Above 90 percent debt, growth is +2.0%, not -0.1%: the correction reproduces, the original does not

The spec, re-run

The panel is JST Release 6: 2,718 country-year rows, 18 advanced economies, 1870 to 2020, with 2,489 public debt/GDP observations (stored as a ratio, here multiplied by 100) and 2,666 real GDP per capita observations from Maddison. Growth is the year-on-year change in real GDP per capita, computed only across consecutive years (a guard found 0 gaps in the year sequence). Country-years without a debt observation, 229 of them, drop rather than defaulting into the lowest bucket; Ireland is missing 52 of its 151 years. The buckets, the 1946–2009 window and both weighting schemes are the papers’ choices: pooled country-year means are HAP’s reading, means of country means are RR’s. Buckets close on the right, so a ratio exactly at a threshold lands in the lower bucket; 1 observation (DEU-1980) sits on a boundary.

Source: Jorda-Schularick-Taylor Macrohistory Database, Release 6 (Jorda, Schularick & Taylor 2017), retrieved 2026-07-09. Computed by scripts/build/build_research_debt_and_growth.py. Mean annual growth, percent, 1946-2009. SEs are pooled standard errors treating country-years as independent, which the limitations section qualifies.

The primary variable is JST’s real GDP per capita; RR’s was total real GDP growth. The one pre-declared robustness variant rebuilds total real GDP as per capita GDP times population, and it moves every bucket up by 0.57 to 0.87 points without changing the shape: above 90 percent debt it gives +2.57% pooled (SE 0.64) and +2.40% RR-weighted, which lands 0.20 to 0.37 points from HAP’s published 2.2 percent. No other specification was run.

No cliff at the threshold

RR’s claim was not just a low mean but a break at 90. Splitting the same 1,132 postwar country-years into 10-point debt bins (our presentation choice; bins close in ascending order until the first holds fewer than 10 observations, and everything above that edge pools into an open top bin) shows a downward drift and nothing special at the threshold: mean growth is +2.23% in the bin just below 90 and +2.21% just above. 1 of the 14 bin means falls below zero, the lowest at 110–120 percent of GDP with -0.02% (SE 1.33, n=19), and the open 130+ bin, where postwar debt ratios run up to 269.8 percent, averages +3.55% across 33 country-years.

Inside the 90+ cell

The postwar 90+ cell holds 120 country-years from 9 economies (AUS BEL CAN GBR IRL ITA JPN NLD USA), and it is anything but uniform: single years range from -21.5% (USA 1946) to +65.9% (NLD 1946), and the pooled standard deviation is 6.80 points. 2 of the 9 country means are negative (AUS, USA), drawn entirely from years between 1946 and 1950, and RR’s weighting gives each the same vote as Belgium’s 29 years. Even so, both weighting schemes land positive, 0.40 points apart in this cell.

Source: Jorda-Schularick-Taylor Macrohistory Database, Release 6 (Jorda, Schularick & Taylor 2017), retrieved 2026-07-09. Computed by scripts/build/build_research_debt_and_growth.py. Real GDP per capita growth in country-years with debt/GDP above 90 percent, 1946-2009. The year span runs first to last such year; for 3 of the 9 countries that span contains years not above 90.

Why RR’s -0.1% cannot appear here

A clean refutation would need RR’s exact data, and this is not it, in three specific ways. First, New Zealand: HAP’s account of RR’s spreadsheet has early postwar New Zealand years excluded, leaving a single 1951 observation of about -7.6 percent (HAP’s figure) that RR’s equal-country weighting then amplified into the negative headline mean. JST carries no New Zealand at all, so the pivotal observation structurally cannot operate here, and this note could not have reproduced RR’s number even if RR were right. Second, the spreadsheet error HAP documented, which dropped five countries from RR’s average, is not a property of the data: a correct computation cannot re-create it. Third, vintage: JST’s Release 6 carries revised Maddison GDP and revised historical public debt series against RR’s circa-2009 compilation of central government debt. What this panel can say is narrower and still worth saying: on 17 of RR’s 20 economies (missing AUT GRC NZL, adding CHE), over the papers’ own window, applying RR’s own weighting, the 90+ mean is +1.60% per capita and +2.40% for total GDP, nowhere near -0.1 percent, while HAP’s corrected 2.2 percent is within 0.20 to 0.37 points of the closest variable.

The full 1870–2020 span, not just the postwar window

The 1946 and 2009 endpoints are the papers’ sample choice, not something this data dictates. Running the same buckets over the panel’s whole 1870–2020 span gives the same picture with lower levels: a gentle decline across buckets and a positive 90+ mean under both weightings.

Source: Jorda-Schularick-Taylor Macrohistory Database, Release 6 (Jorda, Schularick & Taylor 2017), retrieved 2026-07-09. Computed by scripts/build/build_research_debt_and_growth.py. Mean annual growth, percent, 1870-2020, all rows with both a debt and a growth observation.

What this cannot tell you

• Which way causation runs. Bucket means are descriptive. Slow growth inflates a debt ratio through both its numerator and its denominator, and that reverse channel is the standard critique of both papers. Nothing on this page adjudicates it.
• What RR’s own dataset shows. This panel lacks 3 of RR’s 20 economies and carries revised GDP and debt series. HAP’s audit of the original spreadsheet is the source for what RR’s own file contains; this note only tests whether the published pattern survives in current data of the same kind.
• More precision than 9 countries can give. The pooled SE of 0.62 treats 120 country-years as independent draws; they are not (Belgium contributes 29 of them, between 1946 and 2009). Grouping at the country level leaves 9 units, and the two weighting schemes already move the cell mean by 0.40 points. Read the 90+ estimates as roughly two percent, not as 2.00 exactly.
• Anything at debt levels this panel does not visit. Debt/GDP here spans 1.9 to 269.8 percent, in advanced economies only. Emerging markets, and the 229 country-years with no debt observation, are outside every growth statistic above.
• Where a growth break would sit, if one exists. The 90 percent threshold is imposed from the papers, not estimated. A bucket design can only confirm or deny a jump at the thresholds it is given; the bin profile is a finer look at the same question, not a threshold search.