Add earnings-profit → 3–5 day price trend model

[ax3301]

What

A model relating quarterly earnings / profit data to a stock's post-earnings price drift over the next 3 and 5 trading days.

New folder earnings_trend_model/:

File	Purpose
earnings_trend.py	Reusable pipeline: yfinance data → profit features → forward-trend labels → direction classifier + magnitude regressor
earnings_trend_model.ipynb	Self-contained, Colab-friendly notebook (data → EDA → model → scoring)
README.md	Usage, Robinhood data-source swap, and limitations (bilingual EN/中文)

How it works

1. Per earnings event, engineer profit features: EPS surprise (%), revenue YoY, net-income YoY, net margin & its YoY change, pre-earnings 5/20-day momentum.
2. Label each event with the forward return over 3 / 5 trading days after the announcement (entry = first trading day on/after the report), plus up/down direction.
3. Model on the pooled, time-ordered event table: RandomForestClassifier for direction + Ridge for magnitude, with a chronological train/test split so the score reflects real forward use (no look-ahead leakage).
4. Score the latest reported earnings per ticker → up-probability and expected N-day return.

Verified

Ran the full pipeline end-to-end on 8 large caps → 388 real earnings events. Data collection, both models, feature importances, and upcoming-earnings scoring all work.

Honest limitations (also in the README)

• Free yfinance ships only ~5 quarters of income statements, so net_income_yoy / revenue_yoy are NaN for older events and show ~0 importance. The signal that actually drives the model is EPS surprise + pre-earnings momentum (earnings-date data covers ~25 quarters).
• Direction accuracy sits near the majority-class baseline (~50%) — post-earnings drift is genuinely weak signal; the code reports numbers as-is. Widen TICKERS for statistical power, or swap in a longer-history fundamentals source (e.g. the Robinhood connector) to activate the YoY features.
• Research only, not investment advice.

Notes

• Data source is yfinance (no API key). An optional set_session() hook routes calls through a plain requests session for TLS-terminating proxy environments where curl_cffi gets reset.
• To use the Robinhood connector instead, swap the three data-fetch functions in earnings_trend.py (documented in the README).

🤖 Generated with Claude Code

---

Generated by Claude Code