UQO Working Papers · WP06 · v1.1
Intraday Volatility Transmission Across Asset Classes: A Connectedness Analysis of Equities, Foreign Exchange, Commodities, and Cryptocurrencies
- Year
- 2026
- First release
- October 2026
- Latest revision
- October 2026
- Publication status
- Working paper
Volatility flows from U.S. equity futures to everything else, including crypto
Cross-asset shocks account for 54.8% of forecast-error variance; E-mini S&P 500 and Nasdaq 100 futures are the only robust net transmitters (+36.0 and +25.6 points), and bitcoin's exposure to traditional assets rose from under 10% before 2020 to 20–46% since.
Abstract
This paper measures how realized volatility is transmitted across equity index futures, major exchange rates, gold and crude oil futures, the VIX, and the two largest cryptocurrencies. Daily realized volatilities are built from 5-minute bars for ten instruments over 2008–2026, and the connectedness of their logarithms is estimated on the 2,514 trading days common to all series (March 2016–March 2026) with the generalized variance decomposition of a VAR(4) at a ten-day horizon, with residual-bootstrap confidence intervals.
Cross-asset shocks account for 54.8% (90% interval 53.1–56.9) of forecast-error variance. E-mini S&P 500 and Nasdaq 100 futures are the only robust net transmitters (+36.0 and +25.6 percentage points), while the realized volatility of the VIX (−20.6), crude oil (−14.2) and sterling (−13.2) are net receivers; the net positions of bitcoin and ether are statistically indistinguishable from zero. Bitcoin and ether form a tightly linked pair (27–28% of each other's variance) whose exposure to traditional assets rises from under 10% of forecast-error variance before 2020 to 20–46% in 2020–2025, with information flowing predominantly from U.S. equities to crypto: in the 2024–2026 spot-ETF period bitcoin is a significant net receiver (−12.4, interval −20.4 to −2.1). Rolling estimates show that total connectedness rose during the COVID-19 crash and reached its highest level in 2025.
Intraday, after converting exchange time stamps to UTC, equity and VIX volatility concentrate in the U.S. cash session (53–63% of daily variance), whereas bitcoin's hourly profile is nearly flat (peak-to-trough ratio 1.2); U.S.-session equity volatility predicts bitcoin's next Asian-session volatility, but neither bitcoin's overnight nor its weekend volatility predicts U.S. equity volatility. The paper documents how calendar alignment of 24/7 and exchange-traded series materially affects connectedness estimates.
Key findings
2,514
common trading days, ten instruments, 2016–2026
54.8%
total spillover index, 90% interval 53.1–56.9
+36.0 / +25.6
net transmission of S&P 500 and Nasdaq 100 futures
−20.6
net position of VIX realized volatility, a receiver
<10% → 20–46%
bitcoin variance from traditional assets, before/after 2020
53–63%
equity and VIX daily variance in the U.S. session
Scientific contributions
- A volatility connectedness hierarchy with intervals
- On log realized volatility from 5-minute bars, cross-asset shocks explain 54.8% of forecast-error variance; only ES (+36.0) and NQ (+25.6) are robust net transmitters, and VIX, crude oil and sterling are robust net receivers, across lag orders, horizons and sub-samples.
- Crypto is integrated, and the flow runs from equities
- Bitcoin's own-shock share is 67%, the share of its variance explained by traditional assets rises from 6.8% in 2019 to 34% in 2020 and 46% in 2025, and in the spot-ETF period bitcoin (−12.4) and ether (−19.5) are significant net receivers.
- Sessions measured on a correct clock
- After converting Eastern time stamps to UTC, equity and VIX volatility peak at the U.S. cash open and the U.S. session carries 53% (ES), 54% (NQ) and 63% (VIX) of daily variance; bitcoin's hourly profile is nearly flat (peak-to-trough 1.2).
- Equity volatility predicts crypto, not the reverse
- The previous day's U.S.-session volatility of equity futures predicts bitcoin's Asian-session volatility, whereas bitcoin's overnight and weekend volatility do not predict U.S.-session equity volatility (weekend coefficient −0.04, s.e. 0.05).
- Data construction moves the estimates
- Replacing a forward-filled calendar-day panel with common trading days changed the total index by five points and the net spillover of equity futures by fifteen, and removed the apparent 2018 pre-Volmageddon spike (81.6% → 56.4%) and the near-complete isolation of crypto reported in the earlier version.
Methodology
Methodology (4)
Realized volatility and log transformation
Daily realized variance from 5-minute returns per instrument (at least 20 bars; returns beyond 20 sample standard deviations dropped), annualised and log-transformed for the VAR, with ADF and KPSS tests reported.
Diebold–Yilmaz generalized variance decomposition
VAR(4) chosen by Hannan–Quinn, generalized forecast-error variance decomposition at a ten-day horizon, total, directional, net and net pairwise spillovers, with a residual bootstrap (B = 500) for 90% intervals.
Robustness grid, rolling windows and regimes
Fifteen specifications (levels vs logs, lag orders 1–6, horizons 5 and 20, weekend-inclusive crypto, exclusion of VIX or crypto, 2018–2026 sub-sample), rolling windows of 200, 250 and 500 trading days, four regimes and calendar years with bootstrap intervals.
Granger causality and session regressions
Daily Granger tests on log realized volatility with classical F and HAC Wald statistics (16 pairs), and session-ordered predictive regressions between equity futures, the VIX and bitcoin including a weekend-bitcoin to Monday-equity test.
Research data
- FirstRate Data 5-minute bars, ten instruments
- 5-minute OHLCV bars 2008–2026 in U.S. Eastern time for continuous futures (ES, NQ, GC, CL), spot FX (EURUSD, USDJPY, GBPUSD), spot bitcoin and ether (composite of exchanges) and the VIX index, stored in four DuckDB databases (about 235 million bars, not redistributed); daily realized volatility is the sum of squared 5-minute log returns, with the VAR panel restricted to the 2,514 CBOE trading days on which all ten series have an own value.
- 1-minute bars for the intraday stage
- Same-vendor 1-minute bars (shared with Working Paper No. 12) converted from Eastern time to UTC and aggregated to 5 minutes, March 2016–March 2026, for hourly profiles, session shares and session-ordered predictive regressions.
Reproducibility
- Three-stage Python pipeline: 01 builds daily realized volatility from the raw DuckDB bars (about 10 minutes); 02 reproduces every connectedness table and figure from the committed results/daily_rv_all.csv in about 1 minute; 03 runs the UTC-correct intraday analysis from 1-minute bars
- Construction audit table reconciles the v1.0 numbers (forward-filled panel) with the revised ones, and the full 15-specification robustness grid is computed by script
- Eight figures and all LaTeX tables generated from results/rev_*.csv; full LaTeX source of the 48-page paper
- Pre-revision PDF, review report and changelog preserved in docs/; the raw 5-minute bars are not redistributed
Research artifacts
Version history
| Version | Date | Description | File |
|---|---|---|---|
| v1.1current | October 2026 | Current manuscript as published on www.spboucher.ai (revised edition after the 2026-10-17 scientific review for papers 1–17). | PDF · 48 p. |