CalibratedReturns

Prediction intervals for stock returns using conformal prediction.

Overview

Outputs 5-day return intervals for any ticker. The intervals have coverage guarantees - if you ask for 90% coverage, backtesting shows ~90% of actual returns fall within the intervals.

• Model: Quantile regression trained on ~20 features (momentum, volatility, RSI, MA distances, etc.)
• Calibration: Split conformal prediction to adjust intervals based on recent residuals
• Validation: Walk-forward backtesting to verify coverage holds out-of-sample

Backtest results (SPY, AAPL, MSFT, GOOGL, NVDA, 2021-2024):

• Target: 95% coverage
• Actual: 92.3% coverage
• 3,755 total predictions

Run locally

Backend:

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Frontend:

cd frontend
npm install
npm run dev

API

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{"tickers": ["SPY", "AAPL"], "coverage": 0.9}'

Stack

• FastAPI, scikit-learn, yfinance
• React, Recharts
• Data from Yahoo Finance