A parsimonious approach to GDP and unemployment prediction
This paper by Daniel H. Cooper, Giovanni P. Olivei, and Hannah Rhodenhiser presents a parsimonious forecasting framework for U.S. GDP growth and unemployment rate based on just four fundamental economic variables. The authors use a two-equation system that is estimated jointly via full information maximum likelihood:
- Unemployment rate equation: Forecasts changes in unemployment using the small information set
- GDP growth equation: Links GDP growth to unemployment changes through Okun’s law in first differences
The framework uses only four variables as of time t-1:
- Unemployment rate (including both quarterly average and first month of current quarter)
- Consumer sentiment (University of Michigan Index)
- Federal funds rate deviation from estimated neutral rate
- Corporate credit spread (Gilchrist-Zakrajšek spread)
Estimation approach
- Expanding window estimation starting from 1966:Q1
- Multi-step direct forecasting for horizons 2-8 quarters ahead
- Joint estimation of both equations with cross-equation restrictions
- Okun’s law constraint imposed on the relationship between GDP and unemployment
Forecast Accuracy Performance
The simple framework performs comparably or better than professional forecasts over 1984-2019:
- Unemployment rate forecasts: Generally more accurate than Survey of Professional Forecasters (SPF), Blue Chip (BC), and often Federal Reserve (GB/TB) forecasts
- GDP growth forecasts: Competitive performance, especially at policy-relevant 4-6 quarter horizons
Critical methodological insights
- Order of estimation matters: Forecasting unemployment first, then GDP via Okun’s law, yields much better results than the reverse approach
- Okun’s law constraint essential: Without this relationship, forecasting GDP with a small information set is challenging
- Joint estimation superiority: Maximum likelihood estimation of both equations together outperforms separate OLS estimation
Forecasting U.S. Economic Activity with a Small Information Set
Authors: Daniel H. Cooper, Giovanni P. Olivei, Hannah Rhodenhiser
From: Federal Reserve Bank of Boston, Boston University
Nowcasting/early detection methodology for recession timing detection
The SOS indicator, developed by John O’Trakoun and Adam Scavette in this paper, focuses narrowly on recession timing detection using a simple binary signal – when to declare that a recession has begun. In contrast, the paper by Cooper et al., above, aims for continuous quantitative prediction of GDP growth and unemployment rates over 2-8 quarter horizons, providing specific numerical forecasts rather than binary recession signals.
Data source and construction
- Base data: Uses the insured unemployment rate from weekly unemployment insurance claims data compiled by the Department of Labor from state and territorial unemployment insurance programs
- Calculation: The SOS indicator signals a recession when the 26-week moving average of the insured unemployment rate rises more than 0.2 percentage points relative to its minimum over the 52 preceding weeks
- Frequency: Available weekly rather than monthly, providing higher temporal resolution
Perfect accuracy with no false positives
The SOS indicator demonstrates exceptional accuracy in identifying recessions:
- Correctly identified all seven recessions since 1971 (November 1973, January 1980, July 1981, July 1990, March 2001, December 2007, February 2020)
- Zero false positives over the 1971-2024 sample period, compared to the Sahm rule which had false alarms in 2003 and 2024
- SOS signals recessions 2.3 months after official NBER start dates, compared to 3.4 months for the Sahm rule
The SOS indicator offers significant advantages for policy applications, particularly for automatic stabilizer activation and real-time economic monitoring.
SOS! Signaling Recessions Earlier
Authors: John O’Trakoun, Adam Scavette
From: Federal Reserve Bank of Richmond
Why recession probability uncertainty is lower now
According to the New York Fed’s analysis, uncertainty about the current recession probability is below the historical average. This reduced uncertainty stems primarily from the stability and consistency of recent economic indicators used in their recession forecasting model. The model, developed by researchers Richard Crump and Nikolay Gospodinov, relies on yield curve data and other macroeconomic variables that have shown less volatility in recent months compared to historical patterns.
Key factors contributing to lower uncertainty include:
- Improved model calibration: Recent data refinements have enhanced the model’s precision in estimating recession probabilities.
- Stabilized yield curve behavior: The Treasury yield spread (10-year vs. 3-month) has exhibited less erratic movement after the steep inversion of 2023.
- Consistent economic signals: Fewer conflicting signals from employment, inflation, and consumer data have reduced model ambiguity.
How Uncertain Is the Estimated Probability of a Future Recession?
Authors: Richard Crump, Nikolay Gospodinov
From: Federal Reserve Bank of New York
