Institutional Flows, Quality, and Volatility Metrics vs Regression Expected Returns

War & AI Integration Universe  ·  N = 65  ·  Dependent variable: Returns Absolute 1M
Metric Corr FCF Yield R² Distribution
Sigma Score 20 Days 0.760 −0.069 0.578
Short Term GEX-to-Volume 0.397 −0.236 0.158
Pct to Medium Term Call Wall −0.353 0.201 0.125
Medium Term GEX-to-Volume 0.310 −0.170 0.096
IV Term Structure −0.287 0.470 0.082
FCF Yield −0.230 0.053
IV 1M −0.220 0.343 0.048
Beta to SPY −0.169 0.088 0.028
Dark Buy Ratio 1M 0.163 0.046 0.027
Z-Score Absolute 1M −0.157 0.104 0.025
Medium Term GEX 0.146 −0.070 0.021
Dark Buy Ratio 1W 0.143 −0.009 0.021
Dark Buy Accel 2W vs 1M −0.122 0.110 0.015
Dark Buy Ratio 2W 0.112 0.082 0.013
Skew Medium Term 0.095 0.022 0.009
Dark Buy Accel 1W vs 2W 0.089 −0.215 0.008
Medium Term Bullish Pct 0.083 −0.059 0.007
Millennium Alpha Rank 0.014 −0.336 0.000
Dark Buy Accel 1W vs 1M −0.009 −0.093 0.000
Medium Term Bullish Volume −0.004 −0.046 0.000

Notes: Correlation is Pearson r against realized one-month absolute returns. R² = r². The IV Term Structure metric was available for 57 of 65 securities. Green bars indicate the four strongest predictors; red bars indicate the remaining metrics. All metrics are cross-sectional, single-period observations.

Findings

Sigma Score 20 Days dominates. At R² = 0.578, it explains nearly 6× more variance in 1-month returns than the next-best metric. Nothing else in this dataset comes close.

Fundamentals and volatility were poor predictors. FCF Yield and 1-month implied volatility both carry negative correlations but explain under 5% of return variance. High FCF yield stocks (typically value/defensive names) underperformed, consistent with the momentum-driven nature of this period. Higher IV stocks also lagged — the market rewarded calm movers, not volatile ones.

Dark pool flow acceleration was a disappointment. If three derived acceleration metrics (1W vs 1M, 2W vs 1M, 1W vs 2W), none cleared R² = 0.02. The raw 1W and 2W ratios were equally weak. This suggests that for this universe over this window, institutional dark pool activity simply wasn't differentiating winners from losers.

Short Term GEX-to-Volume slots in at #2 with R² = 0.158 and a positive correlation of 0.397 — meaningfully stronger than the medium-term version (0.096) and nearly double the call wall metric (0.125). Both the short-term and medium-term GEX-to-volume ratios show positive correlations, meaning higher gamma exposure relative to volume associates with better returns. A two-factor model combining sigma score with short-term GEX-to-volume might be the natural next test.

IV Term Structure has a 0.470 correlation with FCF Yield — by far the strongest. High-FCF stocks tend to have steeper (more normal) term structures, meaning the market prices less near-term event risk into value names. The top-4 green factors all have low or negative FCF correlations, confirming they're momentum/flow signals orthogonal to fundamentals. That's the right characteristic for this universe — quality metrics simply weren't driving returns over this window.

A multi-factor composite. Realistically, the most likely path past 0.578 isn't a single metric but a simple linear combination — something like 0.7 × sigma + 0.15 × earnings revision + 0.15 × call wall distance. Even modest orthogonal signals, when stacked, can push R² into the 0.65–0.75 range. The fact that sigma alone explains 58% suggests the remaining variance is split across several weak, partially independent signals — exactly the setup where ensemble approaches shine.