Is it worth automating a simple system in MT5?

[Jiesax]

I am testing a very basic H1 media crossing system. It has no extra indicators or complex filters. I wonder if it is really worth automating something so simple in MT5, or if it is better to leave it manual and optimize it with experience.

[Jiesax]

The system enters when a fast EMA crosses a slow, and closes at the opposite junction. Has anyone tried something similar automated? I am interested to know if an EA for something like this is profitable or becomes inefficient by constant whipsaws.

[gesrruda]

Do not waste time automating nonsense. If with two EMAs you think you're going to beat the market, you better stick to something else. 90% of EAs based on that break accounts.

[higaNeskur]

Do not underestimate the simple. Sometimes the most basic works if it is well parameterized and with good monetary management. I have an EA of crossing stockings with trailing stop that gives good results in quiet pairs like USD/CHF.

[SuttSoSTkno]

And why don't you do it manually with alerts? MT5 allows you to create custom signals. So you save yourself programming and keep control at all times.

[Jiesax]

I had considered it, but I prefer to avoid the emotional factor. The idea of automating is precisely to remove the dilemma of �entering or not entering� and to execute mechanically.

[kesusorssttso]

Automate it. Not because it�s going to be profitable, but because it�s the best way to learn how the system reacts in real conditions. You�re going to be surprised at the mistakes your �perfect logic� makes.

[esbodos]

I did it and learned more by programming that EA than by reading 3 technical analysis books. Even if you don�t use it, it will give you insights about the system�s weak points.

[KuCi1]

Can you share the exact rules? Maybe someone already has it implemented and can pass you the base code in MQL5.

[Jiesax]

Of course: EMA(10

[tyvakno53]

It seems out of a 2007 PDF. But look, the simple well tuned can give surprises. Just don�t target it in volatile markets.

[esrStTr]

Use MT5's strategy tester with real broker data, not MetaQuotes's historical ones. So you see how it behaves in more realistic market conditions.

[CuresyC1424]

The best thing would be to use the visual mode of the tester and analyze the inputs/outputs one by one. You will find many that look good, but they are false signals by market noise.

[vugures]

If you want to take it to the next level, I put in a volume filter or higher trend. A simple MA200 can avoid a lot of junk operations.

[Jiesax]

Thank you for all the contributions. I will schedule one basic version and another with filters. As soon as I have backtest results, I share them here for anyone who serves you.

[Jiesax]

I already have the first version of the EA ready. Use EMA crossing(10

[AxSesgiutes85]

It was obvious. Without trend filter you will be filled with garbage entries in consolidations. It is like wanting to run on a highway with traffic lights every 100 meters.

[NesSy100]

That result is not bad if you think of it as a starting point. I would try the same but adding RSI(14

[Cr.]

Did you run it with real spread or with the default spread of the tester? Because in high-frequency systems that detail changes the whole result.

[Trovar]

You need to optimize the parameters.

[Jiesax]

Good observation. I ran the backtest with fixed spread of 1.2. Today I am going to optimize the periods and test with variable spread according to the broker. I am also going to try an ADX-based trend filter.

[SuCau]

The ADX can help a lot. If you set it up to operate only when it's above 25, you eliminate quite a lot of noise.

[sdesvuds1]

Or better yet: use the angle of inclination of the slow EMA. There is an old code at mql5.com that calculates the angle. If the slope does not exceed a certain threshold, you do not open operation.

[jiesnesn510]

What if you use Renko instead of normal candles? It gives you a cleaner reading of the crossing, especially in pairs like EUR/JPY that usually have more noise.

[23]

See that these systems often perform better in 15M graphics with adjusted TP and SL. Perhaps you could test it in 15M or even in 5M with well calibrated filters.

[Jiesax]

I tried what they said: I optimized the EMAs at 12 and 26, I used ADX > 25 as a filter and real spread of the broker. Result in EUR/USD H1: +6% per year with 10% drawdown. It is little, but at least it is no longer negative.

[sosoCiSesTk]

Now yes! That�s already a system that can be polished. Add capital management by variable lotage and you�ll see how it improves.

[esnuutes1506]

And if you also implement a rule to avoid important economic news, you could avoid peaks of volatility that ruin the sequence.

[nesncyaxr]

Be careful to overoptimize. If you touch too many parameters to the millimeter, your EA will work great in backtest and horrible in real. It has to be robust, not perfect.

[Jiesax]

Totally. I�m starting to see the difference between optimization and curve setting. I�m going to leave broad parameters and simple rules. The idea is that it works in multiple pairs, not just in one. Next test: USD/CHF and GBP/USD. Do you want me to follow with more answers or new threads?

[Jiesax]

One of the most undervalued keys in the construction of automated systems is robustness in different market conditions. Many developers, especially those who are starting, fall into the trap of overoptimizing their EA for a specific data set, achieving spectacular results in backtest, but disastrous in forward testing. This happens because the system has been calibrated to perfectly adapt to past data, which does not guarantee that it behaves the same with future data. The important thing is not that the EA wins always, but that it can adapt and survive in different environments. One crucial point is the use of realistic filters, as variable spreads, slippage and execution conditions similar to those of the broker with which it will operate in live. If your EA only works in 2019 and collapses in 2020 or 2021, it is sign that something is wrong. Another crucial point is the use of the system's filters, as it is necessary to challenge the system's varying, slippage and execution conditions similar to those of the broker with which it will operate in live. Many systems can collapse when you face the real price of the system or to news of high impact. In addition, it is based.

[axrcoSa11]

Excellent contribution, this should be fixed on all EAs threads. Thank you for sharing!

[esSbesTo89]

And how would you apply that to a system based only on RSI? Does it work the same?

[KhrusTunes]

Thank you for the clarity with which you explained it. Many should read this before burning their account.

[esnTknuavucag]

This is pure gold. I've been struggling with overoptimization for months and I hadn't seen it that way.

[JiesnEsnTknuua93]

Is it really worth testing in multiple pairs if your strategy is designed for EUR/USD?

[Tes]

Thank you for taking the time. You just avoided weeks of frustration with my first EA.

[cruSaTajo]

Thank you for this comprehensive explanation. It opened my eyes to several mistakes I was making with my backtests.

[jidutdaConoch]

Don't you think you're complicating too much something that can be solved with a good adaptive mobile media?

[resgaSess]

Very good post. Especially the part about dynamic risk management, is something that most do not even consider.

[Jesvuor95]

Thank you, I'm going to print and reread it before making any changes to my system.

[SesCwSXDXD]

This served me more than several paid courses.

[YttaySiTor]

Don't you think that if a system needs so many validations it's because it's unreliable from the start?

[JIEsNOSO59]

What if I used tick data for tick, I wouldn't be solving overoptimization more accurately?

[CugwSTracresstunesTkr]

For someone who starts, this may sound too technical. Wouldn't it be better to simplify the approach at the beginning?

[Rahurru]

Great post. The part about validating the logic of the system made me rethink several of my assumptions.

[CarThois21]

Thank you for sharing your experience with such honesty. You can see that you speak from practice.

[xesvu.ugSosesys]

Very helpful about applying volatility management. I had not considered it in my backtests.

[Coruvuxxesr]

A jewel. Saved in favorites without hesitation.

[EssonsuaA]

Is it not contradictory to say that we must avoid overoptimization and at the same time try multiple combinations?

[UsaxoSaesSbes77]

And what do you think of using Monte Carlo to test robustness? Do you see it useful or only for advanced systems?

[JttoaxnwS199]

How much to trust forward demo accounts test if there is no actual execution? Wouldn't that be misleading?

[EsnesusNJ]

Thanks for the realistic approach. I was already thinking that if I didn�t earn 20% a month it was because something was wrong.

[boestdt]

Impressive how clear you made it, especially the concept of the system�s �no return point� when it becomes fragile.

[S.VoSessca]

Thanks, I just needed perspective to stop chasing the �best EA in the world� and focus on something solid.

[UnaxesT]

You helped me understand that surviving in the market is already an achievement.

[Sesyeses]

But then does it make sense to continue with backtests if in the end everything depends on the real forward?

[grdoSh]

What do you think about price action-only systems without indicators? Can they be validated the same?

[sesstro]

Where would you put the limit of an acceptable drawdown on an automatic system?

[Sicuesyvuxxesvordo]

And how do you not fall into the trap of moving the live TP/SL when the EA goes into loss?

[CesrSttguSguS]

I am left with the phrase �it�s not all about formulas, sometimes you have to apply human criteria�.

[Nurwzonk]

Thank you, this gave me more clarity than many automatic system books I've read.

[Nukadox14]

Great truth about equity versus balance. Many celebrate �floating� profits without realizing the real risk.

[TkrContes2015]

Congratulations on sharing something with such courage.

[HwboS]

Don't you think the problem is not overoptimization, but the lack of market context in systems?

[Tucesssa.Conuness]

So you totally rule out using news-based strategies, just techniques?

[gesrcesy1980j]

Do historical data not lose value in a market as much intervened as the current one?

[Axysor123]

To what extent would you trust a purchased EA if you don't have access to the code to see its logic?

[ResNesSaTwzZz]

I really needed to read something like that after so many frustrations.

[CesnoSes23]

It's impossible not to rethink the way my backtests were operating.

[sesries]

Thanks, I'm starting to cross-validate with my data right now.

[Sittax]

A master class, and free. What a luxury to have content like that on the forum.

[desnuCT]

Is it not better to develop semi-automatic systems where the trader is involved in management?

[CaSunes164]

And how relevant is testing in different brokers? Doesn�t it generate more confusion than clarity?

[TodraqiuSwz]

What is more important to you, the benefit/risk ratio or the percentage of success?