Okun's Law Unemployment-Growth Relationship

What this research found

Okun's law is the rule of thumb linking how fast an economy grows to how much unemployment moves. K-Dense estimated its growth-rate form on 255 quarters of US data from 1960 to 2023, then tested whether the relationship has held steady. Across the full sample a one-point rise in quarterly real GDP growth is associated with a 0.52-point smaller change in unemployment, but splitting at 1990 reveals that the sensitivity nearly tripled between eras, from -0.259 to -0.714, a shift significant at the 1% level.

• Over 1960-Q2 to 2023-Q4, real GDP growth alone explains 59.0% of quarter-to-quarter variation in the change in unemployment, with an Okun coefficient of -0.521 and a robust standard error of 0.151.
• The relationship strengthened sharply after 1990. The coefficient moved from -0.259 in 1960-1990 to -0.714 in 1991-2023, and the fit improved from an R-squared of 0.491 to 0.736.
• That break is statistically decisive. An interaction-dummy specification puts the slope change at -0.455 with a robust standard error of 0.129, giving a t-statistic of -3.53 and a p-value of 0.0004, with a 95% confidence interval of -0.707 to -0.202 that comfortably excludes zero.
• The break-even growth rate — the pace needed just to hold unemployment steady — fell from 3.56% annualized in the earlier era to 2.46% in the later one, with delta-method confidence intervals of 2.70 to 4.42 and 1.99 to 2.94 that barely overlap. The full-sample figure is 2.94%, with an interval of 2.51 to 3.37.
• Only the slope shifted, not the level. The intercept-shift term in the break specification is marginally significant at a p-value of 0.068, so the change is driven by responsiveness rather than a jump in trend growth.
• Adding one quarter of lagged growth confirms a delayed labor-market response that is significant in every sample, and lifts the full-sample long-run coefficient to -0.608. The era gap survives this specification too, widening from -0.358 to -0.907 with non-overlapping intervals.

How it was done

The monthly civilian unemployment rate and quarterly real GDP were pulled from the Federal Reserve Economic Data repository, with unemployment averaged across the three months of each quarter to align frequencies. The quarter-over-quarter change in unemployment was regressed on non-annualized quarterly real GDP growth by ordinary least squares, using Newey-West standard errors with a Bartlett kernel and four lags so inference tolerates the serial correlation and heteroskedasticity typical of macroeconomic series. The break-even growth rate, a ratio of two estimated coefficients, was given confidence intervals via the delta method applied to the robust covariance matrix. Stability was tested with an interaction-dummy version of the Chow test that lets both slope and intercept shift at 1991, and the results were checked against a distributed-lag specification adding the previous quarter's growth.

Data sources

• Federal Reserve Economic Data — UNRATE civilian unemployment rate, US Bureau of Labor Statistics, monthly, accessed 2026-07-10
• Federal Reserve Economic Data — GDPC1 real gross domestic product in chained 2017 dollars, US Bureau of Economic Analysis, quarterly
• Okun, Potential GNP: Its Measurement and Significance, American Statistical Association Proceedings (1962)
• Ball, Leigh & Loungani, Okun's Law: Fit at 50?, Journal of Money, Credit and Banking 49:1413 (2017)
• Foroni & Furlanetto, Explaining Deviations from Okun's Law, ECB Working Paper 2669 (2022)
• Newey & West, Econometrica 55:703 (1987)