TL;DR
This paper challenges the widely cited "90% debt-to-GDP threshold" for economic growth, finding instead that median GDP growth drops significantly when debt exceeds approximately 30%. Using a novel econometric approach on the Reinhart-Rogoff dataset, the authors formally test for this threshold effect, providing statistically significant evidence for a lower threshold.
What Problem Does This Paper Address?
A central debate in macroeconomics and fiscal policy revolves around the relationship between government debt and economic growth. The influential 2010 paper by Reinhart and Rogoff (RR) suggested a critical threshold of around 90% of GDP, beyond which public debt is associated with substantially lower economic growth. This finding has heavily influenced policy discussions. However, subsequent research, including critiques pointing to potential data errors, has questioned the robustness of this 90% threshold. This paper aims to rigorously test for the existence and location of a debt threshold effect on median GDP growth, specifically addressing the RR debate and contributing a formal statistical test for such effects when the threshold is unknown.
Methods
The authors employ a recently developed econometric technique to test for threshold effects in median regression models. This approach is particularly suited for situations where the threshold value is not known a priori. They specify a conditional median function that allows for a potential change in the relationship between debt and growth at an unknown threshold ($\gamma$). They consider two models: an "intercept-only" model where only the intercept changes at the threshold, and an "intercept-and-slope" model where both the intercept and the slope can change.
The core of their testing procedure involves a sup-likelihood-ratio (QLR) statistic. This statistic is designed to detect a threshold effect by comparing the maximized objective function of the unrestricted model (allowing for a threshold) to that of the restricted model (assuming no threshold). They simulate p-values to assess the statistical significance of the detected threshold. The method is applied to both annual and five-year forward average GDP growth rates.
Datasets
The primary dataset used is the updated Reinhart and Rogoff (RR) dataset, specifically a revised and corrected version of the data used in their 2010 paper. This dataset comprises post-war data for advanced economies, including countries like Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, the United Kingdom, and the United States. The authors use country-year equal weights for their median regression estimations.
Key Insights and Ideas
- The 90% Threshold is Questionable: The paper's most striking finding is the lack of robust evidence for a debt threshold around 90% of GDP. This directly challenges a cornerstone of the RR debate.
- A Lower Threshold Emerges: Instead, the analysis strongly suggests a significant threshold effect at a much lower debt-to-GDP ratio, around 30%.
- Median vs. Mean Matters: By focusing on median growth rather than mean growth (as in some previous studies), the authors are able to identify a more robust threshold. Median regression is less sensitive to outliers, which may have influenced earlier findings.
- Abrupt Growth Drop: Countries with debt-to-GDP ratios exceeding this 30% threshold experience a notable and statistically significant decline in median GDP growth.
- Causality vs. Description: While the paper provides strong statistical evidence for a threshold, the authors acknowledge that their findings are descriptive and do not establish causality.
Key Results
- Annual GDP Growth:
- The null hypothesis of no threshold effect is rejected at the 5% significance level for both intercept-only and intercept-and-slope models.
- The estimated debt threshold is 28%.
- Countries with debt above 28% experience a drop in median annual GDP growth of approximately 1 percentage point (intercept-only model) or 0.8 percentage point (intercept-and-slope model).
- Five-Year Forward Average GDP Growth:
- The null hypothesis of no threshold effect is also rejected for this measure.
- The estimated threshold is 32% for the intercept-only model and 18% for the intercept-and-slope model.
- The estimated coefficients suggest that the negative impact of debt on future growth is smaller than on contemporaneous growth.
- Robustness Checks: The results remain largely consistent across various robustness checks, including omitting countries and considering different model specifications.
Limitations and Future Work
The authors explicitly state several limitations:
- Descriptive Nature: The study is primarily descriptive, and the findings do not establish causality between debt levels and growth.
- Country-Specific Heterogeneity: The analysis does not fully account for country-specific variations in economic conditions, such as inflation and interest rates, which could influence the debt-growth relationship.
They suggest future research directions:
- Causality: Investigating the causal link using panel data and structural econometric models guided by economic theory.
- Country-Specific Analysis: Allowing for country-specific heterogeneity and exploring how macroeconomic variables mediate the debt-growth relationship.
- Economic Channels: Further research into the specific economic channels through which public debt affects growth.
Summary updated