Hype Signal

Check the hottest stock signals right now — powered by social/news momentum and ML.

Rolling Signals

Posts after 4pm ET roll to the next trading day to avoid close-price leakage.

Probabilities

Per-post predictions are combined per ticker/day into proba_any.

Evaluation

Outcomes use forward prices: 3-day return or 5-day peak flags.

Goal: Predict if a ticker rises within ~3 trading days after a signal day.
Label (hit): ret_p3d ≥ 2% (D+3 close ÷ D0 close − 1).
Decision (correct/incorrect): compare the hit with proba_any ≥ 0.5.
Aggregation: multiple posts per day roll into one row with n_posts, proba_mean, proba_max, proba_any, and platforms.
Stored in: public.model_ticker_predictions (one row per ticker/day).

Goal: Detect names likely to print a local high in the next ~5 trading days.
Label (hit): is_peak_within5d = 1 (with optional dd_after_peak_p10 if you later require ≥10% DD after peak).
Decision (correct/incorrect): compare the peak flag with proba_any ≥ 0.5.
Stored in: public.model_ticker_predictions (same aggregation fields).

1) Ingest signals

Reddit, Google News, Yahoo News → social_posts (ticker + sentiment).

2) Align day

Convert to ET; posts after 4pm roll to the next trading_day.

3) Features & predict

CatBoost scores each post; per ticker/day aggregate into proba_any.

4) Evaluate

Join forward prices; compute returns/peaks in model_ticker_outcomes.

Picks policy

Threshold (e.g. proba_any ≥ 0.60, min/max picks) or Top-K per day → model_daily_picks.