feat(features): residual-momentum family (W2, L4469) — data module owns features#356
Merged
Merged
Conversation
…wns features (L4469) W2 of the predictor-improvement arc. Adds 3 per-ticker feature-store columns, computed in feature_engineer.py (the SINGLE source of truth for features) — NOT in the predictor. Reuses the existing beta-residualized log-return series (`residual_returns`, the same one feeding idio_vol_60d) — NO beta recompute, no duplicated computation, drift invariant preserved. - residual_momentum_ratio : vol-scaled cumulative residual log-return over the 12-1 skip-month window (∑resid / (σ_resid·√231)) — Blitz/Hanauer residual momentum, the highest-EV signal. Information ratio → `_ratio`. - mom_12_1_pct : 12-1 skip-month raw price momentum (classic factor; the store only had 5/20/60/120d, none skipping the recent month). - sector_mom_pct : sector-ETF own 12-1 momentum (GKX industry momentum, absolute — distinct from the existing sector_vs_spy_* relative features). Full feature-store contract: FEATURES + registry CATALOG + SCHEMA.md §3 + units-suffix compliant (schema-contract test green). A full features.compute run backfills all history; daily_append warms up like any rolling feature. Predictor (alpha-engine-predictor #216) consumes these as read-only ArcticDB columns, observe-gated in the L2 until the standalone leak-free read validates. Suite 1751 passing (+ new test_feature_engineer_residual_momentum.py). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…bust contrast The pure-beta test (close == SPY exactly) produced a 0/0 information ratio dominated by float jitter — value varied by pandas/numpy version (passed locally on pandas 2.3.3, failed in CI). Replaced with the down-drift mirror of the existing contrast test: market up + idiosyncratic drift down → residual momentum NEGATIVE while raw price momentum POSITIVE. Strong drifts dominate noise → version-independent. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
W2 of the predictor-improvement arc. Adds 3 per-ticker feature-store columns computed in
features/feature_engineer.py— the single source of truth for features — reusing the existing beta-residualized log-return series (no beta recompute, no duplicated computation).residual_momentum_ratio∑residual_returns[t-252,t-21] / (σ_resid·√231)— reuses the sameresidual_returnsseries feedingidio_vol_60d(Blitz/Hanauer residual momentum)mom_12_1_pctclose.shift(21)/close.shift(252) - 1(12-1 skip-month; the store only had 5/20/60/120d)sector_mom_pctsector_vs_spy_*relative)Why this lives here (not in the predictor)
The "data module owns features" invariant. An earlier predictor-side draft re-derived a rolling market beta — duplicating exactly what
feature_engineer.pyalready computes forbeta_60d/idio_vol_60d, the drift hazard the invariant prevents. Corrected: the features live here; the predictor reads them as columns.Contract
Full feature-store contract satisfied:
FEATURESlist +registry.CATALOG+SCHEMA.md§3, all three in sync; columns are units-suffix compliant (_ratio/_pct) sotest_schema_contract.pypasses with no grandfathering. A fullfeatures.computerun backfills all history;daily_appendwarms up like any rolling feature.Consumer
alpha-engine-predictor #216 consumes these as read-only ArcticDB columns, observe-gated in the L2 until the standalone leak-free read validates the signal (not auto-promoted).
Tests
Full suite 1751 passing; new
test_feature_engineer_residual_momentum.py(pure-beta → ~zero residual momentum, skip-month exclusion, NaN-when-SPY/sector-missing, warmup).Sequencing note
The predictor reading these needs the columns materialized in ArcticDB — a full
features.compute(DataPhase1) run after merge. Relevant for the Tue 6/2 off-cycle run: DataPhase1 → PredictorTraining → backtester chain.🤖 Generated with Claude Code