% ============================================================================
\section{Introduction}
\label{sec:introduction}
% ============================================================================

How volatility propagates from one market to another matters for portfolio construction, for the design of risk limits, and for the monitoring of systemic risk. Over the past two decades the channels of propagation have multiplied: index futures and exchange-traded funds trade almost around the clock, cross-asset strategies rebalance mechanically in response to volatility, and cryptocurrencies have grown from a marginal curiosity into an asset class with a market capitalization in the trillions of dollars and, since January 2024, spot exchange-traded funds listed in the United States. Whether these developments have altered the hierarchy of volatility transmission---which markets originate volatility shocks and which absorb them---is an empirical question, and one whose answer appears to change with the sample period and with the econometric choices made along the way.

The connectedness framework of \citet{DieboldYilmaz2009,DieboldYilmaz2012,DieboldYilmaz2014} offers a transparent way to pose the question. A vector autoregression (VAR) is fitted to a panel of volatilities, and the forecast-error variance of each series at a given horizon is decomposed into the shares attributable to shocks in every series in the system. Aggregating the off-diagonal shares yields a total connectedness index; summing them by row and by column yields the volatility each market receives from, and transmits to, the others. The generalized decomposition of \citet{Koop1996} and \citet{Pesaran1998} makes these quantities invariant to the ordering of the variables, which has made the framework the standard tool for cross-market spillover studies in equities \citep{Demirer2018,DieboldYilmaz2016}, foreign exchange \citep{Barunik2017,GreenwoodNimmo2016}, commodities \citep{Kang2017,Ji2019b} and cryptocurrencies \citep{Ji2019a,Corbet2018}.

Three features of the existing evidence motivate this paper. First, most cross-asset studies measure volatility with daily ranges or squared daily returns. Realized volatility computed from intraday returns is a far more precise ex post measure \citep{Andersen2003,BarndorffNielsen2002}, and the precision gain is greatest exactly when connectedness is of most interest, namely on turbulent days. Second, studies that include cryptocurrencies alongside exchange-traded instruments must confront a calendar problem: bitcoin and ether trade every hour of every day, whereas index and commodity futures close for roughly one hour a day and from Friday evening to Sunday evening, and the VIX is disseminated only during Chicago trading hours. How the series are aligned---which days enter the panel, how weekend crypto volatility is treated, and in which time zone the day is defined---is rarely documented, although, as this paper shows, it can shift connectedness estimates by several percentage points and change the sign of net spillovers. Third, point estimates of connectedness are commonly reported without confidence intervals, and regime comparisons are drawn from sub-samples of a few dozen observations in which a ten-variable VAR is severely over-parameterized.

This paper studies ten instruments observed at the 5-minute frequency from January 2008 to March 2026: the E-mini S\&P~500 (ES) and Nasdaq~100 (NQ) futures, the EUR/USD, USD/JPY and GBP/USD exchange rates, gold (GC) and WTI crude oil (CL) futures, spot bitcoin (BTC) and ether (ETH), and the CBOE Volatility Index (VIX). Daily realized volatility is constructed for each instrument from 5-minute log returns, and the connectedness of log realized volatility is estimated on the 2{,}514 trading days common to all ten series, from March 2016 to March 2026, with a VAR(4) and a ten-day forecast horizon. Residual-bootstrap intervals accompany every connectedness measure. The time dimension is explored with rolling windows of 200, 250 and 500 trading days, with four regime sub-samples, and with calendar-year estimates. A second block of analyses exploits the intraday data directly: after an explicit conversion of exchange time stamps to Coordinated Universal Time, it documents the hourly periodicity of absolute returns, the share of daily variance generated in Asian, European and U.S. trading hours, and session-ordered predictive regressions between the realized volatility of equity futures, the VIX and bitcoin.

The paper addresses five questions. (i)~What is the structure of daily realized-volatility connectedness across the four asset classes, and which instruments are robust net transmitters or receivers? (ii)~How does this structure change across market regimes, in particular during the COVID-19 crash and the 2022 monetary tightening? (iii)~Has the integration of bitcoin and ether with traditional markets increased, and in which direction does volatility flow? (iv)~How does aggregate connectedness evolve over time, and how is it related to stress episodes? (v)~How is intraday volatility distributed across trading sessions, and does bitcoin's round-the-clock trading carry information for the volatility of U.S. equity markets?

The main results are as follows. Cross-asset shocks explain 54.8\% of the ten-day forecast-error variance of log realized volatility (90\% bootstrap interval 53.1--56.9). ES and NQ are the only robust net transmitters, with net directional spillovers of $+36.0$ and $+25.6$ percentage points; the realized volatility of the VIX ($-20.6$), of crude oil ($-14.2$) and of sterling ($-13.2$) are robust net receivers. The VIX result is not a statement about the VIX level---which is a forward-looking price---but about the day-to-day variability of that price, more than one-fifth of which is explained by shocks to the realized volatility of the equity futures from which it is derived. The net positions of bitcoin and ether are statistically indistinguishable from zero in the full sample and in most sub-samples, and the point estimate for bitcoin changes sign across reasonable specifications. The two cryptocurrencies form a tightly linked pair---each explains 27--28\% of the other's variance---whose exposure to traditional assets has grown markedly: shocks to the eight traditional instruments explained less than 10\% of bitcoin's forecast-error variance in 2017 and 2019, between 20\% and 34\% in 2020--2024, and 46\% in 2025. The direction of this integration is from equities to crypto. In the spot-ETF period (January 2024--March 2026) bitcoin is a significant net receiver ($-12.4$ percentage points, interval $-20.4$ to $-2.1$), and in the intraday analysis the U.S.-session volatility of ES predicts bitcoin's next Asian-session volatility, whereas bitcoin's overnight and weekend volatility carry no incremental information for the U.S.-session volatility of ES. Rolling total connectedness rose sharply in March 2020 and reached the highest values of the sample in 2025; a previously reported peak in February 2018 turns out to be an artefact of forward-filled weekend observations. Intraday, equity and VIX volatility concentrate in the U.S. cash session, which generates 53--63\% of daily variance, while bitcoin's hourly profile is nearly flat, with a peak-to-trough ratio of 1.2 against 3.9 for ES.

Relative to the May 2026 version of this paper, the present version corrects the construction of the daily panel, replaces levels by logarithms of realized volatility, adds bootstrap inference and a robustness grid, corrects the time zone of the intraday analysis, and revises the conclusions accordingly; the changes are documented in the appendix and in the revision log that accompanies the replication code. The remainder of the paper is organized as follows. \Cref{sec:literature} reviews the literature. \Cref{sec:methodology} presents the realized-volatility measure, the connectedness framework, the inference procedure and the session analysis. \Cref{sec:data} describes the data, the alignment of the series and the sample. \Cref{sec:results} reports the results, \Cref{sec:discussion} discusses them in relation to the research questions and the literature, and \Cref{sec:conclusion} concludes.