mega-cap weekly dispersion by climate era — fixed-percent thresholds dilute

Desk Note Ticker Tape
2 min read

mega-cap weekly dispersion by climate era — fixed-percent thresholds dilute

10 tickers referenced · published Jun 12, 2026

Article published Jun 12, 2026. Prices below use latest available snapshots.

AAPL $305.59 -8.4% 30d MSFT $480.35 +22.0% 30d AMZN $261.31 +5.7% 30d GOOGL $344.00 -0.8% 30d META $568.97 -11.9% 30d NVDA $225.01 +10.9% 30d AVGO $392.43 +5.8% 30d TSLA $339.30 -10.9% 30d JPM $360.96 +5.8% 30d XOM $161.46 +9.6% 30d

analysis · self

TL;DR — single-name weekly volatility has run ~1.3× the 2010s baseline in every era since 2020 (big ±5% weeks 1.5×), while SPY's own big-week frequency is back to 2010s levels. The elevation is DISPERSION — names rotating violently against each other under a calm index — and it is why fixed-percent thresholds calibrated on 2010s physics dilute.

This is the founding evidence of the pattern-catalog arc (pattern-vocabulary), re-measured from the engine's own tape. Reproducible sibling: 2026-06-12-mega-cap-weekly-dispersion-era-study.py — code does the math; this file only cites it.

Measurement

Same 14 mega-caps (AAPL MSFT AMZN GOOGL META NVDA AVGO TSLA JPM XOM WMT V LLY UNH), weekly closes from the engine's WEEKLY max.json caches (interval 1wk — the correct interval for a |weekly move| measure; weekly bars stay banned from the daily classifier). Two aggregations:

  • A pooled — all name-weeks in the era pooled; median |weekly move| + share of name-weeks >5% / >10%.
  • B per-week — median across names per week, then median over weeks (the live dispersion-regime detector's shape, lib/patterns.ts).
era A pooled median weeks >5% weeks >10% B per-week SPY median ratio SPY weeks >5%
2010–2019 2.01% 15.8% 2.6% 1.92% 1.01% 1.89 2.5%
2020–2021 covid 2.55% 24.2% 6.7% 2.45% 1.59% 1.54 6.7%
2022–2023 rate shock 2.56% 25.3% 5.9% 2.49% 1.77% 1.40 8.7%
2024–now 2.59% 23.1% 5.0% 2.45% 1.38% 1.77 2.4%

(The brief's pre-refresh table used a slightly different aggregation and read 1.96/2.46/2.39/2.61 with lower big-week shares; the qualitative finding is identical and the .py pins this measurement precisely. Re-run it rather than copying either table.)

The climate finding

  1. Single-name vol stepped up in 2020 and never came back. Pooled median 2.01% → ~2.55-2.59% in all three eras since; >5% name-weeks 15.8% → ~23-25%. Three different macro regimes (covid, rate shock, AI cycle) — same elevated dispersion.
  2. The index normalized; the names did not. SPY's >5% week frequency: 2.5% in the 2010s, 8.7% in the rate shock, 2.4% now — indistinguishable from the 2010s. The 2024-now elevation is pure cross-sectional rotation under a calm index (name/SPY ratio 1.77, near the 2010s 1.89, but on a much bigger absolute base).
  3. Consequence for thresholds: a fixed −10% drawdown line crosses constantly when baseline single-name weekly vol is ~1.3× the old regime with 1.5× the fat weeks — "pullback" trends toward noise (the AMZN 2026-06-11 case: −10.5% off the high with a rising SMA200). Thresholds should either be self-adjusting (percentile-of-own-history, ATR units) or be read against the climate gauge.

Where this stands now

  • The dispersion signal (elevated ≥ 2.2% / low ≤ 1.5% per-week median, 13-week window) currently reads elevated (2.93% vs SPY 1.97%).
  • The fixed −10%/−30% drawdown lines stay canonical: an ATR-unit variant was tested against this era split and rejected — it broke on the intuitive multiplier.
  • Standing follow-up: a quarterly dispersion re-measure.

Sources

  • Detector: dispersion-regime in patterns.
  • Registry + receipts: PATTERNS; task tracker: TASKS-ENGINE #pattern-catalog, TASKS-SCANS.