オートトレードAI-GBPUSD【with cutting-edge AI】ultra-stable GBPUSD-specific EA

～Features of the AI on board～

①A tough AI trained under harsh conditions

This EA trades based on a model learned and built with the latest optimization algorithms.

Historical data for 22 years was used for learning to ensure robustness.

In addition, a variable spread is used for learning,By setting the minimum spread to 3 pips, we reproduce a harsh environment on purposeto

to finish as an EA that can withstand any market conditions.

②Learning evaluation criteria

In AI learning, it is necessary to provide evaluation criteria for what makes an EA good.

For example, when the criterion is win rate, learning proceeds to maximize win rate, but focusing only on win rate can lead to

extreme scores like “1 win 0 losses in a single trade.”

Therefore, this EA constructs and learns a composite evaluation index that emphasizes the following elements.

1. Total gained pips

A fundamental evaluation metric.

Ultimately, the most important item for an EA is “how much profit can be increased.”

Therefore, the basic axis of AI learning is the profit gained.

2. Profit factor

Profit factor (total profit ÷ total loss) is one of the most important elements.

A small PF means larger total losses.

A small PF increases risk, such as losses when only a short period is sampled.

Thus higher PF is taught as a sign of a better EA.

3. Maximum drawdown

Maximum drawdown is an important indicator of stability.

Maximum DD is not “the largest DD at the moment of settlement,” but

“the largest DD including while holding positions as well as after settlement.”and is calculated as such.

In other words, it builds a logic that truly avoids DD.

4. Time efficiency

This is not a common metric, but an original metric introduced to emphasize stability.

Time efficiency is the ratio of how much profit was earned relative to the time a position was held.This is calculated to maximize time efficiency, and learning aims to increase this value.

Often seen in averaging-down or high-win-rate EAs, where losses are avoided until the position turns profitable, leaving profits salt-frozen.

This metric is very helpful in avoiding position-keeping (salted positions).

③Prevention of overfitting

Overfitting is the greatest concern in AI development.

Overfitting means learning too closely to historical data, failing to adapt to new data.

Backtests may look good, but real operation yields poor results.

Especially in tasks like FX with many random elements, overfitting is prone to occur.

Therefore, this EA does not learn from all historical data; it uses validation by splitting data into learning and validation sets.

During learning, only the training data is used, and the validation data is used to verify whether overfitting occurred.

This EA further splits validation data and test data, conducting a two-stage test.

In other words, the latter part of the displayed backtest (the green section) is results on unseen data (about 7% of all data are unknown data).

④Connecting MT4 with AI

This EA includes a simplified AI model.

Because of this feature, backtesting takes slightly longer than with ordinary EAs, and the size of the main ex4 file is larger.

Installation is simple and completed in a few minutes, so please rest assured.

～Backtest result introduction～

Backtests are performed using TickDataSuite data with a variable spread.

The data closely resembles real production historical data, and the model quality is very high at 99%.

The following isthe result of tests with an initial margin of 1,000,000 JPY, fixed at 0.1 lots.

① Avoiding small profits followed by big loss

Maximum DD is as low as 2.94% and very small.

Average winning trade is 1,046 yen, average losing trade is 1,716 yen, and although losing trades are larger,

the risk-reward is 0.61, designed to secure stable profits without losing accumulated gains from a single stop-out.

②High PF, high win rate, high trade count

This EA also features high PF, high win rate, and high trade count.

PF 7.84, win rate 92.78%, number of trades 6819, values achievable only through AI learning.

Monthly average trades: about 25.

～Trading logic～

The logic is not left entirely to AI; it blends developed EA knowledge with AI performance.

The basic logic is contrarian scalping, but detailed decisions are made by AI.

No further details are disclosed. Please understand.

～Three selectable modes

This EA can select from three modes based on parameters: “small profit target mode,” “normal mode,” and “big profit target mode.” (default is “normal mode”)

In the “small profit target mode,” settlements occur with smaller profit margins than in the “normal mode.” Win rate is high and stable, but profits are smaller.

In the “big profit target mode,” settlements occur with larger profit margins than in the “normal mode.”

Because waiting for profit reduces the number of trades and can increase maximum DD, you can aim for larger profits, but trades occur less frequently.

Backtest results for the “small profit target mode” and the “big profit target mode” are as follows.

■Small profit target mode■

■Big profit target mode■

～Compounding feature included～

Below are backtest results using the compounding feature.

To make comparison easier with the non-compounding version, the first position enters at 0.1 lots by changing parameters.

The 22-year backtest results show funds growing from 1,000,000 to 459,000,000 JPY.

Another notable point is that the maximum DD is as low as 4.85%.

This is because the amount of losses remained stable over the 22-year period, so turning on the compounding feature did not have a large adverse effect.

～Frequently Asked Questions～

Q. What specs are recommended for the operating environment?

A. A PC or VPS with at least 3GB memory is recommended.

Although it includes AI, it uses a pre-trained model, so a high-spec environment is not required.

Q. Does this EA perform machine learning?

A. This EA includes a model built by machine learning (AI).

Q. Can it be used on any time frame?

A. This EA is designed for GBPUSD on the 5-minute chart only.

Q. Is the installation easy?

A. An easy-to-follow manual explaining installation is included.

Q. Should I stop trading during major news releases?

A. It is not necessary to stop during news releases.

Q. Are there restrictions by the brokerage?

A.Available only with brokerages whose server times are GMT+3 (DST) or GMT+2 (standard).

Also, use is prohibited on certain prop firms, and some brokerages restrictions apply there.

Q. Can I use it with multiple brokerages?

A. Use is allowed on one account per purchase. If you want to operate with additional accounts, please purchase additional licenses.

Q. How many positions can it hold at once?

A. By default, up to 3 positions are held. Hedging is not allowed.

The maximum number of positions can be changed via parameters, but the default of 3 positions is recommended.

Q. Do you use martingale or averaging down?

A. We never use martingale (lot size is fixed or determined by compounding).

By default, up to 3 positions are held, but minor averaging is performed to adjust for small price movements when counter-trend occurs.

Q. Will the backtest results be as shown?

A. This EA is trained to avoid overfitting and only shows backtest results that are representative, but in some periods it may show losses, and future results cannot be guaranteed.

Q. How is the compounding feature calculated?

A. The compounding feature automatically sets the lot size that can be entered based on the risk percentage of usable margin (％).

Q. What is the logic?

A. The basic logic is contrarian, but the detailed logic is determined by the AI learning.

Q. The number of trades is small

A. Depending on market conditions, there may be days with no entries.

Also, there may be errors in MT4. Please periodically check the terminal's Expert tab for errors.