Simon-Pierre Boucher
All research

UQO Working Paper No. 6· v2.0· November 2026· 74 pages

Realized-Volatility Connectedness Across Equities, Foreign Exchange, Commodities, and Cryptocurrencies: Daily Variance Decompositions and Trading-Session Evidence

Finance & Economics
On SSRN
Quality controlled67.8/100

Cross-asset shocks account for 54.9% of the forecast-error variance of realized volatility (90% interval 52.4–57.2); E-mini S&P 500 and Nasdaq 100 futures are the only robust net transmitters (+33.7 and +24.0 points), and bitcoin and ether are neutral over the decade.

Daily realized volatilities are built from 5-minute returns on a strict grid for ten instruments and decomposed with a VAR(4) at a ten-day horizon, with bias-corrected block-bootstrap intervals and multiplicity adjustment. The hierarchy is descriptive: a static decomposition without any VAR dynamics reproduces the sign and much of the size of the net positions, so it reflects contemporaneous correlation more than lead–lag transmission. Bitcoin's exposure to traditional series is higher since 2020, but the additional rise at the spot-ETF launch is established only at the 90% level.

Simon-Pierre Boucher — contact@spboucher.ai

Paper PDF (revised)Read onlineSSRNCode & data repository

54.9% [52.4, 57.2]

share of the ten-day forecast-error variance of log realized volatility due to cross-asset shocks, with a 90% bias-corrected block-bootstrap interval; ten instruments, 2,512 common trading days

+33.7 / +24.0

net spillovers (percentage points) of E-mini S&P 500 and Nasdaq 100 futures, the only net transmitters whose intervals exclude zero after multiplicity adjustment; they fall to +25.5 and +17.5 when the VIX is excluded

−15.4 / −14.9 / −13.0 / −10.6

net positions of the realized volatility of the VIX, crude oil, sterling and gold, the net receivers; the positions of bitcoin, ether, the euro and the yen cannot be distinguished from zero

1.8% → 22.7%

share of bitcoin's forecast-error variance due to the eight non-crypto series, pooled 2016–2019 against 2020–2026: a difference of +20.8 points (90% interval +13.7 to +28.9)

+16.0 pts, p = 0.052

rise of the same share around the spot-ETF launch of 11 January 2024 (90% Hall interval +2.5 to +30.4; the percentile interval includes zero; not adjusted for multiplicity), against +2.1 and −4.7 at placebo splits; bitcoin's net position does not change at any of these dates

61.5% / 65.3% / 75.2%

share of daily variance generated in the U.S. cash session for ES, NQ and the VIX, in exchange time; bitcoin's hourly profile is nearly flat (peak-to-trough ratio 1.2 against 3.9 for ES)

MDE 0.159

minimum detectable elasticity at 80% power for the effect of weekend bitcoin volatility on Monday ES volatility, for which no effect is detected

Abstract

Which markets originate and which absorb realized-volatility shocks? Daily realized volatilities are built from 5-minute returns for ten instruments (equity index futures, three exchange rates, gold, crude oil, bitcoin, ether, the VIX), and their log connectedness is estimated with the generalized variance decomposition of a VAR on 2,512 common trading days (March 2016–March 2026), with bias-corrected block-bootstrap intervals. Cross-asset shocks account for 54.9% of forecast-error variance (90% interval 52.4–57.2). Equity futures are the only robust net transmitters; the VIX, crude oil, sterling and gold are net receivers; bitcoin and ether are neutral. Bitcoin’s exposure to the eight non-crypto series rises from 1.8% to 22.7% after 2019, and, less firmly (unadjusted p = 0.052), by +16.0 points at the spot-ETF launch but not at placebo dates. In exchange time the U.S. cash session carries 61.5–75.2% of equity and VIX variance; bitcoin’s weekend volatility does not detectably predict Monday equity volatility.

JEL: C32, C58, G12, G15

Key Results

  1. A ten-instrument system over a decade, with intervals on every measure

    Equity index futures, three exchange rates, gold, crude oil, bitcoin, ether and the VIX from March 2016 to March 2026, spanning the COVID-19 crash, the 2022 tightening cycle and the spot-ETF era; every measure carries a bias-corrected interval and a multiplicity-adjusted p-value.

  2. Net positions are mostly contemporaneous correlation

    A static decomposition that sets every VAR coefficient to zero reproduces the sign and much of the size of the main net positions (ES +23.3 against +33.8); 22.5% of the day-to-day variability of the VIX's realized volatility is accounted for by shocks to the equity futures from which the index is derived.

  3. Gross exposure separated from net position around the spot-ETF launch

    Bitcoin's exposure to traditional series rises at the ETF split and not at placebo dates, but its net position does not change; in the spot-ETF period its net spillover of −11.0 (interval −22.8 to −0.7) does not survive adjustment for the ten net positions of the regime (Holm 0.55).

  4. The alignment of 24/7 and exchange-traded series is documented and quantified

    The calendar of the panel, the exclusion of returns that span a market closure, the time zone of the cryptocurrency stamps and the truncation of extreme returns each move the total index or a net position by several percentage points; the vendor's bitcoin and ether stamps are shown to be UTC by an explicit audit.

  5. Session regressions in exchange time, with minimum detectable effects

    Controlling for bitcoin's own intervening sessions, the previous day's U.S.-session volatility of NQ still predicts bitcoin's Asian-session volatility (Holm-adjusted p = 0.030; incremental R² at most 0.0010), the ES block does not after adjustment (p = 0.070), and bitcoin's earlier-session volatility predicts the U.S.-session volatility of the VIX but not of ES or NQ.

  6. Time variation of aggregate connectedness

    Rolling total connectedness rose in March 2020 and reached its highest values in 2025, when the calendar-year estimate was 73.2% against 49–54% in 2016–2019.

Quality control

This paper went through a multi-stage quality control: 6 quality-control reports and 79 points checked. The version presented on this page is the revised manuscript. Scores are on a 0–100 scale and are a diagnostic of the manuscript, not an acceptance probability.

Score after revision

67.8/100

56.3 before quality control · +11.5

Quality control dated November 4, 2026

Quality-control decision
major revision
Scientific integrity
PASS
Reproducibility
R1reproduced from documented analysis data
Maturity
Bone revision cycle from submission
Quality-control reports
6
Points checked
79

Scores by dimension

Before quality controlAfter revision

  • Originality & contribution52.5 → 56.0
  • Scientific rigor55.3 → 70.0
  • Identification & methodology50.8 → 62.0
  • Data & reproducibility60.8 → 78.0
  • Robustness & validation60.0 → 72.0
  • Writing71.2 → 74.0
  • Figures & tables61.5 → 78.0
  • Literature positioning56.3 → 74.0

Revised manuscript74 pages8 figures25 tables15 equations88 references31,753 words

What quality control changed

Quality control led to a rebuild of the data: the bitcoin and ether time stamps, read as Eastern time in the earlier version, are UTC, and realized volatility is now computed on a strict 5-minute grid without truncation. Inference moved to bias-corrected block-bootstrap intervals with multiplicity adjustment, which weakened several cryptocurrency claims; the title was changed.

  • Bitcoin and ether stamps read as Eastern time → established as UTC and converted; every bitcoin daily and session result differs from the earlier version for this reason
  • Panel of 2,514 common trading days with truncated returns → 2,512 days on a strict grid without truncation; total index 54.8% → 54.9% (90% interval 52.4–57.2)
  • Net position of the VIX −20.6 → −15.4 and of ES +36.0 → +33.7; gold joins the net receivers
  • “Bitcoin is a significant net receiver in the spot-ETF period” → net spillover −11.0 (interval −22.8 to −0.7) that does not survive multiplicity adjustment (Holm 0.55); the firmer result is the +16.0-point rise in gross exposure (unadjusted p = 0.052)
  • “Bitcoin's exposure to traditional assets under 10% before 2020” → between 4% and 20% in the calendar years 2016–2019 (1.8% in the pooled window) and 22.7% in 2020–2026
  • “U.S.-session equity volatility predicts bitcoin's Asian-session volatility” → retained for NQ only (Holm-adjusted p = 0.030), not for ES (p = 0.070), with an incremental R² of at most 0.0010

Limitations stated in the paper

  • The decomposition is descriptive: it does not identify structural shocks, and the net positions are largely reproduced by contemporaneous correlation alone.
  • The VAR(4) leaves serial correlation and conditional heteroskedasticity in the residuals; longer-memory specifications raise the total index by about two points and compress the VIX's net position.
  • The 5-minute realized variance includes jump variation and, for the composite cryptocurrency feed, cross-exchange price discrepancies; a single-exchange comparison is untested.
  • The raw bars are proprietary and cannot be redistributed, and the bootstrap does not propagate measurement error in the realized measures, so the intervals are, if anything, too narrow.
Read online (interactive reader)Revised PDFRevised branch on spbgit

Data

One-minute bars for ten instruments

FirstRate Data 1-minute bars for continuous front-month ES, NQ, gold and crude oil futures, spot EUR/USD, USD/JPY and GBP/USD, composite spot bitcoin and ether, and the CBOE VIX index, from 2013 (2016 for ether) to March 2026; aggregated to 5-minute closes they yield 9.40 million returns on the strict grid.

Common-trading-day panel

2,512 days from 14 March 2016 to 11 March 2026 on which all ten instruments have a realized volatility computed from their own bars (in practice the CBOE calendar); no value is forward-filled and weekend crypto volatility does not enter the baseline.

Time-stamp audit

Futures, exchange-rate and VIX stamps are U.S. Eastern time; bitcoin and ether stamps are UTC, established by cross-correlating the vendor series with public one-minute candles from two exchanges around two known-timed episodes, and converted to Eastern time before any day or session is formed.

Methodology

Realized volatility on a strict 5-minute grid

Daily realized variance is the sum of squared 5-minute log returns on an Eastern-time calendar day, with no return spanning a market closure and no truncation of extreme returns; the analysis uses its logarithm.

Generalized variance decomposition of a VAR

A VAR(4) selected by the Hannan–Quinn criterion and a ten-day horizon give the total index and the directional, net and pairwise measures; VAR(22) and HAR-lag specifications are reported because the VAR(4) leaves serial correlation in the residuals.

Bias-corrected block-bootstrap inference

A bootstrap-after-bootstrap with 500 draws at each stage and moving-block resampling of residuals yields 90% Hall intervals, with Holm and Romano–Wolf adjustments for the number of statements drawn from each table.

Rolling windows, regimes and event tests

Rolling estimates, four regime sub-samples, calendar-year estimates and formal change tests around the spot-ETF launch, with placebo splits in January 2022 and January 2023.

Intraday periodicity and session-ordered regressions

Hourly profiles of absolute returns, variance shares of the Asian, European, U.S. cash and post-close sessions in exchange time, and predictive regressions between the session volatilities of equity futures, the VIX and bitcoin, including a weekend-to-Monday test.

Reproducibility

  • A single entry point rebuilds the analysis data from the raw bars, re-estimates every object, redraws the figures and recompiles the manuscript; a second mode runs from the committed daily panel, in about ten minutes on a laptop.
  • The daily realized-volatility panel and its variants are committed with their hashes, so every table and figure of the results is reproducible without the raw bars; the bootstrap seed is fixed and the draws are committed.
  • A test regenerates the committed tables and compares them, and the frozen scripts of the two earlier versions reproduce their archived outputs byte for byte.
  • The raw 1-minute bars are held under a private research licence and are not redistributed; hashes, row counts and first and last stamps are recorded so an independent copy can be checked, and the first stage is reproducible only with access to the vendor files.

Cite this paper

Boucher, S.-P. (2026). Realized-Volatility Connectedness Across Equities, Foreign Exchange, Commodities, and Cryptocurrencies: Daily Variance Decompositions and Trading-Session Evidence. UQO Working Paper No. 6, Université du Québec en Outaouais.

@techreport{boucher2026wp6,
  author      = {Simon-Pierre Boucher},
  title       = {Realized-Volatility Connectedness Across Equities, Foreign Exchange, Commodities, and Cryptocurrencies: Daily Variance Decompositions and Trading-Session Evidence},
  institution = {Université du Québec en Outaouais, Département des sciences administratives},
  number      = {6},
  type        = {UQO Working Paper},
  year        = {2026},
  month       = {nov},
  url         = {https://www.spboucher.ai/research/wp6},
}

More in Finance & Economics