How LearnAImarkets Works
Published by LearnAImarkets Research Team
How LearnAImarkets Works
The prediction system explained step by step
The decision pipeline
What is LearnAImarkets?
An educational platform that uses AI to analyze European markets every day. It does not invest or trade: it analyzes, predicts a regime (BULL / NEUTRAL / BEAR) for the next 42, 84 and 126 days, and explains why it reaches that conclusion.
What data does it use?
Every day the model ingests prices from 12 European ETFs (Core12) together with more than 60 macro indicators (rates, yield curve, inflation, employment, credit spreads) from the Federal Reserve. Everything is turned into a daily feature vector that feeds the ensemble.
Yahoo Finance
FRED
60+ features
Heterogeneous ensemble
How does the ensemble predict?
Instead of a monolithic model, the final signal comes from combining two very different engines — and each one is itself a small committee. This diversity is what reduces bias and captures uncertainty.
5 Transformers · neural network
Five sub-Transformers trained with different seeds. Each reads long sequences of prices+macro and contributes its own reading. Their divergence is not a flaw: it is how the ensemble captures epistemic uncertainty.
XGBoost · decision trees
Complementary tabular model. Captures non-linear interactions and abrupt relationships that the Transformer may smooth out.
Blend weights (currently ≈70% Transformers / ≈30% XGBoost) are recomputed every 30 days by walk-forward, based on real out-of-sample error measured per horizon. If XGBoost underperforms the Transformer by more than 10%, its weight is capped at 30% as a safeguard.
How does it protect capital?
On top of the ensemble there is a guardrail layer that never moves the signal, but does adjust exposure:
VIX Shield
VIX does not have a single red line. Above 25 we treat it as market stress; 35/45 mark alert and stress. The VIX Shield defensive cash overlay only activates in real panic territory (sustained VIX ≥60), because validation showed that triggering it on scares like VIX 25 destroyed returns.
SMA200 Floor
In a confirmed BULL regime (price above the 200-day moving average), the model requires at least 100% invested. It never sits systematically in cash if the trend remains intact.
Risk multiplier
Final exposure is scaled according to model conviction and detected regime (tanh of the signal, different ranges in BULL/NEUTRAL/BEAR). The more conviction, the more exposure — always within hard limits.
What does each signal mean?
BULL REGIME
The model sees a dominant uptrend. Portfolio fully invested in growth ETFs, minimal cash, protections inactive.
BEAR REGIME
The model detects deterioration. It rotates toward defensive assets (bonds, gold, utilities) and raises cash depending on risk intensity.
NEUTRAL
Transition zone or sideways market. Balanced exposure, no extreme bias, waiting for variables to confirm direction.
Can it be wrong?
Of course. No model is infallible. What you CAN verify is how the system validates itself: the backtest is pure walk-forward (trains on past data, predicts on data it has never seen) and every champion candidate must beat the current one on Sharpe, Alpha AND a panel of out-of-sample thresholds. Past returns do not guarantee future results — but the methodology is transparent, reproducible and auditable.
Does it work? The data speaks.
—
Annual return
—
Edge over S&P 500
—
Return quality (Sharpe)
—
Max historical drawdown
Walk-forward out-of-sample backtest 2018–2026. Past performance does not guarantee future results.
Explore the demos
Quick summary
Learn AI Markets in 60 seconds
If you want a concise version of what the platform does, who it is for, and where to start, this guide is built for that intent.