TeX 62.4%
Python 37.5%
1% ============================================================================2\section{Literature Review}3\label{sec:literature}4% ============================================================================56\subsection{From Volatility Spillovers to Connectedness}78The transmission of volatility across markets has been studied since the recognition that large price changes cluster in time \citep{Mandelbrot1963,Fama1965} and that this clustering is correlated across national markets. \citet{Hamao1990} and \citet{Lin1994} documented volatility spillovers between Tokyo, London and New York with univariate and bivariate GARCH models, \citet{King1990} interpreted cross-market transmission as the outcome of rational agents inferring information from foreign price changes, \citet{Longin2001} showed that the correlation of international equity returns rises in bear markets but not in bull markets, and \citet{Forbes2002} showed that measured increases in cross-market correlation during crises are largely a mechanical consequence of higher variance rather than evidence of contagion. Multivariate GARCH specifications, in particular the dynamic conditional correlation model of \citet{Engle2002} and the regime-switching frameworks of \citet{Ang2002} and \citet{Baele2005}, made it possible to track time-varying comovement in moderately sized systems. These approaches, however, describe comovement rather than direction: they do not say which market originates a volatility shock and which absorbs it.910\citet{DieboldYilmaz2009} proposed to answer that question with the forecast-error variance decomposition of a VAR fitted to volatilities. The share of market $i$'s forecast-error variance attributable to shocks in market $j$ is a natural measure of the spillover from $j$ to $i$, and the aggregate of all cross-market shares is a single index of connectedness. Because Cholesky decompositions depend on the ordering of the variables, \citet{DieboldYilmaz2012} adopted the generalized decomposition of \citet{Koop1996} and \citet{Pesaran1998}, which allows for contemporaneously correlated shocks and delivers an asymmetric connectedness table from which directional ``to'', ``from'' and net spillovers follow directly. \citet{DieboldYilmaz2014} reinterpreted the table as a weighted directed network, so that total connectedness is the mean degree of the network and net pairwise spillovers are its signed edges. In their application to four U.S. asset classes, \citet{DieboldYilmaz2012} found that equity volatility transmits to bonds, exchange rates and commodities, and that connectedness rises in crises; this ordering of equities at the top of the hierarchy recurs in most subsequent studies.1112The framework has since been extended in several directions. \citet{Barunik2018} decompose connectedness by frequency to separate short-run from long-run transmission; \citet{Barunik2016,Barunik2017} distinguish spillovers generated by negative and positive returns in equity and foreign-exchange markets; \citet{Antonakakis2020} replace rolling-window VARs with time-varying-parameter VARs that do not discard observations; and \citet{Lastrapes2021} propose a joint spillover index that corrects the row normalization of the generalized decomposition. Statistical inference for connectedness measures has received comparatively little attention. Most applications report point estimates, although the measures are smooth functions of VAR parameters whose sampling uncertainty can be approximated by the bootstrap methods developed for impulse responses \citep{Kilian1998,KilianLutkepohl2017}.1314\subsection{Cross-Asset Transmission: Exchange Rates, Commodities and the VIX}1516Within foreign exchange, \citet{GreenwoodNimmo2016} apply the connectedness framework to the G10 currencies and find that risk spillovers intensify in stress episodes, while \citet{Barunik2017} show that negative-return volatility dominates transmission among the major pairs. The yen's behaviour in crises is central to the present paper's regime results: \citet{Ranaldo2010} and \citet{Habib2012} document that the yen and the Swiss franc appreciate when global risk aversion rises, so that USD/JPY volatility reflects the unwinding of carry positions rather than idiosyncratic Japanese news.1718For commodities, \citet{Arouri2011} and \citet{Mensi2013} document volatility spillovers between crude oil and equity markets and between commodities and equities more generally, \citet{Kang2017} measure spillovers among crude oil, precious metals and agricultural futures and find that they intensify in crises, \citet{Vivian2012} identify structural breaks in commodity volatility, and \citet{Baur2010} and \citet{Reboredo2013} characterize gold as a safe haven for equities and as a hedge for the U.S. dollar. Whether these roles translate into gold being a net receiver or a net transmitter of realized-volatility shocks is an empirical matter addressed below.1920The VIX requires a distinct interpretation. It is the risk-neutral expectation of 30-day S\&P~500 variance implied by option prices \citep{Whaley2009}, and its level is closely related to, but systematically above, realized variance, the gap being the variance risk premium \citep{BekaertHoerova2014}. The present paper does not include the VIX level in the VAR; it includes the realized volatility of 5-minute changes in the VIX, that is, the variability of the implied-volatility price itself. The finding that this series is a net receiver of shocks from equity-futures realized volatility is therefore consistent with the VIX being a derivative of the S\&P~500 and does not contradict the predictive content of the VIX level for future realized variance.2122\subsection{Cryptocurrencies and Traditional Markets}2324Early studies found cryptocurrencies to be largely detached from traditional asset classes. \citet{Bouri2017} conclude that bitcoin is a diversifier rather than a hedge or a safe haven, \citet{Corbet2018} find that connectedness between cryptocurrencies and conventional assets is weak and concentrated within the crypto group, and \citet{Baur2018} show that bitcoin returns are essentially uncorrelated with those of traditional assets. The within-crypto dependence is strong: \citet{Ji2019a} document a dense connectedness network among the largest coins, and \citet{Makarov2020} attribute part of the common dynamics to arbitrage across exchanges. Bitcoin's volatility is also unusual in level and dynamics: \citet{Katsiampa2017} and \citet{Chaim2018} document pronounced jump components, which is relevant for the choice between levels and logarithms of realized volatility in a VAR.2526The COVID-19 episode changed the assessment. \citet{ConlonMcGee2020} find that bitcoin lost value together with equities in March 2020 and did not act as a safe haven, \citet{Bouri2020} find with wavelet methods that neither bitcoin nor gold and commodities offered reliable safe-haven protection for equities at short horizons, \citet{Corbet2020} document contagion from equities to gold and cryptocurrencies, and \citet{Elsayed2022} report increased risk transmission between bitcoin and traditional assets during the pandemic. \citet{Iyer2022} shows that the correlation of bitcoin returns with U.S. equity returns and the share of equity-volatility variance explained by crypto shocks both rose after 2020, and relates the change to the entry of institutional investors. Studies of cryptocurrency--commodity links \citep{Ji2019b} and of meme assets \citep{Yousaf2023} point in the same direction of growing but still modest integration. Whether the direction of transmission has shifted---whether crypto volatility now originates shocks that reach traditional markets, or mainly absorbs them---remains open, and the sign of bitcoin's net spillover reported in the literature varies with the sample and the specification. The present paper addresses this question with confidence intervals, calendar-year estimates and session-ordered predictive regressions.2728\subsection{Realized Volatility and Intraday Periodicity}2930Realized volatility, the sum of squared intraday returns, is a consistent estimator of the integrated variance of a continuous-time price process as the sampling interval shrinks \citep{Andersen2001,Andersen2003,BarndorffNielsen2002}. At very fine sampling intervals, however, bid--ask bounce and discreteness introduce a bias that grows with the number of observations \citep{HansenLunde2006,AitSahalia2005,Zhang2005}. In a comparison of hundreds of realized measures across asset classes, \citet{Liu2015} find that it is difficult to improve significantly on 5-minute realized variance, which justifies the sampling frequency used here. The logarithm of realized volatility is close to normally distributed and strongly persistent \citep{Andersen2003}, a persistence that the heterogeneous autoregressive model of \citet{Corsi2009} captures parsimoniously, and realized measures are themselves estimated with error, with consequences for the dynamics of volatility models \citep{Bollerslev2016}. Jumps add a further component to realized variance that can be separated with bipower variation \citep{Andersen2007}.3132Intraday volatility follows strong deterministic patterns: equity volatility is highest at the open and the close of the cash session \citep{Wood1985,Harris1986}, and the periodicity in exchange-rate volatility tracks the opening hours of the Asian, European and American centres and the release times of macroeconomic announcements \citep{Andersen1997,Andersen1998}. For bitcoin, \citet{Eross2019} find that volatility and volume are highest during U.S. trading hours despite continuous trading, and \citet{Baur2019} document time-of-day and day-of-week effects in returns and volume, with lower activity at weekends. These patterns bear on the alignment problem studied here: a ``day'' of volatility means different things for a market that closes and for one that does not.3334\subsection{Contribution}3536The paper contributes to this literature in four ways. First, it measures cross-asset connectedness with realized rather than range- or return-based volatility for ten instruments spanning equities, foreign exchange, commodities, implied volatility and cryptocurrencies over a decade that includes the COVID-19 crash, the 2022 tightening cycle and the launch of spot bitcoin ETFs. Second, it treats the alignment of 24/7 and exchange-traded series explicitly and quantifies its effect on the estimates: forward-filling weekend observations, a common but seldom documented practice, alters the total connectedness index by five percentage points and the net spillover of equity futures by fifteen. Third, it attaches bootstrap confidence intervals to all connectedness measures and shows that several widely discussed regime results---including the sign of bitcoin's net spillover---are not statistically distinguishable from zero in sub-samples of the usual size. Fourth, it uses the intraday data not only to construct daily realized volatility but to test, with correctly time-zoned session volatilities, whether bitcoin's overnight and weekend trading carries information for the volatility of U.S. equity futures.37