The Energy Information Administration's August 2026 Short-Term Energy Outlook , released on August 11, puts one number at the centre of this winter's natural gas story. The outlook states that “we expect natural gas inventories to be a record 3,985 billion cubic feet (Bcf) at the end of October 2026,” a figure “which is an increase of 19 Bcf compared with the July STEO and 5% above the five-year average.” Ten years earlier the same commodity produced almost the same reading. On the EIA's weekly Lower 48 working gas series , the last October report of 2016, for the week ending October 28, recorded 3,963 Bcf. The two figures sit 22 Bcf apart, about half a percent. What changed is everything the stock has to be measured against. A stock figure is a numerator. On its own it is a volume, not a condition. This piece takes up a narrow question: when the numerator is flat across a decade and the denominator is not, which denominator should a reader use, and do the pl...
The two-year and ten-year Treasury spread has not been inverted since the summer of 2024. On a monthly average basis the last negative reading was August 2024, at −0.10 percentage points; September 2024 printed +0.10 and the series has not gone below zero since. On 18 August 2026 it stood at +0.52 percentage points, with the two-year at 4.19 percent on 17 August and the ten-year at 4.68 percent on 14 August. The ten-year against the three-month bill, the spread the New York Fed's own recession model prefers, was +0.85 on 17 August.
Almost none of the commentary that followed the 2022–2024 inversion has been revisited in light of that. The inversion was treated as the warning; the return to a positive slope was treated as the all-clear. On the historical record, that sequencing is backwards. In every clean post-1976 episode, the business cycle peak arrived after the spread crossed back above zero, not during the inverted stretch. Twenty-three months have now passed since the September 2024 crossing without a dated peak. That is the anomaly worth explaining, and it is a different question from the one the financial press asked in 2023.
The honest framing is not that the indicator failed. It is that a single spread compresses at least three separate forces into one number, and when the mix of those forces changes, the mapping from spread to outcome changes with it. What follows separates the mechanism from the correlation, walks the four historical crossings, and sets out the conditions under which the current reading would start to mean what it used to.
Where the Curve Actually Is, and What the Rest of the Data Says
A slope reading means little without the level it sits on. The federal funds target range was held at 3-1/2 to 3-3/4 percent at the 28–29 July 2026 meeting, unchanged since December 2025. Three participants dissented, each preferring a quarter-point increase. The statement described economic activity as expanding at a solid pace despite elevated uncertainty tied in part to conflict in the Middle East, said job gains had kept pace with the workforce, and attributed elevated inflation in part to supply shocks concentrated in energy.
The surrounding data is mixed rather than deteriorating. Real GDP rose at a 1.5 percent annual rate in the second quarter of 2026 on the advance estimate published 30 July. July payrolls fell 23,000 while the unemployment rate held at 4.1 percent. Initial claims for the week ending 8 August were 209,000 with a four-week average of 199,000; continuing claims were 1,777,000 and the insured unemployment rate was 1.2 percent. The real-time Sahm rule stood at −0.03 in July against a 0.50 trigger. Headline CPI rose 3.4 percent over the twelve months through July while core rose 2.5 percent.
Two features of that set matter for the curve. The first is the payroll and claims divergence: employment growth has stalled without separations rising, which is the signature of a low-hiring, low-firing labour market rather than a contracting one. The second is the 0.9 point gap between headline and core inflation, which points at a supply shock rather than demand overheating. The World Bank's May 2026 commodity note recorded Brent rising roughly 65 percent, about 46 dollars a barrel, through the end of March, described as the largest monthly increase on record, against a global supply decline of 10.1 million barrels a day in March.
How a Slope Carries Information in the First Place
The long yield is, to a first approximation, the average expected short rate over the term plus a term premium. When the curve inverts, one of two things is happening. Either the market expects the policy rate to be materially lower in a few years than it is today, which normally implies expected weakness, or the compensation demanded for holding duration has collapsed toward zero or below, which is a portfolio-preference story rather than a growth story.
Only the first channel has any claim to causal content, and even then the causation runs indirectly. The curve does not cause a downturn. It reflects a market consensus that a downturn is likely enough to justify pricing cuts. The transmission that follows is the familiar one: bank net interest margins compress when funding costs sit above asset yields, credit standards tighten, marginal borrowers are rationed, and investment plans that penciled at a lower cost of capital are deferred.
The dis-inversion is the more informative event precisely because of that chain. A curve steepens out of inversion when the front end falls, and the front end falls when easing gets priced. Easing gets priced when the labour market or credit conditions have already begun to turn. On this reading, the inversion is the market anticipating stress and the un-inversion is the market pricing the response to stress that has arrived. The recession, on this logic, should cluster around the second event.
Two Steepenings That Look Identical in the Spread
Here the single number breaks down. A spread of +0.52 can be produced two ways. In a bull steepener, short yields fall faster than long yields; the two-year drops below the policy target as cuts are priced in. In a bear steepener, long yields rise faster than short yields, driven by term premium or inflation compensation, and the two-year sits at or above the policy target because tightening, not easing, is the priced risk.
The current configuration is unambiguously the second. With the target ceiling at 3.75 percent and the two-year at 4.19 percent, the front end carries roughly 44 basis points of priced tightening rather than any easing. The Kim-Wright ten-year term premium stood at 0.83 percentage points on 7 August, meaningfully positive rather than compressed. Three dissents in favour of a hike at the most recent meeting is consistent with that pricing.
A curve that steepens because cuts are coming and a curve that steepens because hikes are coming are opposite states wearing the same number. Every historical study that links dis-inversion to recessions was estimated on episodes of the first kind.
The Four Crossings Since 1976, Counted Honestly
The constant maturity series begins in June 1976, which yields six inversion episodes. Using monthly averages and NBER reference dates, the crossings and their outcomes are as follows.
| Episode | Last negative month | First positive month | Next NBER peak | Gap |
|---|---|---|---|---|
| 1978–1980 | April 1980 (−1.03) | May 1980 (+0.73) | January 1980 | peak came 4 months earlier |
| 1980–1982 | June 1982 (−0.17) | July 1982 (+0.15) | July 1981 | crossed zero five times |
| 1989–1990 | March 1990 (−0.04) | April 1990 (+0.07) | July 1990 | 3 months |
| 2000 | December 2000 (−0.11) | January 2001 (+0.40) | March 2001 | 2 months |
| 2006–2007 | May 2007 (−0.02) | June 2007 (+0.12) | December 2007 | 6 months |
| 2022–2024 | August 2024 (−0.10) | September 2024 (+0.10) | none dated | 23 months and counting |
Three episodes are clean, in the sense that the spread crossed zero once and stayed positive: 1990, 2001 and 2007. Their gaps to the following peak were three, two and six months. That is the entire empirical basis for treating dis-inversion as a proximate recession signal. Three observations.
The other episodes are instructive for different reasons. In 1980 the recession was already four months old when the spread crossed back, and the crossing landed two months before the July 1980 trough rather than before the peak. The 1980–1982 stretch crossed zero repeatedly: positive in July 1980, negative from September 1980, positive in November 1981, negative again from February through June 1982, positive from July 1982. Any rule keyed to a single crossing would have generated four signals across that window, and the July 1981 peak arrived while the spread was at −1.07.
The 2019 case is worth stating precisely because it is frequently miscited. On a monthly average basis the ten-year minus two-year spread never went negative in 2019 at all; the lowest readings were +0.06 in August and +0.05 in September. The inversion that dominated coverage that year existed only in daily closes, and lasted days. Whether 2019 counts as an inversion episode depends entirely on the sampling frequency chosen, which is a strong hint about how much weight the classification can bear.
The Part This Model Misses
Several things could be true instead of the reading above, and each deserves equal weight.
The peak may already exist and simply be undated. The NBER committee dates cycles in arrears, historically with lags running from several months to well over a year, and it explicitly waits for data revisions before announcing. A July 2026 peak would not plausibly be announced before 2027. The claim that no recession has occurred in 23 months is therefore a claim about the published record, not about the underlying economy. If a peak is later dated to mid-2026, the gap becomes roughly 21 months and the anomaly shrinks without disappearing.
The sample is too small to support the inference either way. Three clean episodes cannot distinguish a three-month median lag from a twelve-month one. Any confidence interval around that median spans the entire plausible range. The correct statement is that dis-inversion has preceded three of the last several peaks by a short interval, not that it reliably does so.
Monthly averaging is a choice that changes the answer. Daily closes would move several of the crossing dates by weeks and would add 2019 as an episode. Nothing in the underlying theory specifies a sampling frequency, and results that flip on that choice should be held loosely.
The bear steepener framing could be wrong about direction. A term premium rebuild that pushes long yields up tightens financial conditions on its own, independent of policy. Mortgage rates, corporate refinancing costs and commercial property capitalisation rates all key off the long end. A steepening driven by the long end rising is not benign; it can transmit stress through exactly the channels an inversion is supposed to warn about, only later and through different balance sheets.
A supply shock can produce a downturn the curve never anticipates. If energy prices force real income lower while nominal policy stays restrictive because headline inflation is elevated, the resulting contraction would have no yield curve precursor at all. The gap between 3.4 percent headline and 2.5 percent core inflation is the measurable version of that risk. The curve has no mechanism for pricing a shock that has not happened yet.
What to Watch Next Week
Four items carry the most information for this specific question.
- The two-year against the policy ceiling. This is the single cleanest test of which steepening regime is operating. As long as the two-year holds above 3.75 percent, the front end is pricing tightening and the historical dis-inversion analogues do not apply. A move below the ceiling would be the first evidence of a regime change, and a move below the range floor of 3.50 percent would mean cuts are being priced with conviction.
- Weekly initial claims and the four-week average. The 199,000 four-week average is the strongest single argument against imminent contraction. Payroll weakness without claims deterioration is stalled hiring; payroll weakness with claims deterioration is separations. The two have very different implications and are distinguishable weekly.
- Continuing claims relative to 1,777,000. In a low-hiring regime, the duration of unemployment rises before the level does. Continuing claims capture that before the household survey unemployment rate moves.
- The headline-core inflation gap. If the 0.9 point wedge narrows because headline falls, the supply shock is passing through and the constraint on policy loosens. If it narrows because core rises, the shock is broadening and the constraint tightens.
Scheduled releases carry more weight than commentary here. The employment situation report, the weekly claims release each Thursday, and the CPI release are the observations that would move the assessment.
Concrete Framework — How to Test It
The following is a monitoring checklist, not a forecast. Each item is falsifiable against a published series.
- Record the regime, not the spread. Each month, note the two-year yield minus the upper bound of the policy target range. Positive means tightening is priced and the bear steepener regime holds; negative means easing is priced and the historical analogues become relevant. The August 2026 reading is +0.44 percentage points.
- Track the term premium separately from the slope. The Kim-Wright ten-year term premium at 0.83 percentage points is the check on whether the long end is rising for risk-compensation reasons or growth reasons. A fall below roughly 0.25 while the slope stays positive would indicate the composition of the spread has shifted.
- Set a labour trigger with a threshold, not a narrative. The four-week claims average sustained above roughly 260,000 for four consecutive weeks, or the real-time Sahm rule reaching 0.50, would each constitute independent confirmation. Neither is close as of the 8 August week.
- Date the crossing, then start a calendar. September 2024 is month zero. August 2026 is month 23. If a peak is eventually dated inside this window, the historical range extends rather than breaks; if the window passes 30 months with no peak, the dis-inversion rule has failed on this episode and should be weighted accordingly in future.
- Hold both scenarios explicitly. The scenario in which the curve is still right and the lag is simply longer than any prior episode is not less plausible than the scenario in which the regime has changed. Neither has been settled by the data available in August 2026, and writing down which observation would move the assessment is more useful than picking one now.
The defensible summary is narrow. The spread that drove two years of recession commentary has been positive for 23 months. The three clean historical precedents for that crossing implied a peak within two to six months, and none has been dated. The composition of the current spread differs from those precedents in a way that is measurable rather than rhetorical: the front end is pricing tightening rather than easing. That does not resolve the question. It changes which series would resolve it.
Every figure cited here is drawn from a published series and dated. Where a claim could not be tied to one, it has been left out rather than estimated.
This article is macroeconomic analysis for general information and is not investment or financial advice.
Comments
Post a Comment