T10Y3M vs T10Y2Y: Which Yield Curve Inversion Signal to Choose

T10Y3M and T10Y2Y are two complementary curve spreads: the former remains the NY Fed’s canonical variable for reasons of mechanism rather than of tally, while the latter inverts earlier and captures more of the expectations gap among shorter maturities.

Choosing between the two requires understanding what each specifically measures and why the academic community prefers one when the financial press more frequently cites the other.

What each spread measures and why they diverge

T10Y3M and T10Y2Y share the same long leg, the 10-year Treasury yield (series DGS10), but differ in their short leg. T10Y3M uses the 3-month constant-maturity yield (DGS3MO), not the bill’s discount rate (DTB3): the difference, immaterial in calm markets, reaches 17 basis points in December 2024, enough to move an un-inversion date. That yield is a proxy for the Fed Funds effective rate, but one whose gap widens precisely when the cycle moves: a 10-basis-point median since 1990, but a 28-basis-point median over 2022-2023 and as much as 115. T10Y2Y uses the 2-year Treasury yield (DGS2), which already embeds partial expectations about the future path of monetary policy, particularly anticipations of Fed hikes or cuts over the next twenty-four months. The detail of the underlying FRED series and their calculation conventions is laid out in the technical meaning of T10Y3M.

This short-leg difference produces a different reading of the curve signal. T10Y3M compares current monetary policy (embodied in the 3-month) with very long-run expectations (10-year). T10Y2Y compares two sets of expectations at different horizons (2-year vs 10-year). In practice, T10Y2Y inverts earlier because the 2-year yield rises faster than the 3-month during anticipated-tightening phases: the bond market “prices in” coming Fed hikes before they materialize in the effective Fed Funds rate. Further on this: the mechanics of yield curve control.

The 2022-2024 episode documents this precocity precisely. T10Y2Y touched zero on April 1, 2022, when T10Y3M was still positive at over 180 basis points; it came back above on April 5 and only inverted without interruption from July 6. T10Y3M did not cross the negative boundary until October 18, 2022, and durably on the 25th: six and a half months after the first touch, fifteen weeks after the second. The gap does not come from a short leg rising more slowly. From April 1 to October 31, 2022, the 3-month gained 369 basis points and the 2-year 207. It comes from where they started, the 2-year sitting 191 basis points above the 3-month. The same asymmetry holds on un-inversion: T10Y2Y turned positive again in September 2024, T10Y3M in December 2024, a three-month gap as well. This sequence is documented month by month in the 2022-2024 T10Y3M inversion, which restores the precise values of both spreads over the same window.

The mechanical decomposition of this 203-day lag illuminates the structural difference between the two spreads. Between April and October 2022, the Fed raised the Fed Funds target range from 0.25-0.50 percent to 3.00-3.25 percent (a cumulative 275 basis points), but markets were already pricing additional hikes through 2023. The 2-year yield, embedding this anticipation, rose from 2.32 percent in early April to 4.28 percent in late October, outpacing the 10-year rise from 2.65 percent to 4.05 percent over the same window. The 3-month yield, meanwhile, only tracked the Fed Funds rate effectively, climbing from 0.55 percent to 4.05 percent. The T10Y2Y therefore mechanically inverted first because the 2-year was running ahead of policy reality, while the T10Y3M waited for the policy rate itself to catch up.

T10Y2Y precocity against T10Y3M robustness: what the five common cycles say

The count needs to be set on the right window. Neither FRED series reaches 1968: T10Y2Y begins on June 1, 1976, T10Y3M on January 4, 1982. Over the common window, five cycles see both spreads invert durably: 1989, 2000, 2006-2007, 2019 and 2022-2024. T10Y2Y touches zero first in four of them, with four to seven months of lead at first touch. The exception is 2019, and it is not a minor one: T10Y3M inverts in May, three months before T10Y2Y spends three sessions below zero in August. The order also flips in 2006 as soon as durable inversions are compared rather than first touches. Related analysis: T10Y3M 1998: The LTCM False Signal, Unique Exception Since 1968.

But precocity is not robustness, and the argument is poorly served by the examples usually given for it. Over 2005-2006, T10Y2Y is not the one that sinks further: it logs 166 sessions below zero for a trough of −0.19, while T10Y3M logs 124 for a trough of −0.60, three times deeper. Over 2019 the gap is wider still and runs the same way: 104 sessions below zero for T10Y3M, trough −0.52, against three sessions at −0.04 for T10Y2Y. The spread presented as the sober one is the one that cried loudest in both episodes meant to prove the opposite. What genuinely separates the two series is not a false-positive count, it is what each of them measures.

The comparative analysis by Estrella and Mishkin (1996, NBER WP 5379), which laid the methodological groundwork for the NY Fed choice, tested T10Y3M against T10Y2Y and several other candidate spreads over the 1959-1995 sample. Their empirical conclusion: T10Y3M dominates on nearly every metric (likelihood ratio, pseudo-R², out-of-sample stability), with a margin sufficient to justify the NY Fed selection as principal variable of its probit model. This dominance was reconfirmed by Bauer and Mertens (Federal Reserve Bank of San Francisco Economic Letter 2018-07 then 2023-14) over the expanded 1959-2023 sample.

Why the NY Fed retains T10Y3M despite T10Y2Y’s precocity

Three academic arguments justify the institutional preference for T10Y3M, despite the tactical convenience of T10Y2Y. These arguments explain why the NY Fed T10Y3M signal remains the empirical reference rather than an alternative substitute.

First argument, alignment with the transmission channel. The causal mechanism documented in the literature operates through bank net interest margins. U.S. commercial banks fund themselves at very short maturities (demand deposits, overnight repo, three-month certificates of deposit) and lend at long maturities. Their effective funding cost is better captured by the 3-month than by the 2-year: a 2-year term deposit is rare, while a 3-month deposit is the norm. T10Y3M thus more faithfully translates the bank-margin compression that is the transmission channel from inversion to recession.

Second argument, market purity. The 3-month yield is closely correlated to the Fed Funds effective rate: the median gap is ten basis points since 1990, even if it triples during active tightening. The 2-year yield, by contrast, embeds expectations that may diverge from the actual Fed path. This divergence introduces noise for current-tightening reading: an inverted T10Y2Y may signal either current tightening or an expectations reversal on the future path, two distinct macroeconomic readings.

Third argument, out-of-sample robustness. The empirical tests by Estrella and Mishkin (1996) precisely measured the stability of calibration coefficients between estimation and test periods. T10Y3M produced the highest stability among all candidate spreads, which is the operationally most important statistical criterion for out-of-sample robustness. T10Y2Y produced less stable coefficients, an indicator of greater overfitting in the calibration. These three arguments rest on mechanism, not on a scoreboard of signals, which is what makes them sturdier than the false-positive counts sometimes offered in their place.

These three arguments stack and converge toward T10Y3M as the principal variable. The academic debate is not closed: Engstrom and Sharpe (Federal Reserve Board 2018) reignited the discussion with the near-term forward spread as a possible alternative, but no challenger has displaced T10Y3M in the official NY Fed publication for three decades.

Why T10Y2Y remains used in financial press

If the NY Fed and academic literature converge on T10Y3M, why does general financial press more frequently cite T10Y2Y? Three non-academic factors explain the media preference.

First, tactical readability. T10Y2Y is the indicator tracked by bond traders as a barometer of short-term Fed expectations. Its daily movement is more volatile than T10Y3M’s, which produces more market events (short inversions, local divergences) that fuel daily market-coverage articles. T10Y3M, more stable and more aligned with Fed Funds, produces less material for daily market columns. On the same theme: the Eco3min framework for short versus long maturities.

Second, operational precocity. For a financial journalist who wants to flag curve inversion to readers, T10Y2Y is the first to cross the negative boundary, typically three months before T10Y3M. The “first inversion” character of T10Y2Y produces a more immediate media signal, even when the NY Fed probit signal has not yet activated.

Third, historical inheritance. The 2-year / 10-year spread has been cited in financial press since the 1980s, predating the academic formalization of T10Y3M by Estrella and Mishkin in 1996. This earlier media anchoring has persisted through editorial inertia, even after the NY Fed installed T10Y3M as canonical variable of its probit model.

A postscript as of September 22, 2026 turns the usual picture of the two spreads around. Since January 2025, T10Y3M has spent 86 sessions below zero, with a trough of −27 basis points on April 4, 2025, while T10Y2Y has not gone back below once and stands at +20 basis points. Over the past twenty months, every inversion signal has come from the spread reputed to be the quieter one. Logged daily since 1982, the 10y-3m spread data allow those 86 sub-zero sessions to be recounted one at a time.

Combined reading: when to use which

Rather than mechanically opposing the two spreads, the rigorous reading uses them as complementary tools covering different clocks. Three configurations deserve distinction for daily reading.

Configuration 1, anticipated signal. When T10Y2Y inverts before T10Y3M, the bond market signals an anticipation of monetary turning point before the Fed has completed its tightening cycle. This configuration was observed from July to October 2022 (T10Y2Y inverted without interruption, T10Y3M still positive) and signals a high conditional probability of subsequent T10Y3M inversion, without yet constituting a recession signal in the sense of the NY Fed probit model. The appropriate reading: heightened surveillance, but no yet activation of the reference signal.

Configuration 2, confirmed double inversion. When both spreads are simultaneously in negative territory, the recession signal is confirmed on both variables: the situation observed over most of October 2022 to September 2024. This configuration is the clearest for macro reading: convergence of both indicators on a coherent signal, alignment with NY Fed probit model calibration generally exceeding 30 percent then 50 percent of conditional probability.

Configuration 3, divergence at un-inversion. When T10Y2Y turns positive before T10Y3M (as in September-December 2024), the market anticipates coming Fed cuts while effective bank funding cost remains high. This configuration is read by economists as transitory, resolved in the following three to six months by T10Y3M un-inversion as the Fed actually cuts rates. This is precisely the profile observed in the 2024 episode. In the same vein: T10Y3M: The August-to-December 2024 Un-Inversion, End of Record Inversion.

This three-configuration typology enriches the binary “inversion vs no inversion” reading by proposing a three-dimensional grid: anticipated signal (T10Y2Y alone), confirmed signal (both), exit signal (T10Y2Y un-inverted before T10Y3M). It allows using T10Y2Y precocity as preliminary alert while retaining T10Y3M robustness as the reference variable for conditional recession probability. For reading the broader cycle of central-bank policy and rate-cycle transmission, the two spreads read together rather than in competition.

Last updated — 3 November 2026