Chen and Luo's connectedness table offers little evidence of crypto moving stocks: roughly 38 points out of a 417.19-point network link the four coins with the five equity indices. Two coin pairs account for 156. The abstract claims that "R² decomposed connectedness analysis reveals substantial cross-market linkages". The equity indices carry those linkages. Their block contributes 185.57 points, including 21.77 and 21.31 for GER-UK alone. By comparison, all the cells connecting coins and stocks total 38.36.

Building the weekly network

The authors use weekly prices from December 9, 2018 to October 26, 2025, sourced entirely from investing.com. The four coins are Bitcoin, Ethereum, Tether and USD Coin. The five equity indices are Shanghai A-shares, the Dow Jones Composite, the DAX, the Nikkei 225 and the FTSE 100, each in local currency. An adjusted BIC selects a GARCH-family model for each series from six specifications and five error distributions. The resulting volatility estimates enter a two-step DCC model to produce a correlation matrix for each week.

For connectedness, the authors follow Cocca et al. (2024), departing from the usual variance-decomposition approach. They calculate each asset's regression R² against the other eight assets from that week's conditional correlations alone. A Genizi decomposition assigns a share of the R² to each regressor; these shares become the table's cells. Rows sum to FROM, columns sum to TO, and their difference gives NET. An average produces the Total Connectedness Index (TCI).

The paper reports a TCI of 52.15%. BTC and ETH contribute 43.05% and 41.9% to one another; the corresponding figures for USDT and USDC are 35.6% and 35.46%. China has a FROM of 3.4% and a TO of 2.51%. The net transmitters are the U.S. (NET +2.27), Germany (+1.77) and BTC (+1.75). Japan (-2.08), the UK (-1.03) and China (-0.9) are net receivers.

The TCI reaches its peaks in early 2019, the first half of 2020 and mid-2023. The authors describe the crypto crash, COVID-19 and regulation only as "possible explanations". Their allocation recommendations are separate risk limits for the two coin types and a liquidity buffer held in stablecoins.

Why divide a nine-asset total by eight?

In Table 3, both the FROM column and the TO row sum to 417.19. Dividing across nine assets gives 46.35%. To get the reported 52.15%, the divisor must be eight.

The notation gives K two roles. Equation 1 makes the return vectors (K+1) by 1 and Hxy K by 1. In that formulation, K is the count of eight regressors, so 1/K yields 52.15. But Equation 7's sum, along with the phrase "K univariate GARCH models are estimated", treats K as nine.

The paper says that "approximately 52% of the variance in each market's returns can be explained". Each FROM entry is an R². Their mean across the nine assets is 46.35%, so that description overstates the average share explained.

The small share crossing between coins and stocks

Our arithmetic on Table 3 puts the coin-to-equity cells at about 19.2 points, with about the same total going from equities to coins. Together, they account for roughly 9% of the 417.19. ETH's contribution to the U.S., at 4.24%, is the largest coin-to-equity cell. In reverse, the U.S. contributes 4.39% to ETH. BTC contributes 1.85% to the U.S.; their Kendall correlation is 0.149. About 37% of the network comes from the two within-subtype pairs alone.

The authors capture this accurately in their own sentence: "spillover effects within the same subtype are far stronger than those across subtypes". The two coin pairs contribute 156.01 points, compared with 38.36 for all cells connecting coins and stocks. Crypto-to-equity contagion is a small residual in this table.

Squaring loses the sign. Contemporaneous regressions lose the lag.

The mutual R² contribution between USDT and USDC is 35.5%, despite a Kendall correlation of -0.296. Squaring cannot distinguish offsetting peg deviations from co-movement. The conclusion goes on to claim a strong positive correlation (>0.5) between USDT and USDC. This conflicts with the -0.296 in Table 1 and the abstract's statement that "stablecoins show negative correlations with other assets".

Direction presents the larger problem. All the regressions are contemporaneous. A pair's cells differ because each asset's R² uses a different set of regressors. The label "transmitter" therefore comes from bookkeeping among correlated regressors; no lead-lag test supports it.

All full-sample average NET values fall within ±2.27 points. We found no standard errors or bands for any of them. The U.S. leads the NET ranking by just 0.52 points over BTC. BTC, meanwhile, has the highest TO across the assets: 61.65%, compared with 59.73% for the U.S.

Five decisions left to the implementer

We could not reproduce the system

Our data lack established index histories for the Chinese, German, Japanese and UK markets, as well as price histories for USDT and USDC. Restricting the network to U.S. equities and the coins we hold would estimate a different transmission structure. We therefore did not run that reduced network.

Useful for monitoring, untested for allocation

China's isolation is the paper's clearest finding. Its FROM is 3.4%, against Japan's 35.38%; the gap is visible without a significance test. For a book trading the U.S., German, Japanese and UK indices, that gap is the one number here worth following weekly.

The allocation recommendations have less support. The paper proposes stablecoins as a buffer and BTC and ETH as diversifiers. We did not find a backtest, hedge ratio or hedging-effectiveness number supporting either recommendation. Our view would change with a connectedness-weighted portfolio that beat equal weight out-of-sample after costs.

Until someone demonstrates that, use Table 3 as a description of which assets clustered together over 2018 to 2025.