GS's AI Confidence Signal for 15M Weekly Report ～2026.9.4
I asked AI to predict "2 hours later" on EURUSD M15
go_stock here.
A verification freak, a personally developed MT4 indicator
for "GS AI Confidence Signal for 15M":
I will show you everything, including verification data and actual performance.
When people hear about tools using AI, many are skeptical, so
I want to explain as clearly as possible.
【1. What does this tool do?】
On the EURUSD 15-minute chart,
a machine learning model predicts whether the price will go up or down in 2 hours,
and it only displays arrows in scenarios with a favorable edge.
Light blue Up = Buy signal
Orange Down = Sell signal
The prediction model (120 decision trees, 3,480 nodes, 54 indicators)
is already integrated into the main indicator,
and no additional files or internet connection are required.
Judgments are made only from confirmed candles, so
it does not repaint by changing arrow positions later.
There are no orders placed at all,
so the decision to enter is ultimately made by a human.
【2. Why is 2 hours to close the position the standard?】
When testing over 12.2 years with varying holding times,
2 hours yielded the best balance.
Holding 30 minutes: win rate 54.9% / per trade +0.47 pips
Holding 1 hour: win rate 56.6% / per trade +0.96 pips
Holding 2 hours (standard): win rate 56.4% / per trade +1.32 pips
Holding 3 hours: win rate 55.5% / per trade +1.62 pips
Holding 6 hours: win rate 53.5% / per trade +1.92 pips
At a glance, extending to 6 hours seems to offer larger rewards.
However, when looking from 2021 onward,
3–6 hours were the same as or worse than 2 hours.
To avoid tailoring only to past data,
the standard setting remains 2 hours.
※ Values are gross values not including spreads or slippage.
【3. Why I deliberately did not set take profit or stop loss】
Setting a take profit increases win rate.
Take profit none / close by time (standard): win rate 56.4% / per trade +1.32 pips
Take profit set at 8 pips: win rate 70.9% / per trade +0.80 pips
Win rate increases by 14 points,
but the per-trade gain drops by about 40%.
When you set take profit, you cut off the run-up too early.
As for stop loss,
the width of favorable movement within two hours after the signal (average 15.4 pips) and
the width of adverse movement (average 14.3 pips) are almost the same.
Rather than forcing a stop loss,
mechanically closing by time yields better results.
【4. Verification results for 12.2 years】
From June 2014 to August 2026, tested on EURUSD M15 for about 12.2 years.
Signals: 20,877
Win rate: 56.4%
Profit Factor: 1.20
This is the figure from the Walk-Forward validation.
At each point, it sums predictions from models trained only up to that point.
It does not use future data to retroactively verify past results.
There is no favorable backtest that looks into the future and matches past data.
【5. Latest performance (weekly automatic aggregation)】
Operational performance after release is
automatically aggregated by Python every week and reported.
Since it is not manual, you cannot cherry-pick a favorable week.
Here are the trends for the last five weeks.
Week 31 (07/27–): 15 trades / 13 wins 2 losses / 86.7% / total +132.0 pips
Week 32 (08/03–): 5 trades / 3 wins 2 losses / 60.0% / total +9.2 pips
Week 33 (08/10–): 1 trade / 1 win 0 losses / 100.0% / total +8.3 pips
Week 34 (08/17–): 2 trades / 2 wins 0 losses / 100.0% / total +5.7 pips
Week 35 (08/24–): 1 trade / 1 win 0 losses / 100.0% / total +2.6 pips
Details for Week 35 show
08/26 00:15 Buy — Confidence 74% — +2.6 pips
Weeks with few trades (1–2) have large number fluctuations,
so please treat the figures as a rough guide.
We plan to continue publishing both good and bad weeks weekly as is.
【6. Summary】
EURUSD M15-specific AI signal with no extra files or internet required.
12.2 years of verification shows a win rate of 56.4% and a profit factor of 1.20.
The standard is "close mechanically in 2 hours".
Not placing take profits or stop losses yields better results.
In the future,
we will continue with practical logs based on this weekly report and
also publish articles detailing concrete methods to increase accuracy through filtering.
Details and purchase information for the product are here.
The following figure is a screenshot of the latest signal.
――――――――――――
【Disclaimer】
This article and this indicator are
information provided to assist trading decisions and are not investment advice.
They do not guarantee profits, and
past verification results and weekly report performances
do not guarantee future results.
The author bears no responsibility for losses incurred by trading using this content.
Please conduct actual trading with your own judgment and responsibility.