US Yield Curve: 10Y-2Y vs 10Y-3M Spread Accuracy

The Fed's Preferred Metric: Why the 10Y-3M Spread Holds Weight

As of mid-July 2026, the US yield curve has fully normalized after one of the most closely watched inversion episodes in modern monetary history. The 10Y-3M spread now sits at approximately +80 basis points, and the 10Y-2Y spread at roughly +40 basis points — both comfortably in positive territory after the prolonged inversion cycle that began in July 2022 and extended well into 2024. We can observe that this normalization carries immediate analytical weight for our forward baseline: it forces a reassessment of how both the Federal Reserve's preferred recession probability model and the broader practitioner toolkit should weight these two term spreads moving into the second half of the decade.

The Federal Reserve Bank of New York uses the 10-year minus 3-month (10Y-3M) Treasury yield spread as the primary term spread input for its official recession probability model. This is not an incidental choice, and it predates the more media-amplified 10Y-2Y spread that has captured retail attention for decades. Given the mandate of producing a probability estimate with a continuous monthly input and a clean interpretation of the term premium, the 10Y-3M spread offers the most defensible signal — the 3-month T-bill sits at the front end of the curve and reacts most directly to Federal Open Market Committee policy decisions, while the 10-year note embeds medium-term inflation expectations, real growth assumptions, and the term premium itself. The 2-year note, by contrast, occupies an awkward middle position: too long to capture pure policy expectations, too short to reflect the structural growth and inflation outlook that anchors the long end.

Consequently, when the Federal Reserve Bank of San Francisco published its 2018 reassessment of curve-based recession forecasting — the Bauer and Mertens study — it confirmed what institutional practitioners had long suspected: the 10Y-3M spread exhibits slightly higher predictive accuracy than its more familiar cousin. The Area Under the Curve metric, a standard measure of classifier performance, sits between 0.85 and 0.89 for the 10Y-3M spread across the historical sample, edging out the 10Y-2Y measure in side-by-side testing.

The 10Y-3M spread is the Fed's anchor: it captures pure policy at the front end and structural expectations at the long end, with the 2Y caught between regimes.

Comparative Accuracy: AUC Metrics and Historical Success Rates

Since 1980 — when the 2-year Treasury note was first auctioned in 1976 and a sufficient post-Volcker sample became available — both spreads have posted a 71% historical success rate in predicting US recessions. Extending the 10Y-3M sample back to 1957 lowers that figure to 64%, but this remains a robust hit rate for a single variable in macroeconomic forecasting. The fact that the longer 10Y-3M sample produces a lower success rate is itself instructive: it suggests that the predictive power of the curve strengthens during the modern inflation-targeting regime and weakens in earlier decades when monetary policy operated under different rules.

The takeaway for our forward-looking framework is straightforward: if we are constructing a recession probability overlay for a portfolio risk committee or a policy memo, the 10Y-3M spread should anchor the model, with the 10Y-2Y serving as a secondary confirmation signal. Given the mandate of producing robust probability estimates rather than headline-grabbing calls, this hierarchy reflects the underlying economic logic rather than market convention.

We would also note that the Federal Reserve's modeling apparatus increasingly incorporates near-term forward spreads — synthetic measures that strip out the term premium component — as auxiliary inputs. These refinements reinforce the structural preference for shorter front-end instruments over the 2-year note when the objective is to isolate the policy expectation channel from the broader risk pricing channel.

The 2022–2024 Anomaly: When the Curve Failed to Signal a Recession

The prolonged yield curve inversion that began in July 2022 did not trigger a US recession by mid-2024, even as historical averages — particularly the 11-to-12-month average lag — implied that contractionary pressures should have materialized by late 2023 at the latest. By any conventional reading of the term spread signal, a recession should have arrived. It did not. We can observe that this episode has been classified by many macro desks as a potential false positive, and the implications for our modeling framework are non-trivial.

Three structural factors likely attenuated the signal. First, the Federal Reserve's quantitative tightening program — running down the central system's balance sheet at sustained pace through 2023 and into 2024 — interacted with the term premium channel in ways the pre-2014 historical sample cannot fully capture. The mechanical transmission runs from duration absorption: by stepping back from the long end of the Treasury market, QT left private investors to absorb more of the long-duration paper the central bank was no longer absorbing. Under standard bond market mechanics, this additional supply of duration should put upward pressure on the term premium — and, through it, on long-end yields — pushing the curve toward steepening rather than toward deeper inversion. The empirical fact that long-end yields remained anchored and the curve stayed deeply inverted through most of 2023 implies that this countervailing supply pressure was more than offset by demand: foreign reserve managers continuing to recycle dollar proceeds into Treasuries, domestic pension and insurance accounts rebalancing into long-duration paper at attractive yields, and — most tellingly — a recession-hedge bid that expanded as the inversion deepened. QT did not compress the curve through a mechanical yield channel; if anything, its mechanical pressure should have steepened it. The structural distortion that mattered was on the demand side of the long end, not on the Fed's supply side — which means the 2022–2024 episode should be read as a curve that was inverting despite QT rather than because of it, complicating the historical mapping between inversion depth and recession indicator yield curve signals that prior literature codified.

Second, the unprecedented fiscal transfers and accumulated household balance sheets that followed the pandemic created a consumption buffer that materially delayed the transmission of monetary tightening into real activity. Third, the labor market remained structurally tight through 2023 and into 2024, with wage growth running above target-consistent levels — an outcome inconsistent with prior late-cycle dynamics and one that suggested the economy was operating along a different short-run aggregate supply curve than the historical record would imply.

An inversion is not a guarantee: it is a probability shift. The 2022–2024 episode is the reminder we needed.

This does not invalidate the yield curve as a recession indicator. It does, however, impose an obligation on practitioners to weight the signal against the structural backdrop rather than treat an inversion as a deterministic trigger. For our baseline, the operational lesson is that the curve's predictive power is conditional on the broader policy regime — and that regime has been anything but conventional since the post-pandemic balance sheet expansion began.

Lag Times and Market Signals: The 11-Month Average in Context

Historically, the average lag between the inversion of the 10Y-3M yield curve — defined strictly as a negative monthly average — and the onset of a US recession is approximately 11 to 12 months. The distribution is, however, wide: some inversions have been followed by recessions within 6 months, while others have waited the full 24-month upper bound before delivering contraction. We can observe that this dispersion is itself an analytical input — the lag length tends to correlate with the depth and persistence of the inversion, with shallower or shorter inversions producing shorter lags, and with the steepness of the pre-inversion yield trajectory shaping the eventual recovery path.

For our baseline, the practical implication is that the inversion signal functions better as a probabilistic regime shift than as a calendar instrument. Consequently, asset allocation and capital preservation decisions should not hinge on the precise month of inversion onset but on the probability-weighted path of recession risk over the subsequent 12 to 24 months. Given the mandate of preserving capital across the cycle, this probabilistic framing is the only intellectually honest way to deploy the signal in a portfolio context.

The transmission mechanism runs from the front end to bank lending, then to corporate borrowing costs, and finally to capex and hiring decisions. When the curve inverts, banks face a structural headwind — short-term funding costs exceed long-term lending yields, compressing net interest margins and tightening credit availability. This credit channel is the primary mechanism by which an inverted curve historically translates into real economic contraction. The 2022–2024 episode did not break this mechanism so much as delay it: banks tightened lending standards, but the magnitude of fiscal stimulus and balance sheet liquidity cushioned the downstream impact on corporate investment.

We would add one further nuance to the lag discussion. The 11-to-12-month average is itself a mean of a skewed distribution — it compresses what is, in practice, a bimodal pattern. Tight inversion cycles initiated under aggressive hiking regimes tend to deliver shorter lags, often in the 6-to-9-month range. Conversely, "soft" inversions driven by collapsing long-end yields — as opposed to front-end repricing — historically produce longer lags, sometimes stretching toward the 18-to-24-month upper bound. Reading the 2022–2024 episode through this lens suggests that the inversion was driven primarily by front-end repricing rather than by a long-end collapse, which should have produced a shorter lag than the historical average. The fact that no recession materialized at all is what makes the episode anomalous, not the length of the lag itself.

Current Market Landscape: Interpreting the 2026 Normalization

As of July 2026, both Treasury yield curve spreads are positive and the curve has returned to a configuration consistent with late-cycle expansion rather than imminent contraction. The 10Y-3M spread at +80 basis points and the 10Y-2Y at +40 basis points represent a meaningful steepening from the deeply inverted levels of 2023, and they align with an FOMC that has moved toward a more neutral policy stance after the tightening cycle of 2022–2023. We can observe that this normalization is consistent with three observations in the broader macro data: core inflation has converged toward the Fed's 2% target, the labor market has rebalanced without breaking, and real GDP growth has stabilized in a 2% handle. Consequently, the recession probability embedded in the NY Fed's term spread model has declined materially from its 2023 peak.

The dovish tilt in Fed communication over recent quarters has reinforced this normalization, with the front end repricing lower as the market prices in a measured easing path. Given the mandate of connecting term structure dynamics to forward policy expectations, we should note that the 2-year note has historically been the most sensitive bellwether for FOMC pricing — its movement tends to lead the 10-year by several weeks when policy expectations shift decisively. The current 40 basis point 10Y-2Y spread reflects this dynamic in real time, with the front end having moved more aggressively than the long end in response to the recent softening of Fed rhetoric.

The broader consequence of normalization is a reallocation of capital across risk assets — including, as we are now observing, a meaningful deployment into the digital asset and Web3 sectors convening this year in Gulf financial hubs such as the Global Games Show and Global Blockchain Show convening in Riyadh. This is not a direct signal from the US yield curve, but it is a downstream capital flow observation consistent with a normalizing term structure: as the cost of carrying duration falls and recession probability recedes, risk budgets expand across asset classes, and the marginal dollar finds its way into venues that a year ago would have been discounted heavily.

A positive curve is not a clean bill of health: it is permission to take risk. The discipline is knowing when to give it back.

Our Forward Baseline and the Next Inflection

Looking ahead, the implication for our forward baseline is that the US yield curve has reverted from a recession-warning configuration to a late-cycle expansion configuration. The next inflection point will likely emerge when the front end begins pricing in the end of the easing cycle — at which point the 10Y-3M spread should be monitored as the primary signal for the next recession probability regime.

We would close with a position that we believe deserves emphasis: the US yield curve remains one of the most informative single-variable recession indicators available to the policy strategist, but its deployment requires methodological discipline. The 10Y-3M spread should anchor the framework, with the 10Y-2Y serving as a secondary confirmation. The 2022–2024 episode underscored that even the most historically reliable signal can fail when structural distortions — duration absorption dynamics under QT, fiscal residue, labor market anomalies — interact with the underlying transmission mechanism. As the curve sits in positive territory today, our recession probability overlay has receded, but the analytical apparatus remains primed for the next inversion. The signal will return; the only question is when, and whether the structural backdrop will once again attenuate it. Given the mandate of connecting macroeconomic shifts to market realities, our job is not to predict that moment but to ensure that when it arrives, the framework is already in place to interpret it correctly.