Factor Lab Daily Brief 2026-09-29

Data Cutoff

2026-09-28 (Monday) — prices_cache.parquet latest date is 2026-09-28, matching the run date. Market data is current.

Sample and Evaluation Methodology

• Universe: S&P 500 constituents, 502 valid stocks (474 for ROE)
• Fundamental factors (EP/BP/FCF Yield/ROE): Based on daily TTM earnings panel, 60 valid IC observations
• Price factors (MOM/VOL/SIZE): 39 valid IC observations
• MOM: Price[t-21]/Price[t-273] - 1 (21-day return, 252 trading days prior)
• Evaluation window: Last 60 trading days (fundamental) / 39 trading days (price factors)
• ic_mean: Average rank correlation (Spearman) across historical cross-sections, NOT daily IC, NOT future returns, NOT downside probability
• p-value: Ordinary t-test, not corrected for overlapping samples
• ICIR: IC mean / IC std, NOT strategy Sharpe or annualized return
• Horizon: Default 21-day forward return; the most recent 21 trading days cannot yet be fully evaluated

7-Factor Summary Table

Status quo persists: EP, BP, FCF Yield, ROE unchanged across four consecutive runs (TTM panel not refreshed). MOM gradually improving, VOL stable. No material change from fresh market data.

Sector Momentum Table (MOM by Sector)

Anomaly Alerts (with regime context)

• 🔴 Information Technology: IC flipped from +0.312 to -0.262 — continuation. IT momentum turned negative since late August, persistently negative since
• 🔴 Utilities: IC flipped from +0.143 to -0.111 — continuation. Turned negative late August, deepening
• 🔴 Industrials: IC flipped from +0.158 to -0.086 — continuation. Turned negative late August
• 🔴 Financials: IC flipped from +0.186 to -0.101 — continuation
• ⚠️ Health Care: IC=+0.064 deviates from historical mean -0.109±0.038 (z=+4.50) — new flip. HC momentum was persistently negative historically; this turn positive warrants tracking
• ⚠️ Consumer Discretionary/Staples: p-value flipped from significant to insignificant — continuation

Important caveat: ic_history.json contains大量 duplicate date entries (e.g., “mom / Health Care” repeats identical -0.128 from 2026-05-29 to 2026-08-19), potentially contaminating z-score baselines. Alerts should be treated as indicative.

Strategy Implications

1. Momentum regime remains negative but narrowing. MOM ic_mean improved gradually from -0.129 (09-23) to -0.063 (09-29), but p=0.12 remains insignificant. Cannot call “momentum recovery”; this is temporary easing within a negative momentum regime.
2. Low-volatility factor remains the strongest signal. VOL ic_mean=+0.117, ICIR=1.05 — highest and most stable ICIR among all 7 factors. Low-vol stocks continue to outperform high-vol.
3. Value factors (EP/FCF Yield) persistently negative. EP and FCF Yield unchanged for four consecutive periods, ROE also stable negative. In the current market, low-value/high-FCF stocks underperform.
4. IT momentum collapse is the most severe. IC=-0.262 is the weakest across all sectors, highly significant. Combined with the prior flip from +0.31, the “强者恒强” dynamic in tech has reversed.
5. Health Care momentum flip positive needs monitoring. From long-term negative to +0.064, p=0.20 insignificant. Requires subsequent observation to confirm regime change.

Conclusions That Cannot Be Drawn

• ❌ MOM negative ≠ market must fall. It only means winners from t-273 to t-252 underperformed losers over the following 21 days in the sample.
• ❌ SIZE negative ≠ small-cap rally confirmed. It only reflects weak negative correlation between market cap and future returns in this sample.
• ❌ EP/FCF Yield negative ≠ value rotation confirmed. Low-value stocks underperforming in this cross-section does not imply mean reversion is imminent.
• ❌ Significant p-value ≠ profitable strategy. Ordinary t-test does not account for overlapping return autocorrelation; ICIR is not annualized Sharpe.
• ❌ Number of negative sectors ≠ today’s new reversals. IT/Utilities/Industrials/Financials negative momentum has persisted since late August.

Statistical and Data Limitations

1. ic_history.json data quality: Contains大量 duplicate-date entries (identical IC values repeated across many dates), potentially diluting z-score baseline mean/std with invalid duplicates. Z-score alerts should be interpreted cautiously.
2. Fundamental factors frozen: EP/BP/FCF Yield/ROE unchanged across four consecutive runs (to 4 decimal places), indicating the TTM panel was not refreshed. The quoted ic_mean values are snapshots from the last panel update, not “today’s” data.
3. Overlapping sample issue: 21-day forward returns have substantial overlap; ordinary t-test p-values are biased low. “Significant” results reported here may not hold after Newey-West correction.
4. Quantile portfolio direction: Q1-Q5 long-short equals “low-factor group minus high-factor group” return (Q1_mean_ret - Q5_mean_ret in code). Positive = low-factor group outperforms.