feat(phase8): Session 3 — centrality penalty overlay + N=140 factor backtest#6
Merged
Merged
Conversation
… session 3 research results (127/128 tests passing) Portfolio construction: - build_portfolio() gains centrality_penalty + lambda_centrality=0.30 params weight_i *= (1 - 0.30 * centrality_percentile_i) before renormalisation missing tickers default to percentile=0.5; no ticker zeroed (min mult 0.70) - 3 new tests; 17/17 pass in test_portfolio.py Paper trader: - _centrality_penalty_df() helper: cross-sectional percentile rank of graph_delta_eigenvector z-score at each rebalance date - graph_gnn_embedding_drift popped from _factor_panel dispatch — stale N=30 embeddings distort z-scores for 110 new tickers at N=140 - Rebalance loop wired to pass centrality_penalty into build_portfolio Registry / rolling: - scripts/update_registry_phase8.py: idempotent UPDATE for graph_delta_eigenvector (t_stat=-3.194, status=monitoring, regime_profile records fragility-indicator role and N=140 findings) - rolling_registry rebuilt: 406 rows (58 dates x 7 factors, N=140) Research: - research/phase8_backtest.py: multi-horizon IC backtest (21d/63d/126d) on 140-ticker universe; excludes gnn_embedding_drift; includes graph_customer_momentum; regime splits for |t|>1.5 - Key findings: Phase 5 t-stats were N=30 artifacts (momentum_12_1 126d: 4.149->1.506); graph_delta_eigenvector sign-flipped to t=-3.194 at 63d; graph_customer_momentum definitively null at all horizons Phase 8 paper trader: CAGR +5.97%, Sharpe 0.374, Max DD -36.68% vs Phase 7A baseline CAGR +10.94%, Sharpe 0.521, Max DD -32.81%. Degradation driven by N=140 factor gates correctly classifying more signals as null, producing extended cash periods (100% cash Sep-Dec 2025).
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.
Summary
build_portfolio(): positions arescaled by
(1 − 0.30 × centrality_percentile_i), usinggraph_delta_eigenvectorz-score rank as the fragility signal.Min multiplier 0.70 — no ticker zeroed. Fully backward-compatible.
graph_gnn_embedding_driftremoved from_factor_paneldispatch (stale N=30 embeddings distort 110 new tickers);_centrality_penalty_df()helper wired into rebalance loop.graph_delta_eigenvectorgated out as alpha (negative rolling mean_ic);correct — it is a risk overlay, not a return predictor.
graph_delta_eigenvector:t_stat=−3.194 (63d, N=140), status=monitoring, regime_profile records
fragility-indicator role and sign-flip vs Phase 5.
(21d t=−1.419, 63d=+0.280, 126d=−0.519, N=1,465 in-universe edges).
vs Phase 7A baseline +10.94% / 0.521 / −32.81%. Degradation reflects
N=140 factor gates correctly classifying more signals as null — extended
cash periods are the honest consequence, not a regression.