The twin deficits hypothesis says that when a government borrows more, the country's current account worsens too. If it holds, the current account should have a positive slope on the fiscal balance.
This page tests that for the United States, Germany and Japan using IMF and BEA annual data (% of GDP, through 2024). Hover any chart to read exact values.
Fiscal balance (government net lending) and current account balance, % of GDP. Below zero means a deficit. Hover to compare both lines in any year.
Germany starts in 1991 (reunification). U.S. fiscal series: BEA total government, Germany and Japan: IMF WEO general government.
Each dot is one year. The line is the least-squares fit of current account on fiscal balance. An upward line is consistent with twin deficits. Hover a dot to see the year.
Fitted slopes here are plain OLS and match the regression table. Significance uses Newey-West standard errors (lag 2).
Rolling correlation between fiscal balance and current account, computed over the years ending in each point. Positive values agree with twin deficits. Drag the slider to change the window.
With roughly 10 observations per window, correlations swing widely and neighbouring windows overlap heavily. Read this as descriptive, not causal.
Slope of the current account on the fiscal balance, with 95% confidence intervals from Newey-West standard errors. An interval that crosses zero means the slope is not statistically different from zero.
Excl. COVID drops 2020–2021. COVID dummy adds an indicator for those years. PO = Phillips-Ouliaris cointegration test p-value (reported between 0.01 and 0.15, so 0.15 means 0.15 or above).
Slopes are positive for Germany and Japan, but zero or negative for the U.S. in every specification.
Germany's strong levels slope (about 0.97) disappears in first differences. Both series trend over 2000–2018, which inflates the levels result.
Japan is the only country with a significant first-difference slope (0.11, 2001–2024), and it fades when COVID years are dropped.
Unit-root tests mostly fail to reject, and no cointegration test has a p-value below 0.11, so first differences are the safer test.