Yield Curve Recession Signal: Reconstructing the May 2023 Warning
The May 2023 signal was strong. It was not a promise.
The 22 reviewed May 2023 observations of the 10-year minus 3-month Treasury spread averaged -1.73 percentage points. Applying the New York Fed probit coefficients maintained in September 2026 produces a retrospective 71.4% recession probability for May 2024. NBER did not classify May 2024 as a recession as of this study's September 12, 2026 evidence cutoff.
The 71.4% result is reproducible from three numbers
The model uses the normal cumulative distribution of a linear probit score. Grizzly Bulls recomputes the May average from the published daily observations before applying the maintained coefficients.
1. Monthly average: sum the 22 May 2023 T10Y3M observations and divide by 22 = -1.734545 percentage points.
2. Probit score: -0.5333 + (-0.633 ร -1.734545) = 0.564667.
3. Probability: ฮฆ(0.564667) = 0.713850, or 71.4%.
Coefficient snapshot: September 6, 2026. New York Fed estimation window shown in the cited source: 1959-01 through 2009-12.
What happened twelve months later?
NBER's business-cycle chronology did not classify May 2024 as a recession. At the September 12, 2026 research cutoff, its most recent U.S. peak remained February 2020 and its most recent trough remained April 2020.
The 22 May 2023 spread observations
The deepest daily inversion in this dataset was -1.89 percentage points on May 4, 2023.
| Date | 10y minus 3m spread |
|---|---|
| May 1, 2023 | -1.68 percentage points |
| May 2, 2023 | -1.80 percentage points |
| May 3, 2023 | -1.88 percentage points |
| May 4, 2023 | -1.89 percentage points |
| May 5, 2023 | -1.82 percentage points |
| May 8, 2023 | -1.79 percentage points |
| May 9, 2023 | -1.76 percentage points |
| May 10, 2023 | -1.81 percentage points |
| May 11, 2023 | -1.81 percentage points |
| May 12, 2023 | -1.79 percentage points |
| May 15, 2023 | -1.71 percentage points |
| May 16, 2023 | -1.69 percentage points |
| May 17, 2023 | -1.69 percentage points |
| May 18, 2023 | -1.66 percentage points |
| May 19, 2023 | -1.59 percentage points |
| May 22, 2023 | -1.68 percentage points |
| May 23, 2023 | -1.64 percentage points |
| May 24, 2023 | -1.64 percentage points |
| May 25, 2023 | -1.55 percentage points |
| May 26, 2023 | -1.54 percentage points |
| May 30, 2023 | -1.86 percentage points |
| May 31, 2023 | -1.88 percentage points |
A recession probability is not a market-timing rule
The yield curve can be useful macro evidence without telling an investor exactly when to sell stocks, how much exposure to cut, or when to buy back. A trading rule still needs a position-sizing policy, execution assumptions, a benchmark, and evidence about what happens between the signal and the eventual economic outcome.
That distinction matters here. The recent inversion became historically deep, yet the reconstructed twelve-month target missed. An investor who treated one macro probability as an automatic all-out equity signal would be making a much stronger claim than the New York Fed model itself makes.
Methodology and limitations
- Use the 22 official FRED T10Y3M daily observations for May 2023 included in this study.
- Recompute their arithmetic mean rather than copying a precomputed summary value.
- Apply the New York Fed probit coefficients shown in the maintained September 6, 2026 probability chart: alpha -0.5333 and beta -0.633.
- Convert the probit score through the standard normal cumulative distribution and compare the twelve-month target month with NBER's business-cycle chronology as checked on September 12, 2026.
- Do not describe the result as the exact real-time probability published in May 2023. The coefficient set is the maintained 2026 set.
This is a one-episode forecast reconstruction, not a historical backtest of every inversion, a calibration study, or an equity-return study. NBER dates turning points retrospectively, so the outcome statement is bounded to the chronology available at the research cutoff.
Primary sources
Download the reconstruction data
The CSV contains the 22 daily T10Y3M observations used in the calculation. The JSON adds the model inputs, reconstructed probability, outcome boundary, and source URLs.
Citation and reuse
When citing the 71.4% figure, describe it as a Grizzly Bulls retrospective reconstruction using the maintained September 2026 New York Fed coefficients.