HOOD state
Trailing returns, volatility, range, volume, drawdown, moving-average distance, skew, and return persistence.
MODEL 2 · STEP 3Build
Build
the features.
Turn the frozen source table into 60 trailing-only measurements of price state, market context, peer behavior, and known events.
MARKET CONTEXT · OUTSIDE MODEL FREEZEHOOD
These readings update independently. They are not model inputs, deployment evidence, or a trading recommendation.
- Latest price
- $122.11 · fallback
- 20D momentum
- +30.9% Close-to-close · 20 completed sessions
- RSI 14
- 65.9 neutral · Wilder method
- Relative volume
- 1.2× Versus the prior 20 completed sessions
TECHNICALS THROUGH · TIMESTAMPED FALLBACK
Can a feature be rebuilt using only the past?
Model 2 accepts a feature only if its value at session t can be recreated after that completed close, before the t + 5 outcome begins.
JOINED MARKET ROWS1,255 sessions
FEATURE-READY1,115 rows
ELIGIBLE LABELS1,110 rows
60 TRAILING FEATURES
Own state, external context, one clock.
The matrix contains 70 columns: 10 date, raw, and target fields plus exactly 60 model features.
Market and peers
SPY, QQQ, XLF, COIN, IBKR, and SCHW context plus HOOD excess returns, rolling beta, and correlation.
Regime and events
Lagged Bitcoin state, rolling probability ranks, weekday context, and next-session SEC event flags.
FEATURE CONTRACThood_feature_matrix_v1
60 FROZEN- Calculation direction
- Trailing and backward-looking only
- Scaling
- Fit again inside every training history
- Event availability
- Post-close disclosure begins next session
- BTC availability
- One UTC calendar-day lag before join
- Training purge
- Five sessions between train and test labels
- Warm-up policy
- Common eligibility window across candidates
- Corporate-action handling
- IBKR four-for-one split is adjusted before peer features
Neither a future label nor a same-day post-close event can enter a feature. Tuning, scaling, and calibration are repeated inside the training chronology rather than fitted once on the full sample.
POST-FREEZE TECHNICAL CHALLENGER
RESEARCH ONLYThe point estimate improved; the evidence remains inconclusive.
A fixed benchmark adds 20-session momentum, Wilder RSI 14 with overbought and oversold flags, prior-20 relative volume, and price-volume confirmation. It is scored on the last 252 eligible labels without changing this model's frozen nine-gate decision.
- Brier error
- 0.265713 → 0.260557 Lower is better
- Brier gain
- +0.005155 95% paired interval −0.004389 to +0.015004
- 0.5 hit rate
- 44.4% → 45.2% Threshold accuracy alone cannot approve a model
- Challenger AUC
- 0.441 Below the 0.50 no-skill line
Promotion rule: not passed. The Brier confidence interval includes zero and discrimination remains below 0.50, so the readings stay a live context layer while genuinely new sessions accumulate.
Open formulas and audit filesAlpaca consolidated stock bars · Alpaca US crypto bars · Alpaca corporate-action announcements · Robinhood SEC filings · Robinhood investor relations. The Excel file preserves the source ledger and frozen outputs.
NEXTStep 4 · Train through time
Continue through the frozen Model 2 workflow.