Skip to content

Fun Fact: The Excel Error Behind an Austerity Argument

Abstract

In 2010 two Harvard economists published a paper showing that government debt above 90% of GDP dragged economic growth to -0.1% a year. Politicians on both sides of the Atlantic cited it to justify austerity budgets. In 2013 a graduate student trying to reproduce the result for a class assignment got the authors’ actual Excel file and found that an AVERAGE formula covered only 15 of the 20 countries: five rows had been left out of the selection. Correcting that error, together with two questionable data choices it sat beside, turned the -0.1% into +2.2%. A spreadsheet cell reference had helped underwrite economic policy for three years.

The Paper

Carmen Reinhart and Kenneth Rogoff published “Growth in a Time of Debt” in 2010 in the American Economic Review Papers and Proceedings, a non-peer-reviewed section. Its headline finding was blunt and quotable: countries whose public debt exceeded 90% of GDP saw average real growth of about -0.1%, against healthy positive growth below that line. In the aftermath of the 2008 financial crisis, that 90% threshold became a political weapon. US Congressman Paul Ryan’s budget cited the paper as evidence for spending cuts; EU economics commissioner Olli Rehn invoked the same number to press European fiscal restraint.

The Student

In 2013 Thomas Herndon, a graduate student at the University of Massachusetts Amherst, was assigned to replicate a published economics paper. He picked Reinhart and Rogoff’s and could not reproduce their numbers. He asked the authors for their working spreadsheet, and they sent it. Opening it, Herndon found three problems, one of them a plain software mistake:

  • The coding error. The AVERAGE formula computing the mean growth of high-debt countries selected a range that stopped five rows short, silently dropping Australia, Austria, Belgium, Canada, and Denmark. Fifteen of the twenty countries went into the number the world had been quoting.
  • Selective exclusions. Several years of high-debt, high-growth data for Australia, Canada, and New Zealand were left out.
  • Unusual weighting. Each country counted equally regardless of how many years it contributed, so one bad year for one country could outweigh two decades of another.

Fix all three and the striking -0.1% collapse turns into ordinary growth of roughly 2.2%. The cliff at 90% was not in the data. Herndon, with advisors Michael Ash and Robert Pollin, published the correction in April 2013. Reinhart and Rogoff conceded the coding error while disputing that their other choices were improper and defending their broader thesis.

The episode is the most consequential entry in a long list of spreadsheet mistakes with real-world cost, alongside JP Morgan’s 2012 trading loss and the 2020 loss of nearly 16,000 English COVID-19 test results to a row limit. See The Spreadsheet Revolution for why the tool that democratized modeling also globalized its errors.

📚 Sources