Factor Lab Daily Brief 2026-09-23

Data Overview

Seven-Factor Summary

IC = Spearman rank correlation (factor value vs. 21-day forward return); ICIR = IC mean / IC std; p-value = ordinary t-test (unadjusted for overlapping-sample autocorrelation); Long-Short = Q1 mean return − Q5 mean return (code computes Q1−Q5; positive means low-factor group outperforms high-factor group).

Note: 09-22 and 09-23 results are highly consistent (fundamental factors identical; MOM/VOL/SIZE differ only in second decimal), as both use the same price cache ending 09-22. MOM recovered slightly from −0.1468 to −0.1290 (Δ = +0.018), not a regime switch.

Continuation of Existing State

• Momentum persistently negative: MOM IC = −0.13, holding negative territory across multiple sessions. The IC history baseline includes a positive-momentum era (mean +0.1 to +0.4) in May–June; after flipping negative in late August, the “positive-to-negative” alarm has fired continuously. Current IC = −0.129 is the same order as prior −0.147 — regime continuation.
• Value factors collectively negative: EP, FCF Yield, ROE all negative with p < 0.001, meaning high-profit/high-cash-flow/high-ROE groups underperformed low groups in this cross-section. This is not evidence that “value rotation is confirmed” — low-valued, low-cash-flow stocks simply have stronger short-term rebounds in the current截面.
• SIZE negative: Small caps outperformed large caps in this sample; cannot extrapolate to a sustained small-cap regime.
• BP persistently useless: IC = 0.003, p = 0.728 — book value does not predict future returns.
• VOL stable positive: IC = +0.093, ICIR = +0.72, the strongest signal across all factors. High-volatility stocks have higher average short-term returns — contradicts the “low-vol anomaly” literature, but the截面 IC only states correlation direction.

Sector Momentum (MOM by Sector)

Coverage: All 8 GICS sectors reported; Real Estate has only 13 valid IC observations (low reliability).

Anomaly alert interpretation:

• Financials, Industrials, IT, Utilities “positive-to-negative” alarms are regime markers — ic_history shows these sectors were positive in May–June and flipped negative in August, firing continuously since.
• Health Care IC deviates from historical mean (z = +3.24)偏高, but absolute value is only −0.002 and not significant — practically meaningless.
• Real Estate IC deviates from historical mean (z = −4.38)偏低, but only 13 observations — unreliable.
• CD and Consumer Staples p-values went from significant to not significant — statistical power declining.

Strategic Implications

1. Momentum unusable currently: IT sector IC = −0.38 extreme negative, Q1−Q5 = +14.49% (low-momentum groups far outperform high-momentum). Chasing strength is counterproductive in tech; short-momentum strategies are effective in IT currently.
2. Value factors short-term失效: EP/FCF Yield/ROE all negative means “buy low valuation” underperforms “buy high valuation” over 21-day holding. This is not long-term evidence of value investing失效 —截面 correlation ≠ long-term holding returns.
3. VOL the only stable positive factor: ICIR = 0.72, highest across all factors. The high-vol = high-short-term-return correlation is strongest in this regime.
4. Significant sector divergence: IT/Industrials/Financials/Utilities momentum deeply negative; discretionary and staples虽不显著 but slightly positive — defensive consumer sectors may be relatively resilient.

Conclusions That Cannot Be Drawn

• ❌ MOM negative ≠ market must decline. It only means winners (over the past 1 year) underperformed losers subsequently in this sample.
• ❌ SIZE negative ≠ confirmed broad small-cap rally.
• ❌ EP/FCF Yield/ROE negative ≠ value rotation confirmed — 21-day horizon too short;截面 correlation ≠ long-term alpha.
• ❌ “4 sectors flipped positive-to-negative” ≠ a new-day risk event. This is the continuation of the late-August regime shift, firing continuously in ic_history for many days.
• ❌ p < 0.001 does not mean “highly certain.” Ordinary t-test does not adjust for overlapping-return autocorrelation; p-values labeled “unadjusted,” actual significance may be lower.
• ❌ ICIR ≠ strategy Sharpe ratio or annualized return.

Statistical & Data Limitations

• Fundamental factors (EP/BP/FCF Yield/ROE) use 60 cross-sectional observations; price factors (MOM/VOL/SIZE) only 39 complete windows (require 273-day lookback + 21-day forward).
• p-values from ordinary t-test; 21-day forward returns overlap, autocorrelation unadjusted.
• Quantile portfolio returns unadjusted for transaction costs — not callable achievable strategy returns.
• Sector decomposition uses GICS mapping only, not full market sector-neutralization.

Charts:

• Factor IC distribution: /charts/factor-ic-2026-09-23.png
• Sector momentum: /charts/sector-mom-2026-09-23.png