Methodology

This page explains in detail how recession.fyi calculates the recession probability score you see on the main dashboard.

Overview

Our recession probability model is a weighted indicator system that combines 13 economic indicators, each contributing according to its historical predictive power. The model outputs a single probability score (0-100%) representing the likelihood of a US recession occurring within the next 12 months.

Step 1: Data Collection

All economic data is sourced from the Federal Reserve Economic Data (FRED) API, maintained by the Federal Reserve Bank of St. Louis. Data is collected automatically 3 times per day during market hours (weekdays only):

7:00 AM ET - Morning collection
2:00 PM ET - Afternoon collection
6:00 PM ET - Evening collection

Step 2: Risk Scoring (0-100)

Each indicator is assigned a risk score from 0 (no recession risk) to 100 (extreme recession risk) based on historical thresholds that have preceded past recessions.

Example: Yield Curve Risk Scoring

if spread < -0.5%: risk_score = 90 (Strong inversion) if spread < -0.2%: risk_score = 70 (Moderate inversion) if spread < 0.2%: risk_score = 40 (Flattening) else: risk_score = 10 (Normal)

Each indicator has its own thresholds based on historical recession patterns and economic research.

Step 3: Weighted Probability Calculation

The final probability is calculated as a weighted sum of all indicator risk scores:

Probability = Σ (indicator_risk_score × indicator_weight) Where weights sum to 1.0 (100%)

Indicator Weights

Weights are based on academic research and historical recession analysis. Higher weights are given to indicators with stronger historical predictive power.

Indicator	Weight	Rationale
Core Indicators (60%)
Yield Curve (10Y-2Y)	20%	Preceded every recession since 1970
Leading Economic Index	18%	Composite of 10 forward-looking indicators
Unemployment Claims	12%	Early labor market signal
Manufacturing PMI	10%	Real economy production activity
Housing Market (15%)
Case-Shiller Index	6%	Home price health
Housing Starts	5%	Construction demand
30-Year Mortgage Rate	4%	Housing affordability
Credit Markets (12%)
High Yield Spread	7%	Credit stress indicator
Investment Grade Spread	5%	Corporate borrowing costs
Supporting Indicators (13%)
VIX (Market Volatility)	7%	Investor fear gauge
Consumer Confidence	6%	Spending sentiment
TOTAL	100%

Step 4: Status Determination

The final probability score is categorized into risk levels:

0-30%: LOW RISK (Green) - Economy appears healthy
31-50%: MODERATE RISK (Yellow) - Warning signals present
51-70%: HIGH RISK (Orange) - Multiple recession indicators
71-100%: VERY HIGH RISK (Red) - Recession highly probable

Example Calculation

Here's a real example from our current data:

Yield Curve: 10 × 0.20 = 2.0 Leading Index: 20 × 0.18 = 3.6 Unemployment Claims: 20 × 0.12 = 2.4 Manufacturing PMI: 15 × 0.10 = 1.5 Case-Shiller: 40 × 0.06 = 2.4 Housing Starts: 35 × 0.05 = 1.75 Mortgage Rates: 35 × 0.04 = 1.4 High Yield Spread: 25 × 0.07 = 1.75 IG Spread: 20 × 0.05 = 1.0 VIX: 25 × 0.07 = 1.75 Consumer Confidence: 85 × 0.06 = 5.1 Total: 24.65% → Rounded to 25% (LOW RISK)

Historical Accuracy

Our model is designed based on indicators that have historically preceded recessions:

Yield curve inversions preceded all 7 recessions since 1970
LEI declines of 6+ months preceded 80% of recessions
Housing downturns preceded the 2001 and 2008 recessions
Credit spread widening signals liquidity stress before downturns

Important Note: Past performance does not guarantee future results. This model can produce:

False positives: High probability without actual recession (example: 1998)
False negatives: Sudden recessions with little warning (rare but possible)
Timing uncertainty: Signal may precede recession by 6-18 months

Model Limitations

Users should be aware of these limitations:

Data Lag: Some indicators (housing, trade) have 1-2 month reporting delays
Structural Changes: Economy evolves; past relationships may not hold
Policy Response: Federal Reserve actions can alter recession trajectories
External Shocks: Black swan events (pandemics, wars) aren't predictable
No AI/ML: Current model uses rules-based logic, not machine learning (planned for future)

Comparison to Other Models

How recession.fyi compares to other recession forecasting approaches:

vs. NBER: NBER declares recessions retroactively (6-18 months after start). We provide real-time probability.
vs. Treasury Yield Curve: We consider 12 additional indicators beyond just yield curve.
vs. Economist Surveys: Our model is rules-based and transparent, not opinion-based.
vs. ML Models: Simpler, more interpretable. Future versions will incorporate ML.

Future Improvements

Planned enhancements to the model:

Machine learning trained on 50+ years of recession data
Sentiment analysis on economic news
6-month forward probability projections
Sector-specific recession probabilities
International indicators (ECB rates, global trade)

Transparency

We believe in full transparency. Our methodology page clearly shows:

All risk scoring thresholds for each indicator
Complete weight distribution across indicators
Step-by-step calculation process
Data sources and update frequency

Questions?

If you have questions about our methodology or spot issues, please reach out through our contact channels listed on the About page.