Impulse ADX with averaging

Intro:

This thread is dedicated to optimize, backtest & forwardtest the EA with averaging based on RSI and ADX.

EA is developed on MT5 so it is possible to optimize the best setups for multiple symbols combinations.

Rules of thread:

Be a nice person.

How to test EA:

1. Use open prices modeling and any timeframe

2. Optimize 3-7 years - find the best/stable setup

3. Then validate 3 years. For example, optimize using 2010 - 2017 data. EA know nothing about 2017 - 2020 - so you can then run Strategy Test for "future" and see how EA and money management plan work in "pseudo-real life"

Learn how to test EAs in this free video/textual course (no registration needed - just watch or read). It is for MT4 but concepts are relevant to MT5 as well.

Money plan example Nr1:

Grid, martingale, averaging - all those systems will fail if not enough money on account. Sooner or later. But it is ok if using an appropriate money plan.

Here is 1 option (we need to figure out more... better):

1. Have in your broker wallet (not account) 500 usd
2. Put in account 100 usd and start EA
3. When the account balance reaches 300 usd - cashout 200 usd to the wallet and continue running EA (risk/reward ratio 1:2)
4. When account stop-out - recharge with 100 usd and continue EA

Money plan example Nr2:

You know from EA backtests max drawdown. Multiply it by 2 and that is capital needed to run 1 instance of EA with parameters you have optimized/backtested.

Money plan example Nr3:

You can optimize and find n symbols setups. Then sum all symbols max drawdowns, divide by n and get average drawdown. Now multiple it by 2.3 and get the capital needed to run n instances of various symbols/setups EA on 1 account.

Running EA with 0.01 lots eventually will go in 1000 - 3000 usd drawdown. You will see in the backtests. If don't have enough initial capital you can use a cent account - then your drawdown will be 100 - 300 usd.


EA strategy:

We try to catch the new trend impulse (beginning of trend). ADX indicator signals market mood (trendy or ranged). RSI indicator signals direction (BUY or SELL).

BUY entry:
1) Previous candle ADX above X (find best X by optimizing parameter adx_level)
2) Previous candle RSI above Y (find best Y by optimizing parameter rsi_up)

SELL entry:
1) Previous candle ADX above X (find best X by optimizing parameter adx_level)
2) Previous candle RSI below Z (find best Z by optimizing parameter rsi_down)


Here EA version Nr 1 for download:



Here EA version Nr 3 for download (opens 2 oposit orders at once):



Will update this post Nr.1 with detailed info.

Here first candidate for GBPUSD.

Optimized best result backtest for 2010 - 2017 period:





And here .set file (rename .txt to .set):



Then forward tested from 2017 till today - started Strategy Tester from 01.01.2017. When stop out - move 1 day after and started tester again. When profit >300, move 1 day after and again started tester. In that way, it is possible to simulate "real life" trading with an unknown future.

Here sequences of loses/profits:

-100
-100
+300
-100
-100
-100
+314
-100
-100
+329
-100
-100
+300
-100
+353
+415
-100
+379
+300
+445
-100
-100
---------------
+3135
-1300
---------------
+1835

So from 2017 till 2020 earned 1835 usd. Max drawdown 300 usd.

Believe could be found some better setup with this EA.

hello, which timeframe ?

EA has parameter is_open_prices_only - if true then EA will open/close trades according to chart timeframe bars opening.

You can use M1, M5 etc timeframe. See what works best. My example is on M5 timeframe.

Somtimes for EA is better not to react on every noise...

Open prices modeling works very fast for optimizations/backtesting. And yes - therefore EA is specially designed to open/close trades when candles open/closes. There might be some turbulences... 1m or 5m is long time for volatile markets... but.. in both directions (sometimes you win sometimes loose because of that). Everything depends on statistics.

But yes - it is possible to work with ticks... then you need to get ticks data from dukascopy or similar.. regular brokers have ticks only starting from 2017..

trade ist not allowed?

by default, EA works on demo and strategy tester. It's not ready to go on live account... it is needed backtesting/forwardtesting to get the most stable parameters and figure out the best money management plan.

You can post here some good setup (.set file + backtest report + validation stats) and will get the key to run EA on a live account without any restrictions. but I believe we are far away now from the moment when it would be safe to run this EA with real money...

yes, for me this first with ea and runs though, but gives no Signals

Here another quite stable setup. First optimize 2010 - 2017. Then run on 2017 - 2020 and see that EA works stable. Max dd 855 usd. Maybe this could be executed on cent account. Find 10 more like this for different symbols (then run on 10 seperate accounts) to diversify risks. This might be another approach..



Here .set file:



Here backtest Excel 2010 - 2017:



And here validation (pseudo-real life) 2017 - 2020:



Please post something similar but for other symbols/indexes/gold/wti. The more setups will have the more posibilities to choose best candidates.

find some tutorials how to backtest EA on MT5 for example this:

here what should look EA setup in strategy tester:

Updated the post Nr.1 with money plans and EA strategy explanation.

Here nice optimization results for EURUSD 2010 - 2017:



Here instructions how to choose best optimization setup:

1. Add column "Recovery factor" in optimizations results grid

2. Order by "Recovery factor" descending

3. See setups where deal count > 1000 (statistical decisions need large amount of deals)

4. MT5 dont show max drawdown in money - but you can easy calculate it: Profit divide by Recovery factor

We are interested in setups where drawdown less than 1000 usd. And recovery factor as possible bigger. From example picture above we see Profit 8772 and Recovery factor 13 (all rounded). So max drawdown for that setup is 674 usd.

In 7 years we risk with 674 usd drawdown (actually our account must be about 1000 usd to survive that) and we have 8772 usd profit. So the risk/reward ratio is about 1:8.

Ok, here I see that this particular EURUSD setup is quite stable because 2010 - 2017 was great and 2017 - 2020 (our validation with pseudo-real life) is ok too:



Here 2010 - 2017 with dd 662 usd and profit 8772 usd:



And here validation 2017 - 2020 with dd 549 usd and profit 1631 usd:



And here .set file (rename .txt to .set):

Here idea for "basket of n symbols"... took 2 good EURUSD setups which run on different entry conditions. And simulate in Excel what would happen if those 2 EAs would be running on 1 account form 2010 - 2020 (2017 - 2020 is pseudo-real life test):



Here how profits split by years:



And here my Excel with analysis for 2 EURUSD EAs setups combination:



As you can see the total drawdown is no more than 673 usd but total profit is almost twice more: 17582 usd.

Idea is to find stable n setups of various symbols and combine them to run on 1 account.

Hello gerards!
Good luck on your thread!
So once we optimize it it will be available free to use for everyone?

Thanks

Hello! Tnx. From post Nr.1:

Several times noticed that EA in Strategy Tester don't run because when 1 symbol provided in the EA parameter "symbols" then EA controls that symbol in chart or strategy tester must be the same. Otherwise, you will have ticks from EURUSD but make decisions on GBPUSD... therefore implemented validation and raise error to avoid such mistake when going on live for example.

Well, didn't started jet multi symbols optimizations.. But yes this can be confusing sometimes. So always check that at least 1 symbol is the same as in Strategy Tester.

One option to eliminate confusions is to rename parameter "symbols" to "additional_symbols" ? And EA always uses chart/tester symbol as main (for ticks)... just thinking now..

When i see the GBPUSD report tester, the adx time frame is 16392. When load EA to run in demo account, the time frame is 8 hours. I try to forward test your set. Why is that happen ???

MT5 enumerates timeframes and you can, therefore, optimize EA for various timeframes. MT5 assigns for each timeframe numerical value. In strategy tester you can optimize by parameters adx_timeframe and rsi_timeframe - you can set start from M1 and end to Month in optimization parameters.

Keep in mind when optimize by timeframes and in tester you set, for example, timeframe M5 and modeling "Open prices only" then for some timeframes Strategy Tester will skip/error because the step is 5 minutes but in MT5 there are 6 minutes and 12 minutes timeframes which will be ignored.

When the backtest report is exported to Excel (or .set file) those timeframe parameters are exported as numbers and not textual values.

Thank you very much

Here some bad examples so you see how important is to divide testing in 2 periods - optimized and then validation with unknown future. It is like simulating forward testing.

Here it is looser AUDUSD - I didn't find any stable setup but has so many "promising" in period 2010 - 2017...

Here 2010 - 2017:



And here bad truth from 2017 - 2020:



But yes this is bad for money management plans Nr 2 and nr 3 (from post Nr. 1) but it might work out with plan Nr. 1... but seems then different parameters needed... so that there might be 100% profit in 1-2 weeks..

Nop... AUDUSD is bad for MM plan Nr1 as well - here iterated manually 2017 - 2020:

2017.01.01 - 2017.02.15 -100
2017.02.16 - 2018.12.20 +293
2018.12.21 - 2019.12.04 0 (because we have profit 393 and cash out 293 so left 100 on account - no need to recharge)
2019.12.04 - 2020.03.12 -100
2020.03.13 - 2020.03.13 -100
2020.03.14 - 2020.03.18 +207
2020.03.19 - 2020.04.02 0
-----------------------------------
-300 + 493 = +193 usd

Edited:

Adjusted because the MM plan Nr 1 is when balance >300 then cashout 200 and 100 stay to continue. So 2 times was not -100. Therefore we could make some small profit. Drawdown -200 and profit of 193 usd in 2.5 years. not exciting... but at least something.