The portfolio metric gives its best score, 1.00, to the Russian Ruble, even though the paper's own Table 1 identifies RUB as the sample's most volatile currency, with mean conditional volatility of 5.496. The spillover map earns attention. The allocation results never establish that trading it makes money.
A volatility spillover map tracks where a jump in one market's volatility travels next, and how much arrives. Ben Ameur and Jamaani construct one for the BRICS currencies: Brazil's real, Russia's ruble, India's rupee, China's yuan and South Africa's rand. All five are exposed to commodity prices and capital flows that can reverse quickly. The surrounding nodes are the US Dollar Index, the yen, Brent crude, oil volatility and a daily geopolitical risk index.
The paper claims that downside shocks travel farther than upside shocks. Rising risk aversion and departing liquidity can make a model that averages both directions understate crisis-period systemic risk. The implied trade reallocates toward assets the network finds least connected when connectedness rises. That rule needs support from the portfolio section. It gets none.
How the map is built
The map has three stages. Each series first receives a conditional variance from an AR(1)-GJR-GARCH(1,1) with skewed Student-t errors, selected using AIC/BIC and log-likelihood. The authors then separate that variance according to the sign of the day's innovation. Good volatility is h_t when the innovation is non-negative; bad volatility is h_t when it is negative. Finally, the resulting volatility series enter a TVP-VAR (time-varying-parameter vector autoregression) at lag one. Diebold-Yilmaz generalized variance decompositions yield TO, FROM, NET and total connectedness.
The daily sample runs from September 2014 to October 2024. It covers BRL, RUB, INR, CNY and ZAR against the dollar, the US Dollar Index, JPY/USD, Brent crude, the CBOE oil volatility index (which the paper labels VOLOIL, and which is OVX), and GPRD, the Caldara-Iacoviello daily geopolitical risk index. Investing.com supplies FX, Brent and OVX; policyuncertainty.com supplies GPRD. Treating GPRD and oil volatility as endogenous nodes makes sense.
Total connectedness reaches 22.93% in the volatility system. RUB and oil volatility remain net transmitters, while CNY and INR are net receivers. In Fig. 2, the time-varying index climbs to nearly 70% during the COVID-19 pandemic in 2020 and the Russia-Ukraine war in 2022. It stays below 20% through 2015-2017. Those readings come from the dynamic series rather than the static 22.93% table.
The paper then builds three optimizers from the TVP-VAR covariance and pairwise-connectedness matrices: minimum variance, minimum correlation and minimum connectedness. Ederington risk-reduction effectiveness supplies the scores.
The transmitter ranking survives inspection
RUB dominates the decomposed system as a net transmitter. Its NET readings are +18.87% for good volatility and +17.74% for bad volatility, alongside TO values of 34.31% and 37.22%. The broader transmitter finding rests heavily on this single node.
CNY occupies the opposite end, with NET of -11.34% and -13.42%. Its own shocks explain roughly 70.3% of variance in both blocks. Table 1 reports mean conditional volatility of 5.496 for RUB, against 1.145 for BRL and 0.464 for ZAR. RUB's variance is 333.251. The difference spans two orders of magnitude, rather than one.
The largest pairwise connection in the bad-volatility block links the two reserve currencies. USDX receives 14.77% from JPY, while JPY receives 11.14% from USDX. Every BRICS pair in the table is smaller.
GPRD barely joins the system it is presented as driving. This headline variable from the abstract retains 95.50% of its own variance in the total conditional-volatility table and sends only 3.94% to all other nodes combined. Its own-variance shares are 90.95% in the good table and 94.57% in the bad table.
Which USDX is the transmitter?
Table 2 identifies USDX as the volatility system's largest net receiver, at -16.30%. Yet the sign changes with the model input. The discussion of Figure 5 says: "The USDX consistently acts as a net transmitter, reflecting its dominant role in the global financial system."
Table 4 labels its rows r_INR through r_JPY, suggesting an estimate based on returns rather than fitted volatilities. USDX records TO of 56.23% and NET +10.53% there, and the paper describes it as the dominant volatility transmitter. Both sets of figures can describe their respective objects. The text never explains that Table 4 uses a different input, while continuing to call USDX the dominant volatility transmitter. Anyone citing the paper as evidence that the dollar dominates transmission needs to identify the table behind the claim.
The static evidence for asymmetry is slight: 25.54% for bad volatility and 25.22% for good volatility. The gap is 0.32 percentage points, and we did not find a significance test for it. The decomposition applies an indicator split to one conditional variance. Good and bad volatility are therefore the same fitted process, sampled on alternating days. This construction differs from the realized semivariances in Baruník et al. (2016), which the paper cites.
The time-varying series carries the asymmetry case. Bad-volatility connectedness exceeds 75% in 2020. No test accompanies the plot.
A single-asset benchmark drives RE
The authors expressly limit the hedging interpretation. Their portfolio section says the minimum variance, minimum correlation and minimum connectedness portfolios "are interpreted as portfolio optimization tools rather than explicit hedging models". It also says RE "captures portfolio-level risk reduction relative to individual assets and is therefore interpreted as a diversification-efficiency measure".
RE, or Ederington risk-reduction effectiveness, subtracts the ratio of portfolio variance to single-asset variance from one. A diversified portfolio is being compared with holding one asset alone.
The abstract nevertheless says USDX, JPY and Brent crude oil "contribute to greater portfolio resilience and risk reduction during periods of uncertainty, while asymmetric portfolio dynamics reveal that adverse oil shocks intensify systemic risk, especially for the Russian Ruble and South African Rand". The conclusion adds that "the results show that USDX and JPY consistently deliver the strongest diversification and risk-minimizing contributions across all portfolio specifications".
The minimum variance RE column gives a different ordering: RUB 1.00, ZAR 0.99, BRL 0.99, JPY 0.96, USDX 0.94, BRENT 0.94, INR 0.91 and CNY 0.81, all at p = 0.00. The reserve currencies finish fourth and fifth.
RUB leads because holding it alone is the sample's worst alternative.
Its conditional volatility series has variance of 333.251 in Table 1. Almost any diversified portfolio improves on that baseline, which is exactly what RE rewards. Cross-asset RE rankings consequently reveal nothing about which asset stabilizes the portfolio, despite the conclusion reading as though they do.
The weights point elsewhere. CNY receives 0.32 in the minimum variance portfolio and USDX gets 0.28. JPY follows at 0.14, INR at 0.13 and Brent at 0.08. The currencies leading the RE column receive almost no capital: RUB 0.01, BRL 0.01 and ZAR 0.02. The optimizer largely rejects the same assets that RE favors.
CNY is the largest holding, yet it posts the weakest diversification score under two of the three specifications. Its minimum correlation RE is 0.26. In the minimum connectedness portfolio (MCoP), RE falls to -1.08, meaning that the connectedness-minimizing portfolio has greater variance than a CNY-only position. In the same MCoP column, USDX scores 0.31 and JPY 0.59. INR's MCoP RE of 0.05 is the table's sole insignificant result (p = 0.20).
Brent receives the MCoP's largest weight, 0.19, despite retaining 91.26% of its own variance in the return system. USDX receives 0.05.
The portfolio discussion also says, "ZAR (0.05) records the lowest mean return". The figure is a mean weight, and the column contains no returns.
Trading evidence is missing
We did not find transaction costs, turnover, or a walk-forward split anywhere in the portfolio section. None of the three portfolios has a reported Sharpe, mean return or drawdown. Yet the recommendation entails daily reallocation, with weights that visibly move during crises.
The paper describes one GJR-GARCH fit across the full sample and never mentions rolling or expanding estimation. The TVP-VAR then treats those fitted volatilities as data. First-stage error is not propagated, and the inputs remain in-sample. Two of the nine nodes, GPRD and OVX, cannot be held as positions. RUB quotes after February 2022 also fall under capital controls.
The concluding section acknowledges the limited scope and the omission of macro fundamentals and monetary policy. Costs and out-of-sample validation receive no such discussion.
We could not run this ourselves. Our coverage includes no FX spot or FX futures for BRL, RUB, INR, CNY or ZAR, while the oil leg uses Brent rather than a supported US contract. GPRD is absent from our macro feed as well. Because the FX legs drive the paper's transmitter result, replacing them with dollar-proxy ETFs and substituting WTI would leave its mechanism untested. We have described a related failure before, in which no instrument spanned the priced object (the correlation rotation premium).
A walk-forward minimum variance and minimum connectedness portfolio on the tradeable subset would change my view. The weights would need to use only information available at time t, then be assessed on Sharpe after a realistic FX spread and compared with an equal-weight portfolio.
The connectedness results still make a useful risk map. Total connectedness of 22.93% and the Fig. 2 crisis peaks near 70% deserve a place on a monitoring dashboard. An allocation claim needs a measure that compares assets with one another, rather than rewarding each according to how bad it is to hold alone.