The optimizer leaves the token at zero or below. Against the S&P 500 ESG Elite index, Chainlink receives an average full-sample minimum-variance weight of -0.01, while the index receives 1.01. LINK gets the same -0.01 / 1.01 allocation against the SDG index, the New Economies composite and the Biodiversity index. BAT gets it against all four. For MKR, the split is 0.00 against 1.00 in each index leg.
Daga and co-authors acknowledge the result in the abstract. Near-zero or negative weights are consistently allocated to DeFi assets, while sustainability instruments dominate cross-category portfolios, confirming that mixed portfolios generate risk-adjusted returns superior to single-category holdings, "a benefit that is resilient even during-crisis periods". The concession and conclusion share a sentence. The conclusion section repeats the same claim: DeFi-sustainable cross-category allocations produce superior, crisis-resilient risk-adjusted returns.
Both parts of that claim rely on conditional correlations in Table 10 and weight signs in Table 11. No return, no volatility and no Sharpe ratio is reported for any of the nine sub-periods.
The construction
Market segmentation supplies the economic story. DeFi tokens operate under different mandates, microstructure and regulation from ESG or biodiversity equity. Shocks should therefore travel weakly between them, allowing a portfolio spanning both categories to carry less variance than one restricted to either. The proposed gain comes from the correlation gap. Return forecasts play no role.
We could not trade the paper's universe. An implementable book would use US-listed clean energy, green bond, ESG/sustainability and innovation ETFs alongside LINK, MKR and BAT where covered. The paper's connectedness estimates, optimal weights and reported risk-adjusted performance apply specifically to its named S&P index universe. Nothing below tests the paper.
Daga, Oben, Seraj and Eyüpoğlu study ten instruments from June 24, 2019 to April 17, 2025, using 1,398 daily observations for each. The sample contains three tokens, LINK, MKR and BAT, plus seven S&P Global indices spanning clean energy, cleantech, green bonds, ESG, SDG, New Economies and biodiversity. Absolute log returns proxy for volatility. A Diebold-Yilmaz generalized variance decomposition, estimated with VAR lag 1 and a 200-day rolling window, produces total, directional and net connectedness. Time-varying pairwise correlations come from a DCC-GARCH with Student's t errors. The authors then use those correlations to calculate two-asset minimum-variance weights for all 45 pairs, averaging them across the full sample and nine crisis sub-periods.
The Total Connectedness Index is 60.6%. In the authors' decomposition, 605.8 of 1000 units of forecast error variance cross between instruments, leaving 39.4% idiosyncratic. Direction matters more here. All three tokens are net receivers: LINK -7, MKR -9.9 and BAT -8.1. All three green instruments are receivers too. At -29, the S&P Green Bond Index absorbs more than any token and is the system's largest net receiver.
Biodiversity leads the transmitters, with TO 100.1, FROM 76.4 and net +23.7. ESG follows at +20, then SDG at +19.6. Cross-category cells remain tiny. BAT receives 2.17 from ESG and LINK receives 2.35, compared with the 22.66 that ESG receives from SDG.
Do the weights mostly rank volatility?
Conditional variances and one covariance determine the two-asset weights.
Expected returns never enter.
Table 4 therefore sets much of the ordering. Daily standard deviation is 6.8221 for LINK, 6.6234 for MKR and 6.3059 for BAT. It falls to 1.3732 for ESG, 1.4062 for SDG and 0.45154 for the Green Bond Index. With a five-to-one volatility ratio and a full-sample conditional correlation of 0.25 for LINK/ESG, the high-volatility asset has almost no function in the minimum-variance pair.
The same machinery produces the stranger cells. Paired with ESG equity, the green bond index receives 0.88 of the full-sample allocation and 0.94 in the post-election window. SDG against Biodiversity yields -2.45 / 3.45. Their volatilities are nearly identical at 1.4062 and 1.3772, while conditional correlation reaches 0.99, shrinking the denominator. Clean energy against cleantech gives 1.18 / -0.18, meaning 1.18 in clean energy financed with a 0.18 short in cleantech.
A low-volatility bond index and two near-duplicates naturally produce such minimum-variance allocations. Those cells do not establish DeFi diversification.
H4 concerns superior risk-adjusted returns. Yet the portfolio evidence consists of conditional correlations in Table 10 and weights in Table 11. No Sharpe ratio, realized return, portfolio volatility or drawdown appears for any of the 45 pairs. The paper constructs no benchmark portfolio, whether single-category or otherwise. Turnover is also unreported, although the weights are re-derived daily. We have priced such trading before when authors supplied the required figure, in a growth-defensive timer that survived 470% annual turnover. A reader cannot do that arithmetic here.
Four labels, one sustainability exposure
Within the sustainability cluster, full-sample conditional correlation reaches 0.96 for ESG/SDG, 0.97 for ESG/BI and 0.99 for SDG/BI. ESG, SDG and BI all select constituents from the S&P 500 through overlapping sustainability screens. The authors identify this issue twice. They write that high pairwise connectedness among sustainable investments "may reflect shared constituent membership" rather than independent cross-category transmission. With single indices representing each category, they cannot separate those channels, though they argue index construction alone is unlikely to explain the co-movement.
The sensitivity check gives clearer evidence. Removing Biodiversity, the largest transmitter, lowers the TCI from 60.6% to 54.8% without changing any remaining net sign.
Six points of connectedness is the contribution of the system's biggest node.
Overlapping crisis clocks
The crisis analysis uses nine sub-periods around what the paper describes as four successive and overlapping crises. Some windows are wholly nested. During-Russia-Ukraine extends from February 24, 2022 to April 17, 2025. It fully contains the during-to-post-banking-crisis period beginning in March 2023, which explicitly includes "the subsequent stabilization phase", along with the November 5, 2024 policy window. Claims about which shock transmitted the most therefore compare nested samples.
The dynamic TCI stands more clearly on its own. It falls from 75% to 54% in December 2020, then reaches 43% by February 2021. From March 2022 to April 2023 it holds within 58-67%, records 51% in April 2023 and begins rising again from October 2024.
One sensitivity result warrants attention. The static TCI remains 60.6% with 100-, 200- and 300-day windows, as a full-sample decomposition would regardless of the window. Changing the volatility proxy moves the estimate. It is 60.6% using absolute returns, 60.4% with 3-day rolling standard deviations, 72.4% on squared returns and 76.8% from GARCH conditional variances. The paper treats these results as convergence, which holds for sign and broad magnitude class. A 16-point range still makes the reported level dependent on the proxy.
Token timing adds another wrinkle. The sample retains only dates with observations for all ten instruments. A Monday token return consequently covers three calendar days, while the corresponding index return covers one.
A tradable test remains missing
We did not backtest this. The universe depends on the S&P Green Bond Index and six other S&P index series, and we lack price history for the green bond index. The closest implementable substitute would pair US-listed clean energy, green bond, ESG and innovation ETFs with LINK, MKR and BAT. The 60.6% TCI, all net spillover signs and every weight in Table 11 belong to the named index universe. They do not transfer to that substitute.
A realized series would change my view. Use the paper's full-sample DeFi-sustainability cells: the -0.01 / 1.01 pairs for LINK and BAT, and the 0.00 / 1.00 pairs for MKR. Rebalance daily and compare them with a 100% index holding after costs, with turnover disclosed. A Sharpe improvement from those weights would establish the cross-category claim.
Until that figure is reported, the paper supports a narrower result that remains useful. DeFi-to-sustainability spillovers are near zero. The sustainability indices transmit rather than absorb on net, while the green bond index absorbs more volatility than any of the three tokens.