A 10-point VIX spike buys too little Bitcoin-ICLN decoupling to move a two-asset book. Al-Harbi estimates that each one-point rise in the VIX accompanies a 0.0009 fall in their rolling correlation. That makes the change 0.009 for a 10-point spike. The sign looks credible, and it persists across the COVID split. The size matters more to a desk than the paper's line that "fear does not unite these assets; it divides them."

The paper studies how volatility passes between Bitcoin and renewable energy markets. Proof-of-Work mining supplies a plausible link: its electricity demand connects crypto to power markets and, potentially, renewable energy stocks. Al-Harbi tests mining alongside investment sentiment and policy news. Mining is insignificant (P=0.612); policy is too (P=0.781). The VIX, used as the sentiment proxy, is the sole channel with a significant coefficient, and its sign is negative.

From daily prices to rolling correlation

Al-Harbi uses daily Yahoo Finance prices from February 10, 2017 to August 29, 2024 for three instruments: BTC-USD, the iShares Global Clean Energy ETF (ICLN) and SPY. Each return series passes through a separate GJR-GARCH(1,1), which allows negative shocks to raise variance more than positive shocks. The dependent variable is the 60-day rolling correlation between the resulting BTC and ICLN standardized residuals.

An OLS regression relates that correlation to four lagged inputs. The paper calls three of them "transmission channels": daily percentage change in the Digiconomist Bitcoin energy-consumption estimate (energy), the VIX level (sentiment), and a hand-coded +1/−1/0 dummy for major regulatory news (policy). The control is the 60-day rolling correlation between BTC and SPY residuals. Errors are Newey-West.

Al-Harbi proposes a "flight to relative quality" inside risk assets. In a panic, investors supposedly distinguish cash-flowless Bitcoin from regulated renewables backed by infrastructure, pulling their returns apart. The VIX coefficient is −0.0009 (P<0.001), while the BTC-SPY control is −0.7012 (P<0.001); R² is 0.541. Energy (P=0.612) and policy (P=0.781) remain insignificant. The GARCH estimates add little to the case. Bitcoin's asymmetry term is marginal (0.0805, P=0.089), and ICLN's variance is highly persistent (beta 0.9407).

What does a VIX spike buy?

A 25-point VIX move, 15 to 40, implies a correlation change of 0.0225 using the full-sample coefficient. The paper gives no mean correlation against which to judge that move. Its conclusion says diversification benefits "may actually increase during periods of market turmoil." The estimated direction supports that wording. In a two-asset portfolio, though, the variance cross term scales with correlation, and a 0.0225 change barely shifts the book's vol.

The P-values warrant some care. Al-Harbi reports a Durbin-Watson of about 0.08 and correctly uses HAC errors. Yet 1,150 observations made from overlapping 60-day windows contain far fewer independent observations. Persistence on both sides inflates the R² of 0.541. The VIX also enters in levels, leaving two slow series in a regression that could produce a spurious fit. We raised the same levels-regression concern about a clean-energy paper in an earlier review.

The COVID split holds up

Al-Harbi earns credit here. Before March 2020, the VIX coefficient is −0.0011 (P<0.01, N=450). Afterward it is −0.0009 (P<0.001, N=700). The BTC-SPY control moves from −0.6511 to −0.7133. Across two regimes, the VIX estimates are strikingly close.

The stable sign still comes with a small effect.

We could not resolve two bookkeeping questions. Forward-filling prices into a continuous daily series over seven and a half years yields about 2,750 calendar days, yet the full regression reports 1,150 observations. We found no explanation for the gap. The paper lists a 90-day window as a robustness check, and its abstract claims robustness across time windows. We did not find the 90-day estimates in the results tables.

The control behind the fear result

At −0.7012, the BTC-SPY correlation control has the paper's largest coefficient. Its negative sign is odd. ICLN is an equity ETF, and equity ETFs co-move with SPY; the naive expectation is that Bitcoin's correlations with equities and clean-energy equities would rise together. Al-Harbi interprets the sign as substitution: when Bitcoin trades more like a tech stock, renewables become the better diversifier against it.

The construction offers another possible reading. The dependent variable and control are rolling correlations sharing the same BTC residual series, which gives them a common input. Forward-filling makes ICLN and SPY post zero returns every weekend while Bitcoin trades. Both residual correlations are then pulled toward zero on the same days.

That control matters because the VIX result is conditional on it. Al-Harbi takes variance inflation factors of 2.3876 for VIX and 2.3872 for the control as sufficient to clear multicollinearity, since both are under 5. That threshold leaves another question open. A VIF of 2.39 means about 58% of VIX variance is shared with the control (1 - 1/2.39). We did not find a VIX-only specification, so the sign without the control is unknown. The discussion nevertheless says heightened fear "causes" decoupling, a strong claim for reduced-form OLS on lagged proxies. The paper concedes a narrower limitation: its crude policy dummy, which Al-Harbi proposes replacing with an NLP-based index. Appendix A, cited for the coded-event list, did not appear in the published text we read.

Trading the BTC-ICLN pair

The tradable version holds BTC and ICLN at inverse-volatility weights, raising gross exposure when the VIX close from day t−1 lies high in its trailing distribution. The test must begin after August 29, 2024, the end of the paper's sample. Its first result should compare the realized BTC-ICLN correlation drop in high-VIX windows with the paper's implied 0.009 per 10 points. We are running this test now.

Our prior is that the overlay adds turnover for little gain: a correlation change of 0.0225 cannot pay for many rebalances. A VIX-only regression retaining the negative sign would strengthen the case, as would a post-2024 correlation drop an order of magnitude larger than the coefficient predicts.