Evaluating financial recommendations
Testing whether a small neural network can score the quality of financial recommendations.
- Status
- Private research experiment
- Year
- 2025
- Role
- Solo
- Area
- Markets
In plain English
AI finance agents can make recommendations, and someone has to judge which ones are good. This experiment explored training a model to score those recommendations, then checking the scores against past market data.
How it works
- Turn each recommendation into features (numbers that describe it).
- Scale the features so no single one dominates.
- Train an MLP (a small, basic neural network) in Keras to score recommendations.
- Backtest the scores (replay them against historical data) to see how they would have done.
What I did
I ran this solo as a private research experiment.
What I learned
Testing without peeking at the future is what gives a financial score any meaning.