Probability of Recession: Key Indicators and the NY Fed Model
The probability of recession is one of the most consequential questions in macroeconomics and investment management. Whether you’re allocating a pension fund, managing a bond portfolio, or evaluating equity risk, understanding how economists and market signals estimate recession likelihood is essential. This guide covers the key indicators investors monitor, the New York Fed’s quantitative model, and how to position portfolios when recession risk is elevated. For background on how central bank policy and inflation dynamics interact with recession risk, see our companion articles in this series.
What Defines a Recession?
The National Bureau of Economic Research (NBER) defines a recession as a significant decline in economic activity that is spread across the economy and lasts more than a few months. The NBER’s Business Cycle Dating Committee examines GDP, employment, personal income, industrial production, and retail sales to make this determination.
A common shorthand is “two consecutive quarters of negative GDP growth,” but this is not the NBER’s actual definition. The 2001 recession, for example, did not include two consecutive negative-GDP quarters yet was still classified as a recession because employment and industrial production declined significantly.
One important nuance: NBER dates recessions retrospectively. The official call often arrives months after the recession has already begun or even ended. For investors, this lag means you cannot rely on NBER announcements for real-time portfolio decisions — you need forward-looking indicators.
| Recession | Dates | Duration | Cycle-Peak Unemployment* | Real GDP Decline (Peak to Trough) |
|---|---|---|---|---|
| COVID-19 | Feb – Apr 2020 | 2 months | 14.7% (Apr 2020) | -10.1% |
| Great Recession | Dec 2007 – Jun 2009 | 18 months | 10.0% (Oct 2009) | -4.3% |
| Dot-Com | Mar – Nov 2001 | 8 months | 6.3% (Jun 2003) | -0.3% |
| Gulf War | Jul 1990 – Mar 1991 | 8 months | 7.8% (Jun 1992) | -1.4% |
*Cycle-peak unemployment often occurs after the recession has officially ended, as labor markets lag the broader recovery.
Leading U.S. Recession Indicators
Investors and economists monitor a range of leading indicators — data series that tend to change direction before the overall economy does. These signals provide advance warning, typically 3 to 24 months before a recession begins.
| Indicator | Recession Signal | Typical Lead Time | Source |
|---|---|---|---|
| Treasury Yield Curve (10Y-3M) | Spread turns negative (inversion) | 6–24 months | Federal Reserve / FRED |
| Conference Board LEI | 6+ monthly declines (3Ds: duration, depth, diffusion) | 6–12 months | The Conference Board |
| ISM Manufacturing PMI | Sustained readings below 50 | 3–6 months | Institute for Supply Management |
| Initial Jobless Claims | Rising trend; sustained above 300K | 3–9 months | Dept. of Labor (weekly) |
| High-Yield Credit Spreads | OAS widens above 500 bps | 6–12 months | ICE BofA Index / FRED |
| Consumer Confidence | Sharp decline from prior readings | 3–9 months | The Conference Board |
The yield curve — specifically the spread between the 10-year and 3-month Treasury yields — has the strongest historical track record as a single recession predictor. But each indicator captures a different dimension of the economy, which is why professional forecasters use multiple signals together. For how yield curve shape influences fixed income strategy, see yield curve strategies.
No single indicator is infallible. The most robust recession assessments triangulate across yield curve signals, labor market data, credit conditions, and survey-based indicators. When multiple indicators align, the signal is far more reliable than any one measure alone.
The Yield Curve as Recession Predictor
The yield curve inverts when short-term Treasury yields exceed long-term yields — the bond market pricing in expected future rate cuts because investors believe the Federal Reserve will need to lower rates to support a weakening economy. The 10-year minus 3-month Treasury spread (10Y-3M) is the specific measure used by the New York Fed’s recession probability model. The 10Y-2Y spread is also widely tracked but is a different measure with its own track record; this article focuses on the 10Y-3M spread because of its direct connection to the quantitative model discussed below.
The 10Y-3M spread has inverted before every U.S. recession since 1970. However, the 1998 inversion did not lead to a recession within the typical lead window — the Fed cut rates aggressively in response to the LTCM crisis, and the economy continued expanding. Policy responses can alter the outcome that the yield curve otherwise signals.
| Inversion Period (10Y-3M) | Spread Minimum | Recession Start | Lag (Months) |
|---|---|---|---|
| Mid-2019 | -0.52% | Feb 2020 | ~6 |
| 2006 – 2007 | -0.64% | Dec 2007 | ~17 |
| 2000 | -0.81% | Mar 2001 | ~8 |
| 1998 | -0.17% | None within lead window | N/A |
| 1989 | -0.30% | Jul 1990 | ~14 |
The variable lag — anywhere from 6 to 24 months — is one of the biggest challenges of using the yield curve as a timing tool. An inversion tells you recession risk is elevated, but it does not tell you precisely when the contraction will begin.
Estimating Recession Probability: The NY Fed Yield Curve Model
The most widely cited quantitative recession probability model was developed by Arturo Estrella and Frederic Mishkin (1996) and is published monthly by the Federal Reserve Bank of New York. It uses a probit regression — a statistical technique that converts a continuous input (the yield spread) into a probability between 0% and 100%.
The spread is entered in percentage points — for example, if the 10-year yield is 4.0% and the 3-month yield is 4.5%, the spread is -0.5 (not -0.005). The model estimates the probability of a recession occurring within the next 12 months.
Scenario: 10-Year Treasury = 4.0%, 3-Month Treasury = 4.5%
Step 1: Calculate the spread: 4.0 – 4.5 = -0.5 percentage points
Step 2: Plug into the model: -0.5333 – 0.6330 × (-0.5) = -0.5333 + 0.3165 = -0.2168
Step 3: Look up Φ(-0.2168) in the standard normal table: ≈ 0.414
Result: The model estimates a 41.4% probability of recession within 12 months.
Interpretation: This reading is well above the long-run average of roughly 15%, indicating elevated recession risk. However, a 41.4% probability also means a 58.6% chance that no recession occurs — probability is not certainty.
The NY Fed model produces a 12-month-ahead probability, not a timing tool. A reading of 40% does not mean a recession will begin in exactly 12 months — it means the model estimates a 4-in-10 chance of recession occurring at some point within the next year.
Why Recession Probability Estimates Can Disagree
Different forecasting models can produce materially different recession probability estimates at the same point in time. The NY Fed model relies solely on the Treasury yield spread, while other approaches incorporate labor market data (such as the Sahm Rule, which triggers when the unemployment rate rises 0.5 percentage points above its 12-month low), credit market indicators, or survey data like PMI readings.
This disagreement is a feature, not a bug. Each model captures a different dimension of economic health. When term-spread models, labor-market indicators, and credit conditions all point toward recession simultaneously, the signal is far more compelling than any single model’s output. When they diverge, it suggests the economic picture is genuinely uncertain.
Leading Indicators vs Lagging Indicators
Understanding the difference between leading and lagging indicators is critical for investment timing. Leading indicators change direction before the economy does, giving investors time to adjust. Lagging indicators confirm what has already happened — useful for analysis but too late for portfolio positioning.
Leading Indicators
- Signal recession 6–18 months ahead
- Yield curve, Conference Board LEI, ISM PMI, building permits
- Used for defensive positioning before downturns
- Limitation: false positives and variable lead times
Lagging Indicators
- Confirm recession after it has begun
- Unemployment rate, corporate earnings, average duration of unemployment
- Used to confirm depth and duration of downturns
- Limitation: too late for proactive portfolio decisions
CPI is typically classified as a lagging indicator in business cycle analysis — inflation tends to fall after a recession is underway as demand weakens. However, some CPI components (particularly shelter and services) can behave closer to coincident, depending on the inflation regime. This nuance is why professional economists avoid rigid categorization.
Portfolio Implications of Recession Risk
When recession probability rises above historical averages, investors can take defensive steps without abandoning their long-term strategy. The goal is to reduce vulnerability to cyclical downturns while maintaining enough equity exposure to participate in any continued expansion or recovery.
| Sector | Typical Recession Behavior | Category | Rationale |
|---|---|---|---|
| Utilities | Modest decline to flat | Defensive | Non-discretionary demand; regulated earnings |
| Healthcare | Modest decline | Defensive | Essential services; inelastic demand |
| Consumer Staples | Modest decline | Defensive | Food, household goods — purchased regardless of cycle |
| Financials | Significant decline | Cyclical | Rising credit losses; NIM compression |
| Industrials | Significant decline | Cyclical | Capital spending cuts; inventory destocking |
| Energy | Severe decline | Cyclical | Demand collapse drives commodity price decline |
Key defensive strategies include extending bond duration (long-duration bonds benefit from rate cuts during recessions), rotating toward quality (companies with strong balance sheets and stable cash flows), and increasing cash allocation to 10–20% to preserve optionality for buying during selloffs. For quantitative approaches to managing downside risk, see our guides on Value at Risk and maximum drawdown.
Defensive positioning is not market timing. Maintaining core equity exposure is essential for long-term returns. Investors who moved entirely to cash during the 2008 crisis often missed the powerful 2009 recovery. Gradual adjustments in allocation are more sustainable than dramatic shifts.
Common Mistakes
1. Treating any single indicator as infallible. The 1998 yield curve inversion did not lead to a recession within the typical lead window. No indicator has a perfect track record, and relying exclusively on one signal creates blind spots.
2. Ignoring the variable lag between signal and recession. Yield curve inversions have preceded recessions by anywhere from 6 to 24 months. Selling equities immediately after an inversion can mean missing months or even years of positive returns before any downturn materializes.
3. Panic-selling based on probability estimates. A 40% recession probability means there is a 60% chance that no recession occurs. Probability is not certainty, and overreacting to elevated but sub-majority probabilities can be more costly than the recession itself.
4. Confusing economic slowdown with recession. GDP growth decelerating from 3% to 1% is a slowdown — not a recession. Growth is still positive. Many periods of slowing growth do not result in actual contractions, and conflating the two leads to unnecessary portfolio disruption.
5. Assuming recession equals immediate market bottom. Equity markets are forward-looking and typically bottom 3 to 6 months before recessions end. Investors who wait for the “all clear” of an official NBER recession-end date often miss a significant portion of the recovery.
6. Assuming probability implies severity. A high recession probability tells you nothing about the depth or duration of the potential downturn. The 2020 recession was extremely severe (14.7% unemployment) but lasted only 2 months. The 2001 recession was mild by comparison but lasted 8 months. Probability and severity are separate dimensions of risk.
Limitations of Recession Forecasting
Recession probability models produce probabilities, not predictions. They are best used as inputs to a broader risk management framework — not as standalone trading signals. A 50% probability means the model is essentially uncertain, not that a recession is imminent.
No model has perfect accuracy. The economy is a complex adaptive system influenced by policy decisions, geopolitical events, pandemics, and technological shifts that no statistical model can fully anticipate.
Structural changes invalidate historical patterns. The economy of 2025 (services-dominant, digitally connected) operates differently than the manufacturing-heavy economy of the 1970s. Patterns that held for decades may lose predictive power as the economy evolves.
Policy responses can delay or prevent recessions. Aggressive monetary and fiscal policy — such as the Fed’s emergency rate cuts in 1998 or massive fiscal stimulus in 2020 — can alter the economic trajectory that leading indicators originally signaled.
Data revisions and dating lags add uncertainty. GDP data is revised multiple times, and NBER recession dates are determined retrospectively. Real-time data can look quite different from the final revised figures, making contemporaneous probability assessment inherently noisy.
Frequently Asked Questions
Disclaimer
This article is for educational and informational purposes only and does not constitute investment advice. Recession probability estimates are inherently uncertain and should not be used as standalone trading signals. Historical patterns may not repeat, and economic conditions change over time. Always conduct your own research and consult a qualified financial advisor before making investment decisions.
