Andrew Bailey said the rise in the UK curve reflects risk premia, not rate-hike expectations. The sentence traveled fast, got summarized as a "stable policy signal," and was filed under macro. Almost nobody asked which number tells us whether he is right.
There is one. It sits in the spread between the 30-year gilt yield and the 30-year overnight index swap rate. Gilts carry supply and credit risk. OIS carries little beyond bank balance-sheet risk. Subtract one from the other and the residual is not a model output. It is a traded quote for the compensation investors demand to hold UK duration. If Bailey's attribution holds, that spread widened while the curve steepened. If the move was a repricing of the policy path, the spread stayed flat while the short end and the SONIA strip moved together.
That is a falsifiable test, published daily, free of charge. It is also missing from the coverage.
When code speaks, we listen for the discrepancies. Here the code is a two-line subtraction, and the discrepancy is the gap between what a central banker attributed and what the market actually quoted.
Term premium is the compensation an investor demands for holding a long bond instead of rolling a short one. In the standard decomposition, a 30-year yield is the expected path of the policy rate plus that compensation. The first term belongs to the central bank. The second belongs to the market. They have different half-lives. A policy-path move mean-reverts when the cycle turns. A premium move does not. It reprices the discount rate for everything with duration — growth equities, property, infrastructure, and bitcoin.
Bailey's remark, then, is not a throwaway. It is a claim about which of the two terms moved, and it carries three functions at once. It tells investors not to extrapolate the long end into a hiking cycle. It preserves the Bank's autonomy over the policy path, because a central bank that accepts the market's steepening as a hike signal has handed the decision away. And it points at the fiscal authority: if the compensation is a risk premium, someone other than the Monetary Policy Committee is being asked to explain it.
The coverage took the first function and stopped. Summaries framed the statement as an implication of a stable policy stance. That reading inverts the arithmetic. A rising risk premium is the market demanding more to lend, not less. It is stability's invoice, not stability.
Attribution is not measurement. It is positioning with a vocabulary. Measurement needs an observable that quotes the premium directly, without a structural model.
Two exist, and both are cheap. The first is the gilt–OIS spread, preferably at the long end, where duration risk dominates and collateral technicals are at least visible. The second is the slope, built as 2s30s with the short leg anchored to SONIA forwards rather than spot yields, so the policy path is priced from swaps and the residual is duration.
I run both. The skeleton below is the version I keep on the desk, pared down to something a reader can reproduce in a notebook.
import pandas as pd, numpy as npdf = pd.read_csv('gilt_ois.csv', parse_dates=['date']).set_index('date') # columns: g2, g30 (gilts), ois2, ois30 (SONIA OIS), be30 (30y breakeven)
df['slope'] = df.g30 - df.g2 # gilt–OIS spread quotes the compensation directly # path captures the priced policy path # resid is whatever the policy path cannot explain df['tp_proxy'] = df.g30 - df.ois30 df['path'] = df.ois2.diff(21) df['resid'] = df.g30.diff(21) - 0.85 * df.ois2.diff(21)
z = lambda s: (s - s.rolling(252).mean()) / s.rolling(252).std() df[['tp_z','path_z','resid_z']] = df[['tp_proxy','path','resid']].apply(z) print(df[['tp_z','path_z','resid_z']].tail(5).round(2)) ```
The beta of 0.85 is the point. It encodes the regularity that long yields move less than one-for-one with the priced policy path, so any excess in the long end is, by construction, non-policy. Replace the fixed coefficient with a rolling regression and the residual becomes a detector for the moment the assumption breaks, which is precisely what September 2022 was. The spread is not pure term premium either. It carries swap-spread technicals: dealer balance-sheet costs, pension collateral demand, index-linked distortions. In 2022 those technicals were the story. Treat the number as a quote, not as truth.
A term premium is not an abstraction. It is the price of finding a buyer, and the UK's long end lost its price-insensitive buyer. Defined-benefit schemes that once absorbed long gilts and long index-linked paper for liability matching ran leverage, then unwound it in a matter of days. The demand that remains is price-sensitive, and price-sensitive demand demands compensation.
Set that against supply. The Debt Management Office keeps long-dated and index-linked issuance heavy relative to peers. Quantitative tightening removes duration from the market even as the Bank signals a slower pace of sales. Both facts can be true and still leave a net duration oversupply sitting with marginal buyers, who price it accordingly.
This is the same mechanism I mapped in 2024 on the Bitcoin side, running in reverse. Then, ETF custody pulled coins off exchanges and shrank float; a price-insensitive wrapper absorbed supply, and the structural squeeze surfaced in the illiquidity premium rather than the spot print. In gilts, the price-insensitive buyer left instead of arriving, so the identical mechanism runs backwards and surfaces as term premium. Mechanics, not mood. It is payment for standing in a market whose natural absorber has gone.
September 2022 is the calibration event. The 30-year gilt repriced more than a full percentage point in under two weeks. The Bank did not raise rates by anything close. It bought long duration on financial-stability grounds and sold it back within months. That episode established two facts. A premium-driven selloff in gilts is fast, convex, and self-reinforcing through collateral channels. And there is no policy put on it. The intervention restored market function; it did not set a yield ceiling.
Compare a hike-driven selloff. In a policy repricing, the market knows a terminal rate exists and the cycle ends. Duration can be held through it. In a premium repricing the terminal rate is irrelevant. Compensation is being repriced against an imbalance in the buyer base, and nothing in the policy toolkit addresses that directly. That asymmetry is the entire content of Bailey's statement, and it is why the word "stable" should be struck from the summary.
Why a crypto desk reads this at all. The marginal bitcoin buyer is now a macro allocator with a duration model. Bitcoin behaves like a long-duration, zero-cash-flow asset whose discount rate is real yields, and the relevant real yield is the long end, not Bank Rate. If a meaningful part of that yield is term premium rather than expected policy, the discount rate does not fall when the Bank cuts. The cut removes the path component. The premium stays.
| Regime | Observable signature | Half-life | Crypto read | |---|---|---|---| | Policy repricing | 2y gilt and SONIA strip move together; gilt–OIS flat | One cycle | Cuts restore the discount rate; buy the pause | | Premium repricing | Gilt–OIS widens; short end anchored | Years | Cuts do not restore it; duration stays expensive |
The second channel is the currency. A premium-driven gilt selloff weakens sterling without the rate support a hike-driven selloff would provide, and a weaker pound alongside rising UK risk premia is a global risk-off input rather than a domestic story. But this is where forensic discipline matters. That is a mechanism, not a correlation. If one-month sterling implied vol is not pricing the same event, the channel is not transmitting, and any bitcoin weakness is being caused by something else.
Attribution has a blind spot, and so does the audience reading it.
Bailey's claim cannot be falsified in real time, and a claim that cannot be falsified is a communication instrument rather than an empirical statement. Term premium estimates are model-dependent by construction. The Adrian–Crump–Moench decomposition, the Kim–Wright affine model, and a plain gilt–OIS subtraction will disagree on the level and occasionally on the sign of the change. So when a governor invokes "risk premium," he is not reporting a measurement he has corrected. He is selecting a narrative that happens to defend his institution's autonomy and shift explanatory burden toward the fiscal side. I spent six weeks in 2017 reverse-engineering testnet contracts because a whitepaper is a claim and bytecode is a fact. An attribution from a podium sits closer to the whitepaper end of that spectrum.
The audience's blind spot is the mirror image. Crypto traders have spent two years treating every gilt and JGB headline as a directional signal for bitcoin, which is a correlation trade wearing a mechanism's clothes. The mechanism is narrow: real term premium to discount rate to long-duration assets, with the dollar as the transmission belt. If the dollar and real yields do not confirm, the trade is noise with a macro vocabulary.
None of this argues against holding duration. It argues against holding it for the wrong reason. The residual is where the story lives, and the story it tells here concerns the price of a buyer, not the path of a policy rate.
The next data point is not a speech. It is the spread.
Watch three things over the coming weeks. The 30-year gilt–OIS spread with the 2s30s slope behind it: a widening spread with a stable short end confirms Bailey's attribution and prices a persistent premium. The short end following the long: SONIA forwards repricing higher would falsify him and turn the curve into a hiking story after all. And one-month sterling implied vol: if it stays anchored while the spread widens, the market is treating this as a technical, and transmission to risk assets is not yet live.
If the spread widens and volatility does not, rate-sensitive UK sectors and every long-duration asset priced in dollars have a problem no cut will fix. That is a stronger claim than the one the coverage made, and it is the only one the data supports.