Precious metals become correlated cargo when diversification matters most. The authors issue that warning themselves: uncertainty and metals have a weak link in ordinary conditions, then a strong one under stress. Yet the paper interprets the symmetry of VIX-gold and VIX-silver connectedness as flight-to-quality evidence. Because a variance-decomposition share has no sign, traders should focus on that inference.
What the model measures
Akgül and Yürük study daily log returns from 11 May 2007 to 27 October 2025. Six series enter the model: the VIX, the CBOE oil volatility index OVX, the Caldara-Iacoviello geopolitical risk index GPR, and spot gold, silver and platinum. Each return is 100 times the log price ratio. Investing.com supplies VIX, OVX and metals data, while GPR comes from the Economic Policy Uncertainty database. Data availability fixes the opening date. The authors state that May 2007 is the oldest OVX series they could obtain.
A disclosure belongs beside that setup. The paper studies spot gold, silver and platinum against the VIX, OVX and GPR indices. An implementation here would trade the US-listed ETFs GLD, SLV and PPLT. It would also require uncertainty inputs we do not all carry, including the daily GPR index and index-level OVX. We ran nothing.
The authors estimate a quantile VAR using two lags selected by BIC, a 200-day rolling window and a 20-step-ahead generalized forecast error variance decomposition (GFEVD). Their extension is the quantile-on-quantile connectedness method of Gabauer and Stenfors. It lets a shock start at one sender quantile and arrive at another receiver quantile. A single-quantile VAR places both variables in the same conditional state; QnQ removes that restriction.
For each variable, the grid contains 0.05, 0.275, 0.50, 0.725 and 0.95. The outputs are total connectedness, net directional connectedness, and a division between directly related and reversely related links. This last split separates positive-comovement from negative-comovement components within the same spillover.
The possible trade follows directly. If metals separate from uncertainty in calm markets and reconnect under stress, a static metals allocation is overpriced as insurance. Position size should then depend on the state of the uncertainty distribution. The paper identifies that possibility without constructing the trade. The authors reserve optimal portfolio weights and quantile-based hedging efficiency for future research.
Its headline finding is clear. Total connectedness remains low under normal conditions and climbs sharply in the extremes. VIX and OVX have their strongest metal links at the 0.95 quantile of the uncertainty index and the 0.05 quantile of metal returns. GPR instead reaches its peak at the joint 0.05 quantiles. At extreme quantiles, the uncertainty indices become net transmitters.
Does 0.95 against 0.05 establish flight to quality?
The abstract states the tail result directly: connections are weak under normal market conditions and quite high under extreme conditions. The conclusion accepts the portfolio consequence. A significant rise in linkage during extreme shocks reduces diversification benefits, and portfolio strategies may become ineffective during market stress. The paper puts the uncomfortable result up front.
The authors respond by recommending dynamic rebalancing, with portfolio weights continually adjusted for market conditions and changing risk-spread patterns among the metals. Nothing in the study tests that prescription. There are no weights, no hedge ratios, no cost assumption and no out-of-sample window.
Elsewhere, the conclusion says the study proves that sudden equity-market uncertainty pushes investors toward precious metals. The cited support is the symmetry of VIX-gold and VIX-silver links during the 2008 crisis, the 2014 to 2016 oil collapse, the 2018 trade war and COVID-19. A GFEVD share attributes variance. Similar directly and reversely related components mean the positive and negative comovement channels have comparable size. They do not establish that gold rises.
For crisis gains, the paper relies on external price context. Gold rose from roughly $600 an ounce in 2007 to over $1000 during the crisis. Silver dropped from between $12 and $14 an ounce to $10, while platinum fell from $1300 to $800. In Q1 2020, gold gained 6.9%, versus silver at minus 2.3% and platinum at minus 20%. Those are useful figures from the IMF and the World Bank.
The model did not produce them.
Another reported result receives little attention. Net risk transmission from the uncertainty indices is strongest at the lowest and highest quantiles of those measures, while it peaks at the average quantiles of the metals. Total connectedness therefore peaks tail-to-tail, whereas net transmission peaks tail-to-middle. The pictures assign different roles to sender and receiver.
Platinum breaks from gold
The descriptive table contains the most practical evidence. Kendall's tau between VIX and gold is minus 0.008 and insignificant. For OVX and gold it is minus 0.018, also insignificant. VIX against platinum is minus 0.129, while VIX against silver is minus 0.100; both are significant at 1%. Platinum has the most negative VIX tau at minus 0.129, followed by silver at minus 0.100. Gold's minus 0.008 cannot be distinguished from zero.
The cross-sectional values also vary widely. Gold and silver co-move at 0.597. Gold and platinum reach only 0.416, with silver-platinum at 0.455.
Connectedness pushes platinum farther away. The paper calls its response to VIX erratic and weak under all market conditions. Its VIX linkage remains asymmetric across the sample apart from the 2018 trade war. Platinum returns also show significant serial correlation, with Q(20) of 25.515. Gold at 5.387 and silver at 10.988 do not. Platinum's squared returns have the table's strongest volatility clustering at 1444.630. The table twice contradicts anyone treating PPLT as cheaper gold because the metals supposedly share safe-haven status.
GPR warrants more scrutiny. Its Kendall taus with gold, silver and platinum are 0.004, 0.014 and 0.016, all insignificant. Return variance is 1629.767, compared with 59.925 for VIX, and Q(20) reaches 939.686. A footnote acknowledges the interpretation problem: these log returns capture sudden increases and decreases, rather than absolute high or low levels of the uncertainty indices. The problem remains. A 0.95 quantile of the daily change in a news-count index is still difficult to position against.
Heat maps without estimates
The sensitivity analysis has substance. It uses a 250-day window, a one-lag version and a 10-step horizon. The authors also rerun the full model with the Financial Stress Index from the Office of Financial Research (OFR), which combines over thirty indicators. This last check is the paper's most informative result. The FSI transmits net risk only in the upper quantiles, while metals transmit net risk at low and average stress. Regime changes reverse the direction, a stronger finding than tail amplification by itself.
The main connectedness results appear as heat maps and time series in Figures 2 through 6. We found no tabulated connectedness values, standard errors or bootstrap bands anywhere in the text. Readers therefore cannot tell whether the 0.95-by-0.05 cell is two points or twenty above the grid's middle. The paper never states the number of observations. It includes no forecasting test, cost assumption or portfolio, omissions the authors describe as scope rather than oversight.
We did not attempt this. Any implementation available to us would trade ETF wrappers rather than the spot metals examined in the paper.
A tabulated grid with bootstrap bands would change my view, followed by a conditional metals sleeve sized from the FSI quantile and tested out of sample with costs. Until then, this paper works best as a warning dial: when uncertainty measures enter their 0.95 tail, treat the metals sleeve as correlated cargo.