Does gold predict crypto risk? First it would have to pick a side.
Short answer: no — and it fails differently from everything else we have scored. Against a forward drawdown label on bitcoin, gold lands at an AUC of 0.537 over 2,280 days — where 0.50 = coin flip and 1.00 = perfect. Pooled, that is the fifth-best score of the 81 series we screen. Split by year, seven of ten years sit below the flip, and the ones above are not scattered: the relationship changes sign, and it has done so twice. In 2019 and 2021, rising gold preceded crypto trouble. In 2020 and 2023–24, falling gold did. Each story, in its years, looked real. Averaged together they cancel — and the pooled number, as we will show, belongs to neither of them.
The claim being tested
"Bitcoin is digital gold" is one of the founding stories of crypto: a scarce, non-yielding store of value, gold 2.0. If that is true, the two assets share drivers, and gold weakening — the store-of-value bid fading — should be a warning for crypto too.
But there is a second story, held just as widely, often by the same people on a different day: gold is the hedge — the safe place money runs when risk assets crack, crypto very much included. If that is true, gold strengthening is the warning.
Notice what just happened: the two mainstream readings of the same indicator predict opposite signs. Our own macro backdrop is honest about this — it labels gold's link to crypto as mixed and narrative-dependent rather than picking a story. This post is us measuring that label.
How we scored it
Same method as the Fear & Greed, VIX and dollar posts, so all four are directly comparable:
- The label. A 10-day forward drawdown on bitcoin in the worst 30% of historical 10-day windows. A research screen, not the number the cockpit shows — our headline drawdown read is a much tighter event (a drop from today's price, at its worst point, within 1 day).
- The base rate. 30% of windows qualify by construction, so a useless predictor scores 0.50.
- The scoring. Combinatorial purged cross-validation, embargo equal to the horizon, out-of-sample throughout.
- The sample. 2,280 trading days, August 2017 to September 2026 — the same overlap the dollar post had.
- The direction. A mixed indicator forces a choice, so we scored the digital-gold reading — falling gold counts as the warning — matching how our panel classifies the series (higher gold = calmer). The hedge reading is the exact mirror image: its AUC is 1 − 0.537 = 0.463. Everything below holds either way; only the labels swap.
The result
Pooled: AUC 0.537. Precision-recall 0.318 against a 30% base rate. On its own that would be one of the better readings we have published — which is exactly why the next list matters.
By year (above 0.50 = the digital-gold reading worked; below = the hedge reading did):
- 2017 (from August) — 0.486
- 2018 — 0.386
- 2019 — 0.342
- 2020 — 0.614
- 2021 — 0.329
- 2022 — 0.363
- 2023 — 0.610
- 2024 — 0.591
- 2025 — 0.451
- 2026 (to 13 September) — 0.366
Mean of the years: 0.454. Best: 0.614. Worst: 0.329 — a spread of 0.285 that straddles the coin flip on both sides. Seven years below, three above, and the pooled figure sits above every one of the seven. That 2021 reading is the lowest full-year score for any indicator in this series: further below a coin flip than the Fear & Greed Index's best year (0.640) sits above it.
The fourth failure shape
Four posts in, every indicator has failed in its own way:
- Fear & Greed — a coin flip whose one strong year is the latest one.
- The VIX — a coin flip pooled and by year; worst in the year it co-moved with crypto most visibly.
- The dollar — switches on and off, and nothing on the chart tells you which state it is in.
- Gold — the relationship itself flips sign, tracking whichever story the market is telling about it.
Follow the sequence. In 2018 and 2019 gold behaved like the hedge: it was strength, not weakness, that preceded bitcoin's rough stretches, and the digital-gold reading scored 0.386 and then 0.342. In 2020 the two assets crashed together in March and recovered together, and the same reading jumped to 0.614. 2021 flipped it straight back — 0.329, the worst reading in the series — and 2022 stayed there. Then 2023 and 2024, at 0.610 and 0.591, as gold and bitcoin increasingly traded as one store-of-value bid: two consecutive above-water years, which is the same recency lure the dollar sets, and if you screened this series at the end of 2024, "gold works now" looked like a finding. 2025: 0.451. 2026 so far: 0.366. The hedge is back.
An indicator with a stable sign and no skill is easy to retire. An indicator whose sign rotates is worse than that, because every year confirms somebody. The digital-gold believers point to 2020 and 2023–24; the hedge believers point to 2019, 2021 and now; both walk away validated, and the pooled record belongs to no one's thread.
Where the pooled 0.537 comes from
It is worth being precise about this, because 0.537 is the kind of number that gets quoted. AUC is a pairwise question — one risky day against one calm day, ranked correctly or not — and pooled over ten years, most of those pairs are cross-year pairs. Gold has more than tripled over this sample. So a calm day from a high-gold year and a rough day from a low-gold year get ranked "correctly" by the digital-gold reading whether or not gold said anything on either day. The pooled figure is largely gold's decade-long uptrend being paired against which years were rough for bitcoin. Inside any one year, where that trend is small, the same reading scores below the flip seven times out of ten. This is the cross-year pairing effect the VIX post describes, and gold is now our clearest example of it.
Why the sign flips
We are speculating past the data here and will say so plainly, but the shape is familiar from the dollar post. Gold is not one signal: it moves on real rates, on central-bank buying, on haven demand, on the dollar it is priced in. And the crypto side of the relationship is not a mechanism at all — it is a narrative, and narratives rotate. When the market trades bitcoin as digital gold, the two share a bid and gold weakness genuinely does lead crypto risk. When the market trades crypto as the risk asset gold hedges, gold rallies after crypto cracks — a lagging symptom of the same event, with the opposite sign. Average across regimes and you learn about neither.
As with the dollar, the honest fix is not a better threshold on gold — it is conditioning, knowing which regime you are in before you read the gauge, which is what a model that also sees rates, credit and cross-asset volatility can attempt and a single chart line cannot. This is the recurring argument: co-movement is not prediction, and an ambiguous co-movement is even less.
What this does not prove
- Not "gold is irrelevant to crypto." It earns its place in the macro backdrop as context. It does not, on this label and horizon, reliably tell you the next ten days are unusually dangerous — in either direction.
- Not "the hedge reading wins." Seven years below 0.50 is the most one-sided count in this series, and it is tempting to call it. But the three years on the other side are the three strongest readings, you only know which story a year rewarded after the year has happened, and our cross-fold standard deviation of 0.036 says nothing about single years. The claim we stand behind is the rotation — a 0.285 spread that has crossed 0.50 twice — not the direction of any one year, and not the count.
- Not a verdict on the decade-scale thesis. "Bitcoin will behave like gold as a long-run store of value" is a claim about years and decades. Our label is a 10-day drawdown. This screen cannot reach that claim, for it or against it — we tested the short-horizon, tradeable version, which is the version people actually act on.
- A univariate screen. Gold alone. Inputs that are ambiguous on their own can still contribute inside a multivariate model that sees the context which resolves the ambiguity.
The rule this series keeps arriving at
Four widely-watched indicators, four different ways of looking better than they are, one defence: ask for the per-period record, not the headline. Gold adds a second question to the checklist: ask what the indicator is supposed to mean, and whether the people citing it agree. An indicator whose believers disagree on the sign can stay popular forever, because it is never wrong for everyone at once.
That is why w4rn publishes per-year skill next to every series, and why our own forecasts are cross-validated and Platt-calibrated so a stated 30% is meant to be a real 30% — a target we check against what actually happened, not a promise. Some of our numbers are modest. Showing them is the point — that is what calibration means.
Method note: figures from our univariate forward-AUC screen, combinatorial purged cross-validation, panel dated 13 September 2026, 2,280 observations for gold, 81 scored series. Offline evidence, not a live trading claim.
Revision note, 14 September 2026: this post originally scored gold against SOL over 2020–2026 (0.511 pooled over 1,484 days; one sign flip, 0.315 in 2021 against 0.585 in 2023). It now scores bitcoin over 2017–2026, the asset most readers mean by "crypto", with three more years of history. The longer record shows the same flip — and a second one before it.
Revision note, 24 September 2026: an earlier version of the paragraph on calibration said that when our forecasts state 30%, those events happen about 30% of the time. That is what calibration aims for, and we had not yet published a check of it against the live record, so the line now states the target rather than a result.
See today's live read in the cockpit → Nothing here is financial advice.