Factor Lab Daily Brief 2026-09-26

Data Cut-off and Sample Definition

• Price data cut-off: 2026-09-25 (Thursday). prices_cache.parquet last row is 2026-09-25; today (9/26) is Saturday, no new market data.
• Fundamental panel: combined_panel.pkl, 502 S&P 500 constituents, TTM rolling financials.
• Evaluation window: 60 cross-sectional observations (~3 months); MOM/VOL/SIZE have only 39 observations due to 273-day lookback lag.
• Factor definitions: MOM=21-day momentum (lagged 249 days); EP=TTM earnings / market cap; BP=book-to-price; FCF Yield=TTM free cash flow / market cap; ROE=TTM net income / shareholders’ equity; VOL=60-day annualized volatility; SIZE=log(market cap).
• IC: Spearman rank correlation between factor value and next-period return (default 21-day forward). ic_mean is the cross-sectional average, not a single-day IC, not a forward return, not a drawdown probability.
• Significance: Ordinary t-test, not corrected for overlapping returns autocorrelation. ICIR = ic_mean / ic_std, not a strategy Sharpe ratio.

Comparison with Previous and Recent Runs

EP, BP, FCF Yield, and ROE ic_mean values have been identical for 5 consecutive trading days since 9/22, confirming no price update this session — a continuation of the existing state. MOM, VOL, and SIZE show minor fluctuations (the price series includes the 9/25 single-day move), but the magnitudes are small.

Seven-Factor Summary

⭐ = p < 0.05

Key Observations

1. EP, FCF Yield, and ROE are all negative and significant. This means low-valuation / low-earnings / low-ROE stocks outperformed their high-valuation counterparts within this sample window. This is not a confirmation of “value rotation” — it is simply the cross-sectional correlation direction for this evaluation period. One cannot extrapolate this to “buying value stocks now will profit.”
2. MOM remains negative (-0.076). Historical winners underperformed historical losers within the sample. This does not imply the broader market must fall, nor does it constitute a signal to buy lagging stocks.
3. VOL is positive and significant (+0.117, ICIR=1.06). High-volatility stocks outperformed low-volatility stocks in this evaluation period. This is the reverse of the typical “low-volatility anomaly” — reflective of a special market environment.
4. BP and SIZE have no explanatory power in this window. BP’s IC is near zero (p=0.73); SIZE is not significant (p=0.24).

Sector Momentum (MOM by Sector)

Coverage: 8 GICS sectors, 502 constituents total. Some sectors (e.g., Health Care) have small sample sizes, reducing IC reliability.

Anomaly Alerts (from daily script)

• 🔴 IT momentum flipped from positive (+0.31) to negative (-0.29) — but this flip has persisted for ~5 weeks since 8/19; a continuation of the existing regime
• 🔴 Utilities flipped from positive to negative — also a continuation signal
• 🔴 Financials, Industrials, Real Estate — all flipped from positive to negative — continuations
• ⚠️ Health Care IC=+0.047 deviates from historical mean (-0.11) at z=+4.24 — elevated, but small sample warrants caution
• ⚠️ CDS/CDSap momentum p-value no longer significant — declining statistical power

Strategic Implications

1. Momentum factor is inactive in the current evaluation window. IT sector is especially pronounced (IC=-0.29). This is not a sudden event — the flip since 8/19 has persisted for nearly a month. Strategic takeaway: do not chase tech stocks on momentum signals.
2. Valuation factors (EP/FCF) are significantly negative. Low-valuation stocks relativeled outperformed. But ic_mean is a 60-day rolling average; it cannot infer tomorrow’s or next week’s direction.
3. Volatility anomaly reversed. Typically “low vol outperforms high vol,” but in this evaluation period high vol led — reflecting a special market environment (e.g., risk-on rebound).
4. BP and SIZE have no explanatory power in this window.

Conclusions That Cannot Be Drawn

• ❌ One cannot infer next-day market direction from IC
• ❌ One cannot claim “all lagging stocks are buyable” from negative MOM
• ❌ One cannot confirm “value rotation has begun” from negative EP/FCF
• ❌ The count of negative sector values ≠ count of new reversals this period — most are continuation flips from ic_history’s positive baseline
• ❌ p < 0.05 does not mean a strategy is profitable (no transaction costs deducted, no portfolio backtest performed)
• ❌ ICIR is not an annualized Sharpe ratio

Statistical and Data Limitations

• Significance tests are ordinary t-tests, not corrected for Newey-West or overlapping-sample autocorrelation; p-values may be artificially low
• ic_mean is a multi-cross-section average rank correlation, not a single-day IC. The most recent 21 trading days cannot be fully evaluated for MOM due to the 273-day lag window
• ICIR = ic_mean / ic_std, not a strategy Sharpe or annualized return
• Sector momentum Q1-Q5 direction follows actual code (Q1 = lowest factor-value group); long-short returns are not net of transaction costs
• ic_history.json contains many duplicate-date entries (multiple rows per run day), which may distort baseline z-scores