The Fed cut rates by 100 bps in late 2024. Mortgage rates rose anyway.

The mortgage rate paradox

Fed funds, 10-year Treasury yield, and 30-year fixed mortgage rate, Jan 2020 – May 2026

During the Sept-Dec 2024 cutting cycle: Fed funds 5.33% → 4.33% (−100 bps), 10-Year Treasury 3.63% → 4.79% (+116 bps), 30-Year mortgage 6.08% → 7.04% (+96 bps).

Sources: FRED, DFF (daily Fed funds effective), DGS10 (10-year Treasury constant maturity), MORTGAGE30US (Freddie Mac PMMS 30-year fixed). Data as of 15 May 2026 (daily) / 14 May 2026 (weekly). Related research: how fixed-rate and adjustable-rate compare.

Standard intuition says that when the Fed cuts, borrowing costs fall. Mortgage rates should follow, eventually. This was the assumption pricing models, policymakers and consumer expectations carried into the fall of 2024.

On 18 September 2024, the Federal Open Market Committee delivered its first cut of the cycle: 50 basis points, bringing the target range to 4.75–5.00%. Two further cuts followed: 25 bps on 7 November and 25 bps on 18 December. The cumulative move was exactly 100 bps. The effective federal funds rate, per FRED data (series DFF), went from 5.33% on 17 September 2024 to 4.33% on 19 December.

The textbook prediction was clear: rates would fall, long-duration borrowing would loosen, refinancing would resume. Articles in the financial press anticipated 30-year mortgage rates returning toward 5%. A companion question: our breakdown of duration.

According to Freddie Mac’s Primary Mortgage Market Survey (FRED series MORTGAGE30US, weekly observations), the 30-year fixed mortgage rate stood at 6.08% the week ending 26 September 2024. By mid-January 2025 it read 7.04%. The move during the Fed’s cutting cycle was +96 bps, in the opposite direction.

The 10-year Treasury yield, which structurally drives long-term mortgage rates, moved in lockstep: from 3.63% on 16 September 2024 (the low before the first cut) to 4.79% on 13 January 2025 (the local peak), a 116 bps increase, again opposite the Fed.

• The Fed cut. Mortgages rose. The intuition was wrong.
• The 10-year Treasury yield moved opposite the Fed target, despite the Fed setting short rates.
• The cause sits in a component of the 10-year yield that does not appear on any consumer-facing screen.

What’s actually in a 10-year Treasury yield

A 30-year fixed mortgage rate is not set by the Federal Reserve. It is a long-duration borrowing rate determined in capital markets, priced off two main components: the 10-year US Treasury yield, and a spread that reflects credit risk, prepayment risk, origination costs and market liquidity.

Over the period since April 1971 (when Freddie Mac began publishing the PMMS; series MORTGAGE30US via FRED, less the DGS10 yield), the average spread between the 30-year fixed mortgage rate and the 10-year Treasury yield has been 1.76 percentage points, with a median of 1.66 pp. As of 14 May 2026, the spread stood at 1.89 pp: slightly above its long-run average but well within historical norms.

Most of the variation in mortgage rates traces to variation in the 10-year Treasury yield. The Fed funds rate enters the picture only indirectly, through its influence on the 10-year. To understand mortgage moves, you need to understand the 10-year. To understand the 10-year, you need to look at what it actually is.

The two components

Buying a 10-year Treasury today commits an investor to lending money to the US government for ten years at a fixed rate. The yield that compensates them for doing so must reflect two things.

First, what investors expect short-term interest rates to do over the next ten years, on average. This is the “expected path” component. If the market expects the Fed to hold short rates near 4% for the next decade, the 10-year yield should sit near 4%, all else equal. This component tracks Fed expectations.

Second, the compensation investors demand for the risk and inconvenience of locking up money for ten years instead of rolling short-term paper. This is the term premium. It reflects uncertainty about future inflation, future policy, future fiscal trajectories: anything that might make holding long-duration paper less attractive than expected returns alone would justify.

The two components can move independently. The Fed controls the short end and, through forward guidance, can shape the expected path. The term premium is a market-determined compensation for bearing duration risk over a horizon longer than any central bank can credibly steer.

This decomposition is not theoretical. The New York Fed publishes a daily estimate of the term premium and expected path components for Treasury yields at maturities from one to ten years, derived from a five-factor no-arbitrage model developed by Tobias Adrian, Richard Crump and Emanuel Moench (NY Fed Staff Report no. 340, 2008). The series begins in June 1961 (779 monthly observations) and is updated daily at newyorkfed.org/research/data_indicators/term-premia-tabs.

As of 30 April 2026, the ACM model decomposes the 10-year Treasury yield of 4.45% into 3.72% expected path and 0.73% term premium. The 0.73% is the focus of what follows.

The decomposition, over 65 years

Plotted across nearly 65 years of monthly data, the decomposition reveals something that does not appear in any individual snapshot: most of the variation in the 10-year yield over time has come from the expected path component (the Fed-determined part), but the term premium has gone through distinct, sustained regimes, and those regimes matter for any forward-looking discussion of long-duration rates.

A 65-year decomposition of the 10-year Treasury yield

ACM model: 10-year yield = expected path of short rates + term premium. Monthly observations, June 1961 – April 2026.

Key Takeaway

The 1979–1985 spike was driven by both components. The Volcker Fed pushed the short rate to historic highs to break inflation; the term premium also surged to record levels as investors required compensation for holding paper denominated in a currency whose purchasing power was eroding by double digits. The all-time peak of the ACM 10-year term premium was 5.15% in May 1984, mid-Volcker. The descent from those highs took roughly two decades. The term premium turned negative for the first time in the recorded series in 2014, and stayed negative or near zero for most of the decade that followed. The 2024 episode is the regime exit.

Source: Federal Reserve Bank of New York, Adrian, Crump, Moench (ACM) term structure model. Series: ACMY10, ACMTP10, ACMRNY10. N=779 monthly observations.

↓ PNG Data: Federal Reserve Bank of New York (ACM model), under its terms. Cite this dataset

Six monetary regimes

Splitting the 65-year series into the canonical monetary regimes makes the regime-shift nature of the term premium more visible.

10-year US Treasury term premium with regime overlay, 1961–2026

NY Fed ACM model. The “risk premium” component of the 10-year yield, across six monetary regimes.

Source: NY Fed ACM term structure model, series ACMTP10. Monthly end-of-period values, June 1961 – April 2026. N=779.

↓ PNG Data: Federal Reserve Bank of New York (ACM model), under its terms. Cite this dataset

Regime-by-regime statistics

Each regime had a distinct mechanism

The 2024 episode, decomposed

The decomposition allows the 2024 episode to be read at higher resolution. Between end-September 2024 and end-January 2025, ACM monthly snapshots show two distinct moves.

The expected path component of the 10-year yield rose by 21 bps (from 3.89% to 4.10%). This is consistent with the market initially expecting more cuts, then re-pricing the path higher as US economic data came in stronger than anticipated and inflation concerns resurfaced.

The term premium component reversed regime. From a still-negative −0.10% at end-September 2024 to +0.49% by end-January 2025: a 60 bps move in four months at the underlying NY Fed precision. This was not a small adjustment within an existing regime. It was the term premium crossing zero from below, exiting the decade-long ZIRP regime and entering normalization. An exit of that kind is one of the hinges on which the history of monetary regimes since 1990 turns.

The total 10-year yield move during the cutting cycle reflected both. The Fed cut. The market re-priced the path slightly higher. And the term premium switched signs.

Explore the data

Hover any month to see the full breakdown into expected path + term premium. Filter by monetary regime to compute mean, median, min, max for the selected window.

Scenario simulator

This section is descriptive, not prescriptive. The exercise is mechanical: holding the latest expected-path component constant at 3.85% (ACM, 15 May 2026), what 10-year Treasury yield would various assumptions about term premium imply? Given the long-run 1.76 pp average spread to mortgage rates, what 30-year fixed mortgage rate would each scenario mechanically imply?

Four anchors worth marking

Robustness: ACM vs Kim-Wright

Term premium estimates are model-derived. Different models can yield different point estimates. The two most-cited estimates in academic and central-bank work are the NY Fed’s ACM model (used throughout this study) and the Federal Reserve Board’s Kim-Wright (KW) model, available on FRED as series THREEFYTP10 since January 1990.

Over the period of overlap (1990–2026, 9,074 daily observations), the two models correlate at 0.86. ACM has averaged 0.26 percentage points higher than KW. As of the most recent KW observation (8 May 2026), the two are within 3 bps of each other, with KW slightly above. The two leading models broadly agree on the current level of the term premium, and both place it solidly above the negative ZIRP-era values.

For the directional reading of this study, term premium has exited the negative regime and partially normalized, the conclusion is robust across both decompositions.

Historical turning points

May 1984: The all-time term premium peak

The ACM 10-year term premium reached +5.15% in May 1984: the highest value in the 65-year recorded series. The Federal Reserve under Paul Volcker had pushed the federal funds rate above 19% to break double-digit inflation. Published in the Federal Reserve’s history of FOMC decisions. The peak term premium reflected investor compensation for holding paper denominated in a currency whose purchasing power was eroding rapidly.

January 2014: Entry into the negative regime

For the first time in the recorded ACM series (since June 1961), the 10-year term premium turned negative on a sustained basis. The trigger: a combination of post-GFC central bank balance sheet expansion (the Fed’s balance sheet had grown from approximately $0.9T pre-crisis to approximately $4T by 2014), regulatory shifts favoring sovereign holdings, weak global demand, and a strong foreign appetite for US safe assets in a world of compressed yields. The negative-regime decade that followed had no historical precedent in the series. Balance-sheet expansion and the scramble for safe assets are the very mechanics of liquidity and financial conditions, from monetary plumbing to QT cycles.

July 2020: The COVID floor

At the depth of the pandemic, the ACM 10-year term premium hit −1.34%, the historical low of the entire 65-year series. The Federal Reserve’s balance sheet response (LSAPs at unprecedented scale), combined with massive flight-to-safety demand for US Treasuries, drove the price of duration to extremes. For a brief period, holding 10-year US Treasuries carried a meaningfully negative risk premium: investors were effectively paying for the privilege of duration, presumably because alternative safe assets were less attractive or because central bank balance sheet operations had distorted price formation.

September 2024 – January 2025: The regime exit

During the Fed’s 100 bps cutting cycle, the term premium went from −0.10% (end-September 2024) to +0.49% (end-January 2025): a 60 bps move in four months, crossing zero from below for the first time after a decade in negative or near-zero territory. Over the same window, the 10-year Treasury yield rose 116 bps (3.63% → 4.79%) and the 30-year mortgage rose 96 bps (6.08% → 7.04%). The Fed cut. Long-duration borrowing rose. The term premium switched signs.

April 2026: Current observation

The ACM 10-year term premium stands at +0.73% in April monthly data (+0.83% in the daily snapshot of 15 May 2026), at the 35th percentile of the full 65-year distribution. Above the negative ZIRP-era values, but well below the 65-year median of 1.41% and the Great Moderation average of 1.65%. The “normalization” of the term premium has occurred, but it has only partially reversed. Whether further normalization toward historical means continues depends on factors largely outside Fed control: Treasury issuance trajectories, foreign demand mix, fiscal trajectories, inflation uncertainty.

How the Fed balance sheet reaches the term premium

The policy rate works mainly through the expected-path component. The residual moves when the quantity of duration held by the market changes, and only the balance sheet changes it. Purchases take duration out. Runoff hands it back.

Three channels carry the effect, and they compound. The portfolio channel acts on the physical stock of duration, as sellers of long Treasuries move the cash into credit, mortgage bonds or equities. Operation Twist isolates it cleanly: the Fed sold short Treasuries and bought long ones without changing the size of its balance sheet, so no cash was injected and no new signal on short rates was sent. The signaling channel acts on expected short rates instead, since purchases read as a commitment to keep the policy rate low for longer, and the ACM model can book part of that effect as a lower term premium. The preferred-habitat channel rests on maturity segmentation. When purchases concentrate on one segment, the investors anchored there accept lower compensation or move to neighbouring maturities.

Stock and flow both matter. The taper tantrum, set off when the Fed chair first raised the prospect of slowing purchases, points to a dominant flow effect in the short run. The low premium that persisted later, while the balance sheet stayed flat, points to a stock effect over the medium term.

Runoff does not simply rewind purchases. Purchases usually arrive in a deflationary or recessionary shock, when safe-asset demand for Treasuries is strong; runoff comes in an expansion, when that demand is loosening. The Fed also presents runoff as a background tool, which dilutes any signal, and banks holding duration for liquidity coverage requirements absorb part of the returned supply.

Three forces cited for the return to a positive premium

First comes the runoff described above. Quantitative tightening (QT) began on 1 June 2022 with monthly caps of $30 billion in Treasuries and $17.5 billion in agency MBS, raised to $60 billion and $35 billion in September 2022. The Treasury cap fell to $25 billion in June 2024, then to $5 billion in April 2025, and the runoff ended on 1 December 2025. The Fed balance sheet went from $8.91 trillion at end-June 2022 to $6.54 trillion on 3 December 2025 (FRED WALCL), about $2.4 trillion of securities handed back to the market.

Net Treasury supply is the second force. In its January 2025 projections, built on current law, the Congressional Budget Office put the total federal deficit at 6.2% of GDP in 2025, 5.2% in 2027 and 6.1% in 2035, projections that leave out budget measures adopted later. Auctions absorbed the extra duration in an orderly way: the 10-year bid-to-cover ratio averaged 2.54 in both 2024 and 2025 (Treasury Department).

The third force, the debate over r-star, the long-term real equilibrium interest rate, runs through another component. A higher r-star implies a higher neutral policy rate and higher expected short rates, so the ACM model books it in the expected path. It reaches the term premium only if uncertainty about r-star raises the compensation required to carry duration.

At the 1 November 2023 press conference, Jerome Powell said QT “is certainly on that list” of candidate explanations for higher long rates, while judging that “it may be playing a relatively small effect.”

The literature divides over the weight and durability of these forces. A structural reading (Charles Goodhart and Manoj Pradhan, The Great Demographic Reversal, 2020) ties higher rates to population aging, as retirees dissave and global net savings fall. A cyclical reading treats the runoff, the deficits and r-star as reversible, which leaves room for the premium to narrow as they fade.

What the term premium is not

Three misreadings recur when the term premium is discussed. The first equates it with the slope of the curve, the gap between the 10-year and the 2-year yield. That slope blends short-rate expectations at both horizons with the term premium, so a steep curve can coexist with a negative premium and an inverted curve with a positive one.

The second confuses it with the real yield on inflation-indexed Treasuries. The two decompositions work at different layers: one separates expected short rates from duration compensation, the other separates the real yield from expected inflation. Read together, they show which component moved when the 10-year yield moves, and therefore which mechanism was repriced.

The third treats any rise in the premium as a loss of confidence in US debt. It can also reflect a return to normal after an anomaly, a structural shift in demand for duration, or a re-estimation of the model itself. The estimate also includes a convexity adjustment that is not, strictly speaking, compensation for risk.

The decomposition that separates the real yield from expected inflation has its own series: DGS10 ≈ DFII10 + T10YIE, the real yield on 10-year inflation-indexed Treasuries (TIPS) plus the 10-year breakeven, which FRED computes as the difference between the two yields. The two decompositions do not add up. ACMTP10 measures the total nominal term premium and DFII10 contains a real term premium, so a sum of the expected path, ACMTP10, DFII10 and T10YIE has no economic meaning.

What bends under stress is the label of expected inflation on the breakeven. TIPS are less liquid than nominal Treasuries, and that liquidity premium lifts the real yield and lowers the breakeven, while the inflation risk premium demanded by holders of nominal Treasuries pushes the other way. D’Amico, Kim and Wei (2018) find that the liquidity premium dominates. On 19 March 2020, the 10-year breakeven fell to 0.50%, against a February average of 1.62% (FRED T10YIE).

A positive premium does not transmit evenly. In a discounting model it weighs most on assets whose cash flows lie far out, such as growth equities, listed real estate and long investment-grade credit. The all-else-equal condition is rarely met, so the effect reads more clearly over several quarters than across sessions. Mortgage-backed securities respond asymmetrically, as higher rates slow prepayments and lengthen their duration. Defined-benefit pension funds, whose liabilities are discounted at long rates, see their funding improve as yields rise, which can reduce their appetite for additional duration.

Methodology

The 10-year Treasury yield decomposition follows Adrian, Crump and Moench (2008), Pricing the Term Structure with Linear Regressions, Federal Reserve Bank of New York Staff Report no. 340. The model decomposes Treasury yields at maturities from 1 to 10 years into the expected path of short-term rates (under the risk-neutral measure) and a term premium, using a five-factor no-arbitrage term structure framework. The series is updated daily by the NY Fed Research division.

Data sources

• NY Fed ACM Term Premium 10-year (series ACMTP10, ACMY10, ACMRNY10): 779 monthly observations June 1961 – April 2026, daily through 15 May 2026.
• FRED DGS10: 10-year Treasury constant maturity, daily.
• FRED MORTGAGE30US: Freddie Mac PMMS 30-year fixed mortgage rate, weekly April 1971 – May 2026.
• FRED DFF: Effective federal funds rate, daily.
• FRED THREEFYTP10: Kim-Wright 10-year term premium, daily January 1990 – May 2026, used for triangulation.

Dataset variables

Python reproduction code

import pandas as pd

# Fetch ACM monthly from NY Fed (xls)
url = "https://www.newyorkfed.org/medialibrary/media/research/data_indicators/ACMTermPremium.xls"
acm = pd.read_excel(url, sheet_name='ACM Monthly')
acm = acm[['DATE', 'ACMY10', 'ACMTP10', 'ACMRNY10']]
acm.columns = ['date', 'y10', 'tp', 'rn']
acm['date'] = pd.to_datetime(acm['date'], format='%d-%b-%Y')

# Verify the 2024 regime exit on monthly snapshots
sep24 = acm[acm.date == '2024-09-30'].iloc[0]
jan25 = acm[(acm.date >= '2025-01-01') & (acm.date <= '2025-01-31')].iloc[0]
print(f"TP move: {(jan25.tp - sep24.tp)*100:.0f} bps")  # 60
print(f"RN move: {(jan25.rn - sep24.rn)*100:.0f} bps")  # 21

Dataset Download & Reproducibility

The complete monthly ACM decomposition is provided in open format. Updated when NY Fed publishes new monthly observations.

Data terms: © Federal Reserve Bank of New York. Content from the New York Fed subject to the Terms of Use at newyorkfed.org (attribution kept, modifications labelled, same terms for any onward distribution); not under a Creative Commons licence. Eco3min text and analysis: CC BY 4.0.

Data Sources & References

• Primary Federal Reserve Bank of New York: Adrian, Crump, Moench (ACM) term structure model. Series ACMTP10, ACMY10, ACMRNY10. Monthly and daily, 1961–present.
• Primary FRED: DGS10: 10-Year Treasury Constant Maturity Rate (daily, 1962–present).
• Primary FRED: MORTGAGE30US: 30-Year Fixed Rate Mortgage Average in the United States, Freddie Mac PMMS (weekly, April 1971–present).
• Primary FRED: DFF: Effective Federal Funds Rate (daily).
• Primary FRED: THREEFYTP10: Federal Reserve Board’s Kim-Wright 10-year term premium (daily, January 1990–present).
• Research Adrian, Crump, Moench (2008): “Pricing the Term Structure with Linear Regressions,” Federal Reserve Bank of New York Staff Report no. 340.
• Research Kim & Wright (2005): “An Arbitrage-Free Three-Factor Term Structure Model and the Recent Behavior of Long-Term Yields and Distant-Horizon Forward Rates,” Finance and Economics Discussion Series, Federal Reserve Board.

Methodological Limitations

• Model-derived estimates. The term premium is not directly observable. It is the residual between the observed yield and the model’s estimate of the expected short-rate path. Different models (ACM, KW, DKW, others) produce different point estimates. This study uses ACM throughout and triangulates with KW (correlation 0.86 over 1990–2026), but the residual model risk cannot be fully eliminated.
• Mixed-frequency decomposition. The headline “+116 bps” 10-year move uses daily extrema (Sept 16 low to Jan 13 peak). The decomposition into “+21 bps expected path + +60 bps term premium” uses ACM monthly snapshots (Sept 30 → Jan 31). These are different windows. The decomposition reports relative contributions to a comparable monthly move, not an exact arithmetic split of the daily headline number.
• The term premium is not constituent-decomposed. ACM separates the 10-year yield into expected path and term premium, but does not further attribute the term premium itself to its drivers (inflation uncertainty, real rate uncertainty, supply-demand, fiscal). A constituent decomposition requires separate empirical work.
• Scenarios are mechanical, not predictive. The simulator holds the expected-path component constant. In practice, expected path and term premium are correlated. The simulator answers “what 10-year yield would be implied if X were the term premium, holding all else constant”: a sensitivity exercise, not a forecast.
• Mortgage spread is not constant. The implied 30-year mortgage figures use the long-run mean spread of 1.76 pp. In practice this spread has ranged historically and currently sits at 1.89 pp. Mechanical implied mortgage figures will diverge from realized values if the spread shifts.
• Three monetary regimes, one dataset. The 65-year series spans late Bretton Woods, post-Bretton Woods fiat under successive frameworks, and the post-GFC balance-sheet era. Term premium drivers have differed structurally across these regimes. Cross-regime comparisons of point estimates should be made with this context.
• Past distributions are not predictive. Percentile rank statistics describe historical position; they do not predict future levels. The 35th-percentile reading places today within history but does not constrain where the term premium goes next.

Frequently Asked Questions

Cite this dataset

Data: © Federal Reserve Bank of New York, under its Terms of Use, not a Creative Commons licence. Eco3min text and analysis: CC BY 4.0.

No investment advice: informational analysis only. The decomposition presented is descriptive: it reports the model-implied components of the 10-year Treasury yield and their historical distribution. Mechanical scenarios in the simulator are illustrative computations, not forecasts. Past distributions are not predictive of future levels.

Related Eco3min Research

• US CPI Inflation History: 113 Years of Purchasing Power Destruction
• Yield Curve Inversions Dataset (T10Y2Y, 1962–Present)
• Federal Funds Rate History (1954–Present)
• US Real Interest Rates Dataset (1962–Present)
• US PCE Inflation Dataset
• Eco3min Macro Data Hub: Full Index