What Is the Neutral Interest Rate (r*)? Meaning, Estimates, and Limits

In a standard monetary-policy framework, r-star describes a hypothetical equilibrium: the real short-term interest rate that would be expected to prevail after temporary shocks have faded, when output is near its sustainable or potential level and inflation is stable. The New York Fed summarizes the Laubach-Williams and Holston-Laubach-Williams approach in similar terms. This is a model-based reference point, not an observable rate that can be read directly from a bond screen or a central-bank announcement. {source:nyFedNaturalRateRstar}

Economists often use “natural rate” and “neutral rate” for closely related ideas, but terminology and the precise horizon can differ across papers and institutions. Some estimates refer to the short-run equilibrium rate that changes with economic conditions; others emphasize a longer-run rate after cyclical forces have passed. When comparing two figures, first check what each study calls neutral and which equilibrium it estimates. Do not assume that the same symbol guarantees the same concept.

The benchmark is “real” because it adjusts for expected inflation. That makes it different from the nominal policy rate, a bank’s quoted loan rate, a Treasury yield, or a household’s realized borrowing cost. The separate guide to real versus nominal interest rates explains how the inflation adjustment works and why its horizon and measure matter.

The neutral benchmark is usually stated as a real rate. A nominal neutral policy rate can be derived only after making an assumption about inflation over a matching horizon. If long-run expected inflation is assumed to equal a central bank’s inflation objective, the exact one-period Fisher relationship is:

Nominal neutral rate = (1 + neutral real rate) × (1 + expected inflation) − 1

When rates are modest, people often use the approximation nominal neutral rate ≈ r* + expected inflation. The approximation is convenient, but the exact calculation includes compounding. The inflation assumption must refer to the same currency and compatible horizon as the real-rate estimate; an inflation target is not automatically the right input for every period or model. The Fisher relationship appears in the Federal Reserve’s discussion of how expected inflation enters natural-rate estimation. {source:fedNaturalRateExpectations}

This distinction prevents several common mix-ups. The observed nominal policy rate is a rate set or influenced by a central bank today. The ex-ante real policy rate is estimated by adjusting a nominal rate for expected inflation over its term. r-star is a separate, unobserved real equilibrium benchmark. A nominal “neutral rate” is a model-derived counterpart under an inflation assumption, not a second policy instrument.

An estimate of r-star can help economists describe whether a real policy rate is above or below an estimated neutral benchmark. A real rate above r-star may be described as restrictive relative to that benchmark; one below it may be described as accommodative. But that comparison is conditional on using compatible horizons, inflation expectations, and definitions. It is not a complete verdict on the policy stance: credit spreads, lending standards, asset prices, fiscal settings, and other forces can also affect financial conditions and demand.

The benchmark also appears in monetary-policy rules. In a common Taylor-rule expression, r-star plus inflation forms part of a baseline nominal policy rate, with additional terms for inflation relative to an objective and economic slack. That makes the estimate consequential, but it does not make one rule or one r-star value mandatory. See how the Taylor rule frames a policy-rate decision for the formula and its input choices.

Central banks can use neutral-rate estimates as one input to analysis and projections without targeting the estimate itself. For example, Bank of Canada staff describe the neutral rate as an input to projections and explicitly state that the Bank does not target it. The application is analytical: an estimate can organize a discussion about policy and the economy, but it cannot determine a rate decision by itself. {source:bocNeutralRateAssessment2026}

The economy never provides a clean observation of “the interest rate that would prevail if cyclical shocks had disappeared.” Researchers see actual output, inflation, market rates, and policy decisions while the economy is experiencing changing demand, supply, risk, and institutional conditions. The equilibrium benchmark is inferred from how those observed series move together under a model.

In the Laubach-Williams family of models, for example, researchers use real GDP, inflation, and a short-term policy rate to estimate unobserved trends such as potential growth and r-star. These methods can separate a slowly changing trend from temporary movements only by imposing statistical and economic assumptions. New observations, revised GDP data, or a different specification can therefore change the historical estimate as well as the current one. The New York Fed publishes both real-time estimates based on the data available then and updated model estimates. {source:nyFedNaturalRateRstar}

“Unobservable” does not mean arbitrary or useless. The models are disciplined by observed data and explicit theory, and they can help compare scenarios. It means the estimate is not directly measured like a posted policy rate, and the uncertainty should remain visible. A precise-looking decimal from a model is not proof that the true equilibrium rate is known to that precision.

There is no single accepted estimator. One family of semi-structural models, including Laubach-Williams and Holston-Laubach-Williams, combines an aggregate-demand relationship with an inflation relationship and extracts trend components from data such as output, inflation, and short rates. In the New York Fed’s presentation, the original LW estimates concern the United States; later HLW work also provides estimates for Canada and the euro area. The model’s geographic coverage does not turn its results into a universal rate. {source:nyFedNaturalRateRstar}

Structural models build the benchmark from choices and constraints that shape saving, investment, production, and demographics. Overlapping-generations models, for instance, can represent how age structure and household decisions affect desired saving and investment. Market-based or term-structure models instead infer long-run rate components from traded bond yields and related financial data. A bond yield also contains term, liquidity, and risk compensation, so a market-based estimate is not simply a direct quote of r-star.

These approaches answer related but not identical questions. A macroeconomic filter, a structural model, and a term-structure model use different data and assumptions; their estimates can differ even when they refer to the same country and broad horizon. Bank of Canada staff’s 2026 assessment illustrates a model suite that combines structural approaches with empirical and term-structure evidence. Its report describes the estimates as model-based staff assessments, not a direct reading from markets. {source:bocNeutralRateAssessment2026}

<!-- learn:illustration -->

The figure shows schematic routes from observed economic data through different model approaches to a range around an inferred benchmark. It is conceptual: no path or marker is a measured r-star estimate, and the range is not universal.

The neutral rate can change when the economy’s long-run balance between desired saving and investment changes. Trend productivity and potential output growth can influence expected returns on investment and the supply of saving. Demographics can matter through life expectancy, the balance between workers and retirees, and saving plans over a lifetime. These are structural or slow-moving forces, not necessarily reactions to the latest policy meeting.

Risk and the demand for safe assets can also matter. A stronger preference for safe assets or greater perceived risk may affect the return investors accept and the cost of financing investment. Research reviews also discuss income inequality, public debt, and other factors that can affect saving and investment. The direction and size are not mechanical: factors interact, can offset one another, and may enter differently across model designs and economies. {source:ecbNaturalRateOccasionalPaper217} {source:bocNeutralRateDiscussionPaper2024}

It is therefore risky to tell a one-factor story, such as “aging always lowers r-star” or “higher productivity always raises it.” A model may show a particular channel under stated assumptions, while another model gives different weights or captures international spillovers. Treat explanations of drivers as model-based mechanisms, not a formula that maps one demographic or fiscal statistic into a certain r-star change.

Estimates can differ because researchers choose different model structures, measures of inflation expectations, data vintages, time horizons, and definitions of the equilibrium state. A real-time estimate based only on data available to policymakers at the time may differ from a retrospective estimate using revised data. A single point estimate also hides uncertainty that a range or confidence interval can make more visible.

Inflation expectations are one concrete source of sensitivity. In a Federal Reserve study of U.S. data, replacing backward-looking proxies with survey-based inflation forecasts changed the level and path of estimated r-star; adding short-horizon survey information narrowed uncertainty bands in the specifications studied. This is evidence about that framework and sample, not a claim that surveys always produce a more accurate estimate in every country. {source:fedNaturalRateExpectations}

The Bank of Canada’s May 2026 staff assessment illustrates why a published range needs a label. It reported a nominal neutral-rate assessment of 2.25%–3.25% for Canada and 2.50%–3.50% for the United States, using several methods. Those numbers are country-specific, nominal, dated staff estimates; the paper says the ranges do not capture all uncertainty and that the authors’ views may differ from the Bank’s official position. They are not current rates for every economy or recommended policy targets. {source:bocNeutralRateAssessment2026}

An older ECB review likewise compares a range of euro-area estimates and reports substantial uncertainty across methods. It is useful for understanding why estimates vary, but its 2018 estimates should not be presented as a current euro-area reading. Keep the institution, date, method, horizon, real-or-nominal basis, and uncertainty range attached to any quoted number. {source:ecbNaturalRateOccasionalPaper217}

Start by asking whether a published value is real or nominal, short-run or long-run, and tied to which country or currency area. Then identify the method, data vintage, horizon, and uncertainty range. A number from a macro model, a staff assessment, a survey, or bond-market pricing may be useful for a different purpose; none should be substituted silently for another.

Consider a deliberately hypothetical calculation. Suppose a fictional economy has an estimated long-run neutral real rate of 1.00%, and suppose expected inflation over the same horizon equals an assumed objective of 2.00%. The exact nominal counterpart is (1.01 × 1.02) − 1 = 3.02%; the simple additive approximation is 3.00%. Both inputs are invented for teaching arithmetic. This is not an estimate for a real country, not a policy target, and not a claim that the current policy rate should equal 3.02%.

Even a correctly calculated nominal counterpart does not settle whether policy is restrictive. To compare a current real policy rate with r-star, use a compatible inflation expectation and horizon, then keep the model uncertainty in view. For the nearby concepts, see how the output gap compares actual and potential output, how monetary policy reaches financial conditions and prices, what TIPS breakeven inflation does and does not measure, and how CPI, PCE, and the GDP deflator differ. Those measures inform parts of the analysis but do not directly reveal r-star.