EA artisan's EA Course 【026】 What is a reproducible EA? Dangers of using indicators in EAs (Part 1)

Why Use Indicators

In EA development, the basics are that it is developed based on past raw data, and, of course, it is not overly optimized.


Proving that it is not overly optimized is difficult, but as a basis for this, the simplicity of the rules is helpful.

To further consider reproducible EAs, one point is that indicators are not excessively used.

Even in discretionary trading, I have a question for those who trade using indicators.

Why do you use that particular indicator? Can you reasonably explain the basis for it?

“No, I use this indicator, I don’t know why, but it works well with past data. I’m not sure why.”

Putting yourself in the position of the questioner, how would you feel if you received an answer like this?

Is that reproducible? Can you make money in the future with that?

That’s what you’d think, right.

Super Processed Food

What I especially want to ask traders who use multiple indicators is…


Even with just price data based on the four prices, you can trade, so why rely on many indicators?

More than 90% of indicators in existence are merely recalculations based on the four-price data. Candlesticks are fresh produce. Indicators are processed foods.

Why this combination?

The same applies to parameters.


Even a single moving average has an infinite number of parameter combinations, right?

Simple Moving Average (SMA), Weighted Moving Average (WMA), Exponential Moving Average (EMA), and others exist, but once you decide which to use, you…

decide what period to use. 25, 75, 200, etc.

Up to here these are common parameters, but for the applied price, you may choose as well.

“Close” Close price, “Open” Open price, “High” High, “Low” Low, “Median Price (HL/2)” the midpoint of high and low, “Typical Price (HLC/3)” (High+Low+Close)/3, “Weighted Close (HLCC/4)” (High+Low+Close+Close)/4…

This is the most famous and simplest moving average. There are more complex indicators with more combinations, too.

Overlaying three or four such indicators… that’s not sane.

Beginning traders understandably want to stack many indicators.

By making the chart screen complex, they feel like a skilled trader.

They stack so many indicators that the candlesticks disappear, capture the screen, and post it on social media…

“See! I analyze this much! You can’t understand it, can you? (I don’t understand it either)”

Something like that, perhaps.

So, even just deciding one parameter…

Among the various combinations, why did you pick that one? Does that combination have a rational basis grounded in market principles and reproducibility?

You aren’t choosing it arbitrarily like “it seems to work.” Right?

If you can’t answer that, you should be able to.

If you think, “This is tiresome… I don’t want to think… I want to逃げ out” (run away), then FX is not for you, so you should quit lol

There is no non-optimized method

Back to the reproducible EA.


Forward test results should resemble backtest results.

This seems obvious, but aiming for exact matches is like chasing the Holy Grail, so be careful.

In this course, we have warned about the dangers of over-optimization, but does that mean there are EAs that are not optimized at all? No.

Because we develop based on past data, it is impossible to program without any optimization.

AI automated trading is popular, but do you know how it trades?

It’s big data, isn’t it? What is big data? It’s past data.

They collect data on what happened in the past and how the market reacted at that time…

Then, the next time something similar happens, you’ll think, “If I do this, I can make money!”

Yes, at this point you can notice that it is being optimized, right?

If you think, “If I had done this in the past, I would have made money, so I’ll do the same in the future…”

then you are developing methods that are optimized to past data. In other words, there is no such thing as an unoptimized EA, nor a non-existent method.

And as long as you optimize even a little, there will never be a perfect match between past and future trading results.

Therefore, as long as forward test results do not diverge greatly from backtests, it falls into the category of “reproducible EA.”

This has become long, so I’d like to continue next time.

■ Concept and Operating Policy of the EA I Developed

EA Craftsman’s EAs (Three Arrows) are here