AI-like chart prediction EA development notes 1｜First, I made it to the point where it “moves.” However, it is not a winning EA yet.

Hello, this is Tsumo.

This article isthe first in the AI Similar Chart Prediction EA Development Notes.

This time, I will write about a newly started EA.

The name is

AI Similar Chart Prediction EA

.

Straight away, I will honestly say at the outset.

This EA is not a finished product yet.
It is not even at a stage where it can be used in actual operation.
It is not yet something that can be called a “profitable EA.”

But in EA development, one big step forward has been made.

I managed to get it to move.

In EA development, this point is surprisingly important.

If it’s just ideas, there are endless possibilities.

Saying that alone is easy.

But actually getting it to run on MT4.
Read past data.
Compare with the current chart shape.
Look for similar patterns.
Infer direction from subsequent price movement.
Proceed to trading decisions.
Run in Strategy Tester.

When you get here, the talk becomes reality all at once.

And reality is quite harsh.

In this backtest, I confirmed that the EA can operate.

However, the results were negative.

In other words, the current evaluation is like this.

Operation verified. Profitability not verified.

Development will begin from here.

If you’re interested in MTP integration, please also take a look.

• MTP: Master [Free]
  https://www.gogojungle.co.jp/tools/indicators/79103?via=users

• MTP: Slave [Paid]
  https://www.gogojungle.co.jp/tools/indicators/79106?via=users

What is being built this time

What I’m building this time is an EA that looks for scenes in past charts that resemble the current price movement.

The idea is simple.

Look at the shape of the current chart.
Search past charts for similar shapes.
Then see whether prices rose or fell in the past.
Use that as a reference to determine the current direction.

In other words,

“If there is a similar past pattern, I want to see how it moved afterward.”

That is the concept.

Even humans sometimes think while looking at charts like this.

This shape, I feel I’ve seen before.
This move resembles a prior rise before.
This breaking action previously led to a sharp drop afterward.
I’ve seen this range breakout somewhere.

I am attempting to quantify this intuition into an EA.

Of course, a human’s intuition cannot be translated directly into an EA.

EA cannot understand “vaguely similar.”

Therefore, the sequence of candlestick patterns and price movements is numerically compared.

Still too early to call it AI

The name includes AI.

However, to be frank,

as of now, we are not yet using a full-fledged AI model, but

nearly a past-chart similarity search logic

instead of forecasting with a neural network.

We are not predicting the future with neural networks.
We are not outsourcing trading decisions to an external AI.

At this stage, the flow is as follows.

Quantify the current chart shape into numbers.
Compare with past chart shapes.
Search for similar past patterns.
Aggregate subsequent price movements.
Determine a sense of direction.

This is the flow.

However, the direction is AI-like in nature.

Search for similar scenes from past data.
Look at tendencies from multiple samples.
Use those tendencies as a basis for decision making.

This way of thinking is easily AI-izable in the future.

Later, there will be room to increase features, adjust weights, or connect with external analyses.

Therefore, for now I’m moving forward with the name “AI Similar Chart Prediction EA.”

I will nurture it so it lives up to the name.

A flashy-sounding EA tends to quickly lose its appeal in the market.

So, I will verify it modestly here.

First, operation verification succeeded

The biggest achievement so far is that it ran without compilation errors.

In EA development, this is the first major hurdle.

There is a concept.
Code has been written.
But compilation does not pass.
It doesn’t run in MT4.
It does not display in Strategy Tester.
Nothing appears on the chart.

Stalling at this stage is common.

This time, I have reached the point where it runs without compilation errors.

And, it can operate as an EA in Strategy Tester.

Furthermore, in backtests, trades and settlements were recorded over a long period.

This at least confirms that the EA can progress to execution in the Strategy Tester.

This does not mean it is profitable yet.

But it is a significant milestone indicating a solid development foundation.

An EA that does not run cannot be verified.

Only after it runs can it be verified.

Current backtest conditions