Stock Recommendations: Explainable AI with Boosting Methods

As an exercise to demonstrate the explanatory power of Shapley values, I trained an agent capable of giving stock recommendations from a ticker! The agent, an XGBoost model, takes a ticker as input and reads some key stats from Yahoo! Finance before giving a natural language recommendation as a BUY, HOLD, or SELL. It is also capable of applying different weights to different features for an investment, growth, or ESG portfolio.

I am not a finance person by trade, so the accuracy of these recommendations/the features chosen as representative of future success are certainly subject to scrutiny - the intent was more to demonstrate how computers can generate explanations from inscrutable technical processes. If you have advice on how to improve the model, never hesitate to reach out!

You can scroll through the notebook here.

• Skills: Explainable AI, model design, quant/finance
• Tools: XGBoost, yfinance, shap, Pandas, NumPy