CVRP: Convex Recession Probability Index

1. Abstract

The CVRP is a 0–100 composite leading indicator of US recession risk. It synthesizes five independent channels, each scored on a 0–20 scale: (1) yield-curve inversion depth, (2) Sahm-rule proximity, (3) initial claims momentum, (4) high-yield credit spread z-score, and (5) the Conference Board Leading Economic Index. The composite is normalised to account for missing components and publishes only when at least three of the five channels have recent data. The design philosophy is multi-channel triangulation: no single signal can push the index to extreme levels, mitigating false positives from any one dominant regime (e.g. yield-curve inversion in 2022 without corresponding labour-market deterioration).

CVRP is not backtested as a classifier. It is intended as a dashboard gauge for real-time monitoring, not as a forecasting model with calibrated probabilities. Readers seeking a probability of recession over a defined horizon should pair CVRP with horizon-specific academic forecasts (Estrella-Mishkin spread model, Chauvet-Piger smoothed NBER probability) rather than reinterpret the 0–100 score as a probability.

2. Motivation

Single-signal recession indicators have well-documented failure modes. The 2s10s yield-curve inversion of 2022 produced a record-deep inversion without an accompanying recession, as of the time of this writing; the gap between signal and outcome was filled by unusually strong labour markets, large fiscal transfers, and net wealth effects from post-pandemic home-price appreciation. A practitioner watching only the yield curve would have been positioned for recession throughout 2023, missing one of the strongest nominal-earnings years on record.

Conversely, a practitioner watching only the unemployment rate would have been late. By the time non-farm payrolls turn negative, the recession is already in progress; NBER typically dates the business-cycle peak one to two quarters earlier than the unemployment-rate trough. Sahm (2019) formalised this with a rule that triggers earlier than NBER dating by observing acceleration in unemployment rather than levels, but even the Sahm rule is a coincident-to-very-slightly-leading indicator, not a leading one.

The CVRP addresses both failure modes by requiring agreement across channels that capture different transmission mechanisms: the yield curve captures Fed-policy-induced tightening, the Sahm rule captures labour-market momentum, initial claims captures the weekly pulse of layoffs, the high-yield credit spread captures the market’s real-time pricing of default risk, and the LEI captures the real-economy aggregate. When at least three of these simultaneously signal recessionary conditions, the false-positive rate drops sharply.

3. Channel definitions

3.1 Yield-curve inversion depth (0–20 points)

Uses the more inverted (worse) spread between the 10-year minus 2-year Treasury spread (FRED series T10Y2Y) and the 10-year minus 3-month spread (T10Y3M). When either spread is positive, the channel contributes 0 points. When either is negative, the channel scales linearly to 20 points at a 150 basis-point inversion.

Academic basis: Estrella and Mishkin (1998) showed that the 10-year minus 3-month spread is the single most powerful leading indicator of US recession, with out-of-sample predictive power at 4–6 quarters ahead. Subsequent work has confirmed robustness across refreshed samples and alternative spread definitions. The 2s10s captures a similar signal but responds more to market expectations of the front-end path, while 10y-3m captures term premium and direct Fed policy stance. The CVRP takes the worse of the two as the conservative signal.

3.2 Sahm-rule proximity (0–20 points)

Uses the real-time Sahm indicator (SAHMREALTIME), which is the difference between the 3-month moving average of the unemployment rate and its minimum over the prior 12 months. A Sahm indicator at or above 0.50 historically coincides with NBER-dated recessions with a 100% hit rate since 1950, though it produced a close call in August 2024 when the reading briefly crossed 0.53 during a summer hiring slowdown that did not progress to a full recession.

The CVRP scales the Sahm indicator linearly from 0 at 0.00 to 20 at 0.50. A reading of 0.25 therefore contributes 10 points, half the maximum. Readings above 0.50 cap at 20. This conservative shape ensures Sahm alone cannot push the composite to 60+ (the “high” threshold) without corroboration from another channel.

3.3 Initial claims momentum (0–20 points)

Uses the 4-week moving average of initial unemployment claims (IC4WSA). The channel computes the 13-week percentage change and maps 0% change to 0 points, +30% change to 20 points. Negative changes (claims falling) contribute 0 points, they are never recession-signalling.

Initial claims is the highest-frequency labour-market indicator available (weekly, with a one-week lag). It typically leads non-farm payrolls by two to three months and leads the Sahm rule by a similar margin. The 13-week window smooths out holiday noise and single-week measurement error while remaining responsive enough to catch turning points before the monthly payroll print.

3.4 High-yield credit spread z-score (0–20 points)

Uses the ICE BofA US High Yield Option-Adjusted Spread (BAMLH0A0HYM2) as a z-score against its own 1-year (252 trading day) rolling history. The channel uses the z-score rather than absolute spread levels to adjust for secular regime shifts: a 400 bp spread signals something different in 2024 than in 2008, because the level of the risk-free rate, the composition of the index, and the baseline liquidity premium have all drifted.

The channel maps z = 0 to 0 points, z = 3 to 20 points, capping at 20 for z > 3. A z-score of 2 (roughly the 97.5th percentile of the rolling distribution) contributes ~13 points. Using a rolling z-score means the channel measures acceleration of credit stress relative to the recent regime, catching inflection points earlier than a fixed-threshold approach would.

3.5 Leading Economic Index (0–20 points)

Uses the Conference Board Leading Economic Index (USSLINDor successor; see “Known limitations” below). Channel score is based on the 6-month annualised rate of change: 0% maps to 0 points, −6% (a historically deep decline rate) maps to 20 points. Positive annualised changes contribute 0 points.

The LEI is a 10-component (now 4-component on the FRED pull after the 2020 discontinuation) aggregate that blends equity prices, manufacturing new orders, consumer expectations, building permits, and several other leading series. It is the only channel that simultaneously incorporates real-economy data and financial-markets data, which gives it a different failure signature than the other channels.

4. Composite formula and normalisation

The composite score is the normalised sum of available channel scores, expressed on a 0–100 scale:

where n_availableis the number of channels with a recent data point (≤ 60 days old). The index requires at least three of five channels to be available before publishing; when fewer are available, the index is marked as low-confidence and suppressed from public display.

The normalisation means a 3-channel composite where all three max out at 20 points produces CVRP = 100, the same score as a 5-channel composite where all five max out. This is deliberate: if three channels agree on extreme recessionary conditions, the signal is already strong enough to report at maximum. The confidence flag communicates the narrower evidence base to the reader.

5. Interpretation thresholds

These thresholds are descriptive labels, not backtested probabilities. Users should not interpret “CVRP at 65” as “65% probability of recession.” For probability-calibrated recession models, see Estrella-Mishkin or the Smoothed Recession Probabilities (Chauvet-Piger) available on FRED.

6. Known limitations

Soft landings and no-recession yield-curve inversions. The CVRP’s multi-channel design mitigates but does not eliminate the risk of false positives during episodes like 2022–2023 where one or two channels signal recessionary conditions that fail to materialise. The LEI, in particular, has signalled recession multiple times since 2022 without producing NBER-dated downturns, a pattern that may reflect structural changes in the US economy (services-dominated labour market, large fiscal buffers, private-sector balance-sheet strength) rather than a broken indicator.

LEI component attrition. The Conference Board LEI has undergone multiple methodological revisions. The FRED series USSLIND was effectively discontinued in February 2020; later readings rely on a reduced component set. When fewer than three LEI sub-components are available, the channel is marked missing and excluded from the composite.

Real-time vs. vintage data. The Sahm rule calculation in particular is sensitive to the distinction between real-time unemployment data and later-revised vintage data. The CVRP uses the FRED real-time series SAHMREALTIME, which matches what a practitioner would have seen at the time of publication, but this means historical CVRP readings cannot be perfectly reconstructed from today’s revised data.

Not a classifier. CVRP has not been calibrated against a specific recession-probability target. Users who need a probability in the “what are the odds of a recession in the next 12 months” sense should use a purpose-built probit model on the 10y-3m spread or a smoothed probabilities model, not re-scale CVRP.

7. Data sources and update cadence

• T10Y2Y, T10Y3M: FRED, daily (business days)
• SAHMREALTIME: FRED, monthly (first Friday)
• IC4WSA: FRED, weekly (Thursday)
• BAMLH0A0HYM2: FRED (ICE BofA), daily (business days)
• USSLIND / successor: FRED / Conference Board, monthly

CVRP is recomputed on every FRED sync (currently every 6 hours). The channel-level breakdown is published alongside the composite at /indicators/cvrp.

8. References

1. Estrella, A. and Mishkin, F. (1998). “Predicting U.S. Recessions: Financial Variables as Leading Indicators.” Review of Economics and Statistics, 80(1), 45–61.
2. Sahm, C. (2019). “Direct Stimulus Payments to Individuals.” In Recession Ready: Fiscal Policies to Stabilize the American Economy, Brookings.
3. Chauvet, M. and Piger, J. (2008). “A Comparison of the Real-Time Performance of Business Cycle Dating Methods.” Journal of Business and Economic Statistics.
4. The Conference Board. Business Cycle Indicators Handbook (LEI methodology).
5. ICE Data Indices. ICE BofA US High Yield Index Methodology.