. Barnum
  • #32
  • Jun 7, 2014 10:03pm Jun 7, 2014 10:03pm
I can save you a lot of time and trouble, but I imagine you think this is fun, so you decide.

I built a custom EA several years ago that was capable of measuring crossovers between two moving averages, and two sets of moving averages. For instance, if you wanted to trade 5/8 but only when 50/100 agreed. I also built in a pip distance filter to limit flutter at crossovers. Since nearly every indicator is written against a moving average array, then finding the profitable crossover, or set of crossovers, would also find the profitable settings for nearly any indicator. I tested all crossovers and combinations up to 200 and not a single one was profitable.

At first, I was discouraged. I realized this is why it is nearly impossible to be consistently profitable in trading. I actually stopped trading and just thought about how to overcome this problem. It took me about 3 weeks. A problem with my analysis, and probably yours too, is you are also measuring low volatility periods when anyone smarter than an egg does not trade. If you dismiss those trading hours, your results will differ. Removing those hours from my design, I discovered the method to consistent profit, however it cannot be measured, by your technique. It is too complicated I think. Sorry.

I do however, completely agree that the settings you are using for your test are completely inaccurate. My guess is they are wrong on purpose. I have seen custom indicators changed before they are "offered" for use, so why would a "free" indicator have the correct settings? Also, let's not forget the number of periods were likely set by a programmer and not a trader.

While I think it is very cool that you can test indicators, in my experience this is a waste of time. But hey, don't let me discourage you!! I do agree 100% that the 30M is the right chart for testing. I cannot tell you why in this simple post, but probably you already know. Good Luck
You cannot be extraordinary by being normal
  • #33
  • Jun 7, 2014 10:08pm Jun 7, 2014 10:08pm
Quoting Rap Skallion
{quote} Moved my medium until it was 1.2 times the fast. The superior results were immediately visible. I am quit sure this will improve my useage of this indicator. ( my slowest ma is a trigger signifying I should close at loss). I should closely look at the recrossing for close at loss. Definitely close in profit. Thank you for your help!
What moving average indicator do you have that allows for decimal point settings?

This information is fairly accurate. In the old days I used SMA/LWMA of the same period setting and it was pretty good.
You cannot be extraordinary by being normal
  • #34
  • Jun 7, 2014 10:12pm Jun 7, 2014 10:12pm

The TOP 10 indicators from MT4



1) Parabolic SAR (PSAR)

2) Single MA (simple)

3) Bullpower

4) Bearpower

5) Relative Vigor Index (RVI)

6) MA Crosses (Simple)

7) Ichimoku (Kijun Tenkan cross)

8) Relative Stength Index (RSI)

9) Moving Average Convergence Divergence (MACD)

10) Money Flow Index (MFI)


Now this list may be a bit strange because as you saw that actually the MFI made the biggest profit right, so why is it on the 10th place ? The answer is simple, because although it made 48139 pipettes, it only had 93 trades, which is too little.

So i didnt compared them based on the net profit which it made during the 12 years, rather on the yield or return which it made.So i made a simulation if it would have been a real trading career with 1% growth after each trade, since we need to consider the scaling of the LOT size aswell.

So i simulated a 1% growth/ trade situation, and because the PSAR had the most trades and relative good average net profit, it is logical that it has the highest yield, which far exceeded every one of the other indicators.

By default i didnt included the non-profitable ones, so all 10 of them is profitable, but the difference in profit between them is huge, probably the first 3 made more profit than the other 7.

Here is a spreadsheet where you can see the yield of each one

Attached File(s)
File Type: xlsx INDICATOR COMPARISON.xlsx   10 KB | 1,211 downloads
"There's a sucker born every minute" - P.T. Barnum
  • #35
  • Jun 7, 2014 10:17pm Jun 7, 2014 10:17pm
Interesting, but not surprising about the SAR. I use settings Step 0.1, Maximum 1.0, Precision 7. Can you run those numbers and see what happens please?
You cannot be extraordinary by being normal
  • #36
  • Jun 7, 2014 10:23pm Jun 7, 2014 10:23pm
Quoting dkrock
Interesting, but not surprising about the SAR. I use settings Step 0.1, Maximum 1.0, Precision 7. Can you run those numbers and see what happens please?
Well the algoritm doenst use just 1 single parameter because the market structure changes, so it adapts to different market structures.Its all about adaptation and survival, the market is like a jungle

However i can test it with 1 singular parameter aswell, just gimme a few minutes to reconfigure it.

But what is Precision 7? My PSAR has only 2 parameters

Attached Image (click to enlarge)
Click to Enlarge Name: XXXX.png Size: 62 KB
"There's a sucker born every minute" - P.T. Barnum
  • #37
  • Jun 7, 2014 10:51pm Jun 7, 2014 10:51pm
Quoting dkrock
Interesting, but not surprising about the SAR. I use settings Step 0.1, Maximum 1.0, Precision 7. Can you run those numbers and see what happens please?
Done, and i have bad news, althoug my PSAR has only 2 parameters, so its totally different from yours, but with my PSAR, i set the Step to 0.1 and maximum to 1.0 and no adaptation, the same parameter for 12 years this was the result:

Pair: Eur/usd
Net pipettes acumulated (win-lose): -690840
Total trades: 28916
Timeframe: M30

Pair: USD/JPY
Net pipettes acumulated (win-lose): -613707
Total trades: 27244
Timeframe: M30

Pair: AUD/USD
Net pipettes acumulated (win-lose): -875459
Total trades: 28127
Timeframe: M30

You can see that the adaptation is very important, the same parameter wont be profitable for a long time, even if now it seems like a holy grail.
"There's a sucker born every minute" - P.T. Barnum
  • #38
  • Jun 7, 2014 11:31pm Jun 7, 2014 11:31pm
OK GUYS, MT4 INDICATORS ARE TESTED, ITS TIME FOR CUSTOM INDICATORS!

So please submit custom indicators here, upload them and i`ll test them for you.This thread will never end because there are so many indicators, but i am not going to test all cr*p indicators, only those which atleast are popular or some resemblance to the price movement.

So dont upload indicators based on tarot cards or stuff like that, as interesting as it sounds, it has nothing to do with price movement (yet i saw guys here who were trading based on them

Ok so here are the criteria of the indicators what you will send to me, if they dont meet these criterias i will not test them:

 

  1. Only indicators for MT4, since i use Metatrader 4, i will and can only test these
  2. Must be .mql4 format, .ex4 formats or other encrypted formats cant be used with my algo, because the buffers are hidden, and i`m not going to guess which buffer is for what, so please only .mql4 files.
  3. Must not repaint, this is not so important, because my algo will detect any repainting itself, but generally the repainting indicators are cr*ap by themselves, so only to save a little time, i wont test them (Stuff like ZigZag, Semaphor, Fractals usually repaint).

A repainting indicator doesnt really follow the price , so in this sense its no better than tarot cards used for trading

  1. Must have maximum 2 parameters ( it can have 0 but usually a fixed parameter is not that good, check AwesomeOscillator for example, otherwise 1 or maximum 2 parameters)

Now what i mean is 2 numerical parameters like in the Parabolic Sar, boolean (true/false) or string/text parameters dont count, if it isnt very complex.

But it would be good if it had only 2 parameters.


More than 2 parameters take ages to compute, if 1X1 has 200x200 values, which is 40.000 combinations are usually calculated in less than 5 minutes, then a 3rd parameter would have 3 dimensions 200x200x200, which is 8 million combination which would take 16 hours to calculate, and i wont do that

Neither too complex (with too many parameters) nor too simple ( with too few) are good, the holy grail must be a very well balanced indicator

Ok guys so start sending custom indicators to test them, and lets find out which one is the HOLY GRAIL!
"There's a sucker born every minute" - P.T. Barnum
  • #39
  • Jun 8, 2014 1:01am Jun 8, 2014 1:01am
Quoting Proximus
The TOP 10 indicators from MT4 1) Parabolic SAR (PSAR) 2) Single MA (simple) 3) Bullpower 4) Bearpower 5) Relative Vigor Index (RVI) 6) MA Crosses (Simple) 7) Ichimoku (Kijun Tenkan cross) 8) Relative Stength Index (RSI) 9) Moving Average Convergence Divergence (MACD) 10) Money Flow Index (MFI) Now this list may be a bit strange because as you saw that actually the MFI made the biggest profit right, so why is it on the 10th place ? The answer is simple, because although it made 48139 pipettes, it only had 93 trades, which is too little. So i didnt...
Prox, where is the stochastic oscillator?
If you are good at something, never do it for free--Joker
  • #40
  • Jun 8, 2014 1:06am Jun 8, 2014 1:06am
Quoting jmn5611
{quote} Prox, where is the stochastic oscillator?
It has 3 parameters, its not good.
"There's a sucker born every minute" - P.T. Barnum
0 traders viewing now