Statistically testing every indicator with my Genetic Algoritm

Yup like the title suggests, i`m going to test every major indicator in the MT4 platform, also you can submit custom ones (preferably those which give BUY/SELL signals and can be mechanical, not stuff like pivot points or like that).

Mission

I've build in the last 4 months a complex Artificial Inteligence which can trade upon indicators profitably, also it can check and compare the each indicator's validity.Now i`m not going to put my AI in public, nor any information how it made it,but i will use it to help you folks out who are using indicators for trading.Certain myths about indicators must vanish, some of you say indicator are BS and cant make constant profit with it, some of you say they are the Holy Grail, and can make immense profit.

I say, you can make money by trading them mechanically, but, you cannot use them statically, meaning that you cant use the same parameters like in the MACD 12,26,9 which were usen in 1970, the market changed much since then.

Method

My genetic algorithm, segments price into clusters randomly, and computes the best parameters to the cluster, then after new clusters/ data comes, it learns from those too, and tries to adapt to the best possible parameters, to the most fitting enviroment.

It is pretty inteligent at doing this.

So i will compare those indicators, by their formula not by their parameters, so dont ask me parameters/ settings of the indicators which are the best, because there is no such thing, in different conditions, different parameters are good,but i will tell you atleast the minimum parameters which must be used to be atleast breakeven, but you still must optimize the indicator after every trade or cluster of trades.

Its not like you can use MACD 12,26,9 in 2002 trending market with same succes as in 2010 ranging market.Thats nonsense.

But what i`m going to show is that although parameters can vary, certain indicators will still be overall better than others.

Because their formula, guesses the price flow more accurately, with any parameter, better than other indicators.

So i might make a top 10 list after we are done, but first lets just test them.

Also not that Drawdown and stuff like those are not measured, so it may be that certain indicators are more dangerous than others, but its usually not higher than 25%, if it is then that indicator has a very hart time predicting the price and will probably not be profitable anyway.

Criteria

Simple, 2 numbers are key here, i will measure each indicator, how much pips/pipettes can it acumulate in the last 12 years of time ,by opening a trade at each signal and closing it at opposite. No TP and SL will be used, so Drawdown is not measured!
Second , i will measure the accuracy rate, how accurately predicts price flow.

I will use the EUR/USD only, and M30 timeframe.I have 12 years of very accurate historical data on M30.

Also i chose M30 because i found out that you cant make money in the long term on TF's under 30 with these indicators,or atleast you would be very limited.None of the MT4 indicators can filter out noise under M30 efficiently.

Now it may be that certain indicators will perform better on USD/JPY or AUD/USD, especially the trend spotters.

But i think the EUR/USD is still a good benchmark, and since many people trade it, i will test it only here.Also if i were to test it in every pair that would take forever, and also it takes much computing power to do this, so EUR /USD only

1 other number will be the nr of trades, the sample size.

Why?

Why i do it ? Simple, its free, and i want to help you out guys who are still novices and have doubts about indicators.Also i dont spoil anything, i have way better indicators than the classical ones, but those who only have classical MT4 indicators, this is a great performance analysis of them

TOP 10 INDICATORS OF ALL TIME:
1) Parabolic SAR (PSAR)
2) Single Simple MA
3) Bullpower
4) Bearpower
5) Relative Vigor Index (RVI)
6) Simple MA Crosses
7) Ichimoku (Kijun/Tenkan cross)
8) Relative Stength Index (RSI)
9) Moving Average Convergence Divergence (MACD)
10) Money Flow Index (MFI)

Results on H4:
http://www.forexfactory.com/showthre...67#post7572267

Ok lets start

1) MACD =Moving average convergence divergence (mains strategy histogram above/below 0 to give signals)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +26024
Total trades: 461
Accuracy rate: 32.537%
Minimum parameters/price cluster: Fast and Slow period must be atleast 100 or more to be atleast breakeven depending on market structure, anything under that will not be profitable in long term, and must change parameters to fit ranging and trending markets, Signal period doesnt matter with this strategy
(12,26,9 is a myth )
Profitable: Yes

2) MACD =Moving average convergence divergence (secondary strategy with signal line above/below 0)

Sorry i cant test this, 3 parameters are too much, i dont have enough CP to finish it in reasonable time.It would take days to compute that , too many permutations

The trade sample can be questioned, but thats all we got, and we span trought 12 years so, atleast we tested it in every condition.

Every type of market is available during each period in time. Flip timeframes to get the market you desire. Just keep in mind that as you go down available pips per swing tend to go down.

I cant go lower than m30 anything lower than that is just too noisy.And if i go higher, the sample size becomes way too small, and the trade duration , months/years.Thats not a very fast way of making money , better put your money in bank then

Count me in like your approach ..

Sure, after i finish with the basic MT4 indicators you can submit custom ones too.But preferably in .mq4 format, since i cant attach the .ex4 files to my algo.

And preferably those which have 1 or 2 parameters, anything more than that could take days/weeks/years to compute, and probably be too complicated to even fit the market.

Continuing...

3) RSI= Relative strength index (Above 30 BUY / Below 70 SELL)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +35332
Total trades: 2 (yep only 2 four year long trades, 1 won 1 lost)
Accuracy rate: 50%
Minimum parameters/price cluster: Not really good in this TF (14 setting is a myth ), higher timeframes advised, unless you want to hold a trade for 4 years open...
Profitable: Yes
Comments: Although it gave more profits than the MACD, it only had 2 trades, i tested it on higher timeframe where it gave pretty good results, also in other pairs it was ok, but because we test it on EUR/USD M30, i will discard this from the TOP 10 because of the lack of data on M30 timeframe

4) Bullpower (buy above 0/ sell below 0)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +21337
Total trades: 3056
Accuracy rate: 23.52
Minimum parameters/price cluster: High period is needed, but it can make profit on M30
Profitable: Yes
Comments: High periods needed like on the RSI, but it has way more trades than the RSI, although its accuracy rate is low, i consider it more reliable than the RSI on EURUSD M30

5) Bearpower (buy above 0/ sell below 0)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +29986
Total trades: 2796
Accuracy rate: 22.46
Minimum parameters/price cluster: High period is needed, but it can make profit on M30
Profitable: Yes
Comments: Less accurate then Bullpower, but it made more profit, with bit longer but fewer bad trades

Excellent idea I was always curious of win rates of those in built indicators Also if you can, can you test them on Daily TF and When testing Moving Averages test with periods of 5, 10, 20 , 50 , 100 and 200...exponential would be ok. For Moving Average Crossovers test groups of (5, 10); (10, 20) and (50, 200) I am a silent fan of Exponential Moving Averages but used them a little

I will leave the moving averages for the last, because there will be many combinations there.First i test the "Oscillators" group.

I choose the M30 because it is the smallest best TF.No doubt the Daily and higher timeframes would make every indicator profitable, but the problem is the profit would be slow.

Would you hold a trade open for 4 months or like that ? I dont think so, because its better to have 100 small trades in 4 month then 1 big trade, because you can compound the profit and make more.

If every trade in average would give 10-15 pips and had 1000 trades in a year would make atleast 1000x more profit, than 15000 pips in 1 trade which lasts the whole year.

Why ? Because you increase the lot size in the meantime as your equity grows.While if you had only 1 trade, the swaps would probably eat all profit, and you had to worry the entire year for that trade...Bad idea.

On Daily TF its easy to trade, but we want quick profit right

Heh, never had a problem with quick profits Also would be good if all results are put in first post after finishing

EDIT: Just a thought, could you put also average monthly profits as well as highest monthly and lowest...

Sure i will list them and put links to every post where i detailed them

6) DeMarker (Above 0.3 BUY / Below 0.7 SELL)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +48139
Total trades: 93
Accuracy rate: 49.462 %
Minimum parameters/price cluster: Above 100 atleast
Profitable: Yes
Comments: So far the best indicator, beats the MACD too


7) Bollinger Bands, Force Index, ATR,Momentum, and Envelopes will be skipped. The BB and Envelopes will be skipped because the price is not normally distributed, so looking at the standard deviation of them is pointless, and probably wont be accurate.

The others are not signal indicators, they measure other stuff.

8) Ichimoku (Kijun & Tenkan Cross)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +18404
Total trades: 1093
Accuracy rate: 37.45%
Minimum parameters/price cluster: Averaging around defaults, looks like the ichimoku's defaults are pretty good
Profitable: Yes
Comments: Looks like the default settings can be acceptable, although the profit is not much, but atleast its precise.

9) Ichimoku (Kumo Cloud strategy)

Skipped, because it needs 3 parameters.

Also we will skip the OSMA too, its similar to MACD, however it has also 3 parameters so, thats too much to calculate with.

Addition: Honestly i`m very dissapointed about its performance i thought the ichimoku would perform way better, i mean all the 300 years of japanese rice trader tradition and wisdom is worth almost nothing nowadays is sad. Maybe it works better in rice futures, but sadly the ichimoku is probably an overhyped myth in forex.

Sorry the AI doesnt work like that, it segments the whole historical data into random packets, or clusters, and computes the best parameters, then selects the best and moves on, and it gets better and better with more data, just as a child gets smarter by learning more, its an artificial inteligence using a heuristic algoritm to learn.

I have no way of knowing the intra-week or intra-month profit, i know the overall stats at the end

That is why its better to have as many trades as possible to not have an entire month in minus, that would s*ck very badly.

That is why i cant say that 189 is the best setting of the RSI, because it isnt, in different times different settings are good, but i can measure the minimum setting, so its logical that if i say anything below 100 is bad, is because the algoritm never picked any value below 100 so they must all be losing parameters, but this still doesnt mean that every parameter above 100 is good , you have to optimize, but you atleast have a benchmark

10) RVI =Relative Vigor Index (Above 30 BUY / Below 70 SELL)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +42040
Total trades: 1529
Accuracy rate: 34.401%
Minimum parameters/price cluster: Above 150 to BE
Profitable: Yes
Comments: A bit bad at filtering, it needs high settings but the second best indicator still after DeMarker

11) WPR = Williams Percent Range (Above -80 buy / below -20 sell)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): -33321
Total trades: 3351
Accuracy rate: 57.20%
Minimum parameters/price cluster: None ,atleast not for this TF
Profitable: Nope
Comments: The first crappy indicator so far, it was either designed for higher timeframes or it doesnt work at all.A bit surprising since its accuracy is very high, yet it cant make profit, whereas all other indicators so Far, no matter how bad but made profit, this one didnt.

It is confirmed, on AUDUSD it made -69862, its really not working, sorry.Chaos theory strategy might be obsolete aswell.

11) MFI = Money Flow index, dont confuse it with market facilitation index (Above 20 buy / below 80 sell)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): +47802
Total trades: 46
Accuracy rate: 42.3%
Minimum parameters/price cluster: Close to default
Profitable: Yes
Comments: At first glance it looks like it made to 2nd place after the Demarker, but actually it didnt, the sample of trade is too small and probably unreliable too long trades with huge DD.So 2nd place is still reserved for RVI

12) Accelerator Oscillator (Buy/Sell above/below 0)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): -352618
Total trades: 16134
Accuracy rate: 38.94%
Minimum parameters/price cluster: It has no parameters
Profitable: Nope
Comments: Complete disaster and waste of time, since it has no parameters, you cant adjust it to market price, and the initial formula itself just doenst follow the price, its very unprofitable indicator

13) (NOT)Awesome Oscillator (Buy/Sell above/below 0)

Pair: EURUSD
Timeframe: M30
Years tested: 12
Net pipettes acumulated (win-lose): -102321
Total trades: 6075
Accuracy rate: 33.33(3)%
Minimum parameters/price cluster: It has no parameters
Profitable: Nope
Comments: Total opposite to its name, its not awesome at all, its a bullsh*t instead

Let's recap a little bit.

We finished all groups of oscillators and volume indicators, only the "trending" group left, and of course the custom indicators.

We skipped a few indicators here and there, for example the Gator which had 6 parameters OMG, even with 3 parameters the total permutations would be like 500^3 because my maximum parameter cap is 500.

It is reasonable because if an indicator needs more than 500 then it was probably not designed for M30 timeframe, so why bother calculate further.

So with 3 parameters my 8 core processor PC has to calculate 125000000 combinations of 12 years of M30 data, like 200.000 candlesticks + the individual processor work that an indicator requires for each calculation + the CPU that the AI needs to learn and combine the clusters.

Now with 2 parameters usually a calculation takes 5-15 minutes, however with 3 parameters it takes 8 hours atleast.

With 6 parameters it could take 300 years

So the Gator Oscillator is out of question

The CPU requirement grows quicker than exponential if you add more parameters...

I wish i had a PC like this for these tasks...



And the Fractals indicator just laggs like hell, i have a few custom ones, but just not fit for AI trading, but i never understand why use them if they lagg 2-3 bars.

Wow. thanks for the result. Must be tedious to work so much but would encourage your work by saying i really appreciate it. Thanks

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?

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