AI Research AAPLMSFTNVDAGOOGLAMZNMETAQQQ

Mega-cap tech pairwise correlation on QQQ worst-decile down days vs ordinary sessions (last ~3y)

Surprisingly, the six mega-cap names in QQQ were less synchronized on the worst-decile QQQ down days than on ordinary sessions: the mean pairwise Pearson r was about 0.239 on selloffs versus about 0.410 on other days. That gap runs counter to the common intuition that correlations spike toward 1 when the tape breaks.

I looked at close-to-close returns for AAPL, MSFT, NVDA, GOOGL, AMZN and META over roughly three years and compared the average of the 15 pairwise correlations on QQQ’s bottom-decile days to all other days. The full methodology, pairwise breakdowns, statistical test (paired t) and charts are below.

The research question

For QQQ over the past ~3 years, does Big-Tech diversification vanish in a selloff — is the average pairwise daily-return correlation among AAPL, MSFT, NVDA, GOOGL, AMZN, and META meaningfully higher on QQQ's worst-decile down days than on ordinary sessions? Thesis: pairwise correlations lurch toward 1 when the tape breaks, so the mega-caps move as one exactly when you'd need them to diverge, making intra-tech diversification a fair-weather illusion.

How this was measured

Constructed daily close-to-close returns for AAPL, MSFT, NVDA, GOOGL, AMZN, and META from minute bars, aligned over the past ~36 months. Identified QQQ bottom-decile days using the 10th percentile of QQQ daily returns (threshold -0.0144) as 'selloffs'; all remaining days are 'ordinary sessions'. Within each regime, computed the 6×6 Pearson correlation matrix of the mega-cap returns, extracted the 15 off-diagonal pairwise correlations, and summarized their average. A paired t-test across the 15 pairs (selloff vs ordinary) tests whether pairwise correlations systematically rise in selloffs.

The key numbers

Overlapping trading days
740
2023-07-19 to 2026-06-30
Worst-decile day count
74
QQQ return ≤ -0.0144
Other-session day count
666
Mean pairwise corr — selloffs
0.2391
Average of 15 pairs
Mean pairwise corr — ordinary
0.4095
Average of 15 pairs
Difference (selloff − ordinary)
-0.1704
Positive = higher correlation in selloffs
Paired t-stat (15 pairs)
-3.524
Selloff vs ordinary across pairs
Paired p-value (two-sided)
0.0034
p=0.0034 < 0.05 → average pairwise correlation is higher in selloffs
Share of pairs higher in selloffs
20.00%
15 pairs

The charts

Mean pairwise correlation by regime
Distribution of pairwise correlations across 15 mega-cap pairs
Per-pair correlation shift (selloff − ordinary)

Pairwise correlations by regime (15 pairs)

paircorr_selloffcorr_ordinarydelta
AAPL-MSFT0.29480.3194-0.0246
AAPL-NVDA-0.05770.3014-0.3591
AAPL-GOOGL0.2720.329-0.057
AAPL-AMZN0.10820.3617-0.2535
AAPL-META0.43490.28680.1481
MSFT-NVDA-0.0090.4675-0.4765
MSFT-GOOGL0.38160.31790.0637
MSFT-AMZN0.32560.5102-0.1846
MSFT-META0.37420.5212-0.1469
NVDA-GOOGL0.17760.3537-0.1761
NVDA-AMZN0.11220.4417-0.3294
NVDA-META0.05360.4994-0.4458
GOOGL-AMZN0.34540.4498-0.1044
GOOGL-META0.45320.40080.0524
AMZN-META0.32070.5823-0.2616

The takeaway

No — over the past ~3 years the mega‑caps did not converge toward r=1 on QQQ worst‑decile down days; average pairwise correlation was lower in selloffs, not higher. The mean pairwise Pearson r across the 15 stock pairs was about 0.239 on selloffs versus 0.410 on ordinary days, a difference of −0.170. That gap is statistically robust in this test: the paired t‑stat is −3.52 with a two‑sided p ≈ 0.0034 — only about a 3‑in‑1,000 chance this average shift is random under the paired comparison. The pattern isn’t uniform: only 20% of pairs were more correlated in selloffs, with large drops in some pairs (AAPL–NVDA fell roughly 0.30→−0.058, delta −0.359) while a few pairs rose (AAPL–META ~0.287→0.435). Practical takeaway: in this 740‑day sample (74 worst‑decile selloffs, 666 other days) intra‑tech diversification did not evaporate during the worst QQQ down days — correlations tended to fall on average — but that conclusion is specific to this period and these methods.

The fine print