Factor Lab Daily Brief 2026-09-22

Data Cutoff and Sample Definition

• Price cutoff date: 2026-09-18 (Friday). prices_cache.parquet latest date is 9/18. Today 9/22 is a US trading day, but the incremental pull did not write new prices to cache (re-run confirmed output identical to 9/19).
• Fundamental data cutoff: combined_panel.pkl contains TTM financial panel for 502 S&P 500 constituents, updated quarterly.
• Evaluation window: EP/BP/FCF Yield/ROE based on most recent 60 cross-sectional regressions; MOM/VOL/SIZE based on most recent 39 cross-sections (21-day forward return requires 273-day lookback, limited by cache start date).
• Holding period: MOM/VOL/SIZE use default 21-day forward returns. The most recent 21 trading days (9/1–9/18) are not yet fully evaluable.
• IC definition: Spearman rank correlation between factor value and next-period return. The ic_mean reported is the average rank correlation across multiple cross-sections in the evaluation window, not a single-day IC, future return, or downside probability.
• t-test: Ordinary t-test, not adjusted for autocorrelation in overlapping returns (21-day forward returns exhibit serial correlation). p-values cannot directly support “highly significant” claims.
• ICIR: IC mean / IC standard deviation, not a strategy Sharpe ratio or annualized return.
• Long-short direction: Code computes Q1 (lowest factor value group) minus Q5 (highest factor value group). Positive value = Q1 outperforms Q5 (high-factor group lags).

New vs. Ongoing Signals

Ongoing status unchanged: Price data identical to 9/19 run (through 9/18). All 7 factor ic_mean values match 9/19 exactly. No new price-driven changes observed.

7-Factor Summary

⭐ marks p < 0.05 (not adjusted for overlapping sample autocorrelation).

Key observations:

• MOM remains negative (-0.147),延续了 since flipping from positive to negative in late August. Q1 (worst performers over 273–252 days ago) avg return 4.63%, Q5 (best performers) only 0.07%. MOM negative does not mean the market must fall; it only indicates that historical winners lagged historical losers within the sample.
• EP remains negative (-0.048), earnings factor ineffective in current sample. Cannot be interpreted as value rotation confirmation—this only reflects the sign of cross-sectional correlation between earnings/price ratio and next-period returns.
• FCF Yield negative (-0.035), free cash flow yield factor also negative. Same reasoning as above.
• BP ineffective (IC=+0.003, p=0.728), book-to-price factor shows no significant relationship with next-period returns.
• ROE negative (-0.025), return on equity negatively correlated with next-period returns.
• VOL positively effective (IC=+0.086), high-volatility stocks outperformed low-volatility stocks on average—contrary to the low-vol anomaly, worth monitoring whether the sample period contains volatility clustering phases.
• SIZE negative (-0.060), small-cap factor favored within the sample. SIZE negative only indicates relative size within this sample, not a general small-cap rally prediction.

Sector Momentum (MOM by Sector)

IT sector momentum collapse: IC=-0.415, ICIR=-1.45, p<0.0001, the weakest signal across all factor-sector combinations. Q1 (weakest IT stocks 273–252 days ago) avg return 13.42%, Q5 (strongest IT stocks) -2.53%. Note: IT sector Q1/Q5 absolute return differential is enormous; this long-short signal does not deduct trading costs.

Coverage: All 8 GICS sectors covered. Health Care (59 stocks) and Consumer Staples (36 stocks) MOM near zero with p>0.5, no valid conclusions可 drawn. Real Estate has only 14 IC observations (not 39), low statistical power.

Anomaly Detection Notes

The following alerts repeat from prior runs and represent ongoing regime continuation, not today’s突变:

• 🔴 MOM / Financials, Industrials, IT, Utilities: IC flip-from-positive-to-negative alerts continue triggering.对照 ic_history, Financials flipped ~mid-August, Industrials/IT ~late August, Utilities ~Aug 19—all occurred weeks ago.
• ⚠️ MOM / Consumer Discretionary, Consumer Staples, Real Estate: p-value flip from significant to insignificant,延续 trend from mid-September.
• ⚠️ MOM / Health Care: IC=-0.014偏离历史均值 -0.112±0.033 (z=+2.94), higher than historical mean, but absolute value is trivial—no practical strategy significance.
• ⚠️ MOM / Real Estate: IC=+0.035偏离历史均值 +0.137±0.032 (z=-3.23),偏低. Only 14 observations, z-score baseline may be distorted.

Strategic Implications

1. Momentum regime remains negative. IT sector (-0.415) contributes the largest negative signal. This reflects the market structure of Aug–Sep: year-ago winners underperformed losers in the recent 21-day window.
2. Value factors (EP, FCF Yield, ROE) all negative. “Cheap” stocks did not outperform “expensive” ones in the current cross-sectional sample. This does not equal a universal conclusion that value strategies are broken—it only describes the cross-sectional relationship for S&P 500 constituents in the evaluation window.
3. Volatility factor is the only consistently positive effective factor (IC=+0.086, ICIR=+0.65). High-volatility stocks outperformed low-volatility stocks in this sample period.
4. BP factor completely ineffective (IC≈0, p=0.73), should not be weighted in the current sample.

Conclusions That Cannot Be Drawn

• ❌ MOM negative ≠ market must fall. It only describes relative performance of past winners vs. losers within the sample.
• ❌ EP/FCF Yield negative ≠ value rotation confirmed. This is merely the sign of cross-sectional correlation between factor and next-period return.
• ❌ SIZE negative ≠ broad small-cap rally ahead. Only indicates relative size relationship in this sample.
• ❌ p < 0.05 ≠ “highly significant, not random”. t-test not adjusted for overlapping return autocorrelation; p-values may be biased.
• ❌ Number of negative sector momentum values ≠ today’s new reversal count. ic_history contains大量 duplicate records (multiple identical IC values per day), which dilutes z-score baseline. Alerts are ongoing regime markers.

Statistical and Data Limitations

• Price data lag: Cache through 9/18, today’s 9/22 trading day not included. Root cause under investigation (incremental pull script may be filtered by min_rows threshold when yfinance returns short windows—see skill for the known issue).
• ic_history.json contains大量 duplicate date entries (multiple identical IC values per same day), affecting z-score baseline stability.
• MOM’s 39 IC observations mean the evaluation window covers ~39 trading days (~7–8 weeks). Cross-sections from ~7–8 weeks ago are only now being evaluated.
• Real Estate has only 14 IC observations (not 39), insufficient statistical power.
• Quantile portfolio returns not net of trading costs; not claimable as implementable strategy returns.