As global debt continues to climb, evaluating debt risks is critical. This column introduces a novel ‘debt-at-risk’ framework that uses information on current macro-financial and political conditions to construct the full distribution of potential future debt outcomes. The framework shows that debt risks are high, rising, and tilted to the upside. Debt levels are projected to reach 117% of GDP by 2027 in a severe adverse scenario, nearly 20 percentage points above projections. Moreover, debt-at-risk is a key predictor of oncoming fiscal distress. The framework can flexibly accommodate additional debt drivers and has broad country coverage, making it a valuable tool for policymakers and researchers.
Global public debt is once again at the forefront of the policy debate (Larch 2024). It surged past $100 trillion in 2025 and could hit 100% of global GDP by 2030. Add in rising geopolitical tensions, ageing populations, defence spending, and tighter financial conditions, and the debt outlook could be even more worrisome. High debt levels constrain fiscal space, raise borrowing costs, heighten inflation expectations (Brandao-Marques et al. 2023), and leave countries more exposed to shocks (Brunnermeier and Reis 2023, Mitchener and Trebesch 2023, Kose et al. 2021, Kenny et al. 2021). Moreover, when debt levels spike unexpectedly, the economic consequences can be severe. Given this context, understanding not just where debt is heading, but the risks around that path, is essential.
In Furceri et al. (2025), we introduce a new framework to assess these risks: the ‘debt-at-risk’. Building on concepts from the growth-at-risk literature (Adrian et al. 2019), debt-at-risk constructs the full distribution of potential future debt outcomes. Thus, it allows us to quantify how high public debt could rise in an adverse scenario and to identify the sources of debt risks.
The approach offers a richer picture than traditional forecasts, which typically provide a single point estimate for a country and year. Instead of focusing only on the usual economic drivers of debt such as economic growth, our approach incorporates underlying financial and political factors — such as financial stress episodes or increased policy uncertainty — that could shape these drivers. The framework also estimates how these conditioning variables influence the expected level of debt and its dispersion. Ultimately, it lets policymakers compare the evolution of debt risks across countries and over time.
The empirical framework
Our empirical strategy is based on a location-scale model, which allows us to estimate multiple quantiles of the future debt distribution. The location parameter captures the average shift in debt outcomes associated with changes in a conditioning variable, while the scale parameter tells us how the variable affects the uncertainty (variance) around the debt distribution. For instance, if a variable capturing financial stress episodes has both a positive location and scale parameter, then an increase in financial stress increases both the average level and the uncertainty around forecasted debt levels. As discussed in Adrian et al. (2019), when the scale parameter is different from zero, the model features asymmetric risk. Specifically, if the scale and location parameters have the same (opposite) sign, then the mean and the variance of the predictive distribution are positively (negatively) correlated implying that upside (downside) risks dominate.
Our baseline specification uses annual data from 1980 to 2024 for a panel of 90 economies, covering more than 90% of global government debt. We estimate up to five-year-ahead conditional quantiles of the debt-to-GDP ratio using a set of financial, political, and economic variables of interest. 1 We then construct full probability distributions for each country using a skewed t-distribution fitted to the estimated quantiles. These are pooled into a single country-specific debt-at-risk distribution using weights based on the predictive power of each variable (following Crump et al. 2023). This approach allows us to flexibly incorporate different drivers across countries without overfitting the model. Debt-at-risk is defined as the 95th percentile of a country or country group’s predicted debt-to-GDP ratio. This tail measure captures how high debt could plausibly rise in an adverse scenario.
Drivers and size of debt risks
Our results show that unfavourable financial developments significantly raise debt-at-risk — particularly in the short term. These effects are nonlinear and larger in the upper tail of the distribution. For instance, tighter financial conditions are associated with a statistically significant increase in both the mean (location) and variance (scale) of the predicted debt distribution, resulting in larger changes in the upper debt quantiles compared to the lower ones (Figure 1). Beyond financial variables, episodes of social unrest and rising policy uncertainty are associated with upward shifts in the debt distribution. Economic fundamentals also matter, and their effects persist. Stronger primary balances are associated with lower debt across the distribution. Meanwhile, weak growth and high initial debt levels push up debt-at-risk in a sustained and asymmetric manner up to a five-year forecast horizon.
Figure 1 Financial conditions and debt-at-risk
Notes: Bars denote estimated coefficients, and whiskers in bars show the 90% confidence interval. Panel A plots the location and scale coefficients for the Financial Conditions Index based on the location-scale model. The dependent variable is the three-year-ahead debt-to-GDP ratio. The independent variables include the IMF’s Financial Conditions Index (standardised to have a mean of zero and a standard deviation of one) and initial debt. Panel B plots the quantile regression coefficients corresponding to the location-scale coefficients.
We aggregate debt-at-risk across countries to examine the full distribution of global public debt. Our estimate for global debt-at-risk in 2027 is 117% of GDP, around 20 percentage points higher than the IMF’s projections (Figure 2). Debt risks have been rising steadily over time, largely in tandem with the increase in actual debt. But risks spike more sharply than debt itself during major global shocks, such as the Global Crisis and the COVID-19 pandemic. These episodes further highlight the nonlinear and asymmetric effects of the conditioning variables (such as financial stress and growth) on the upper tail of the debt distribution.
Figure 2 Global debt-at-risk 2027
(Probability density of three-year-ahead government debt-to-GDP ratio)
Notes: The probability density functions are estimated using panel quantile regressions of the debt-to-GDP ratio on various political, economic, and financial variables. The global sample comprises 47 countries for which data on the conditioning variables are available. The quantile estimates are fitted to a skewed t-distribution. Dots indicate the predicted 5th, 50th (median), and 95th quantiles of the debt-to-GDP ratio.
Debt-at-risk also varies widely across country groups. In advanced economies, debt-at-risk has eased since the pandemic and now stands at around 131% of GDP, close to its average over the last 15 years. In emerging market and developing economies, the level of debt-at-risk is lower at around 96% but has been rising steadily over the past decade. The dominant drivers differ as well: financial conditions play a larger role in advanced economies, while political and policy uncertainty are more influential in emerging markets.
Three extensions: Predictive power for fiscal crises, global coverage, and nonlinear effects
We assess the usefulness of debt-at-risk in predicting fiscal crises and find that it outperforms other common economic variables as a leading indicator of oncoming fiscal distress. Using both Bayesian Model Averaging and machine learning techniques, we find that debt-at-risk is robustly correlated with future fiscal crisis episodes and consistently outperforms other variables — such as debt levels, growth, or spreads — as a predictor of fiscal distress (Figure 3).
We also extend the analysis to 175 countries, leveraging the subset of conditioning variables available for each. Even with reduced data, the model continues to capture key asymmetries and identifies significant upside risks.
Finally, we explore nonlinear interactions, finding that high initial debt levels magnify the effects of negative shocks. For example, the impact of a growth slowdown on debt-at-risk is much larger when a country starts from an already elevated debt level. This result highlights the compounding nature of debt vulnerabilities in a low-growth global economy.
Figure 3 Variable importance by group of predictors
(random forest model predicting a fiscal crisis)
Notes: The figure displays (grouped) variable importances from a random forest model used to predict a fiscal crisis. The independent variables are grouped into the categories shown on the horizontal axis. Higher values indicate that a variable has a higher predictive power. Variable importance is calculated using a predictor importance function in R using a random forest with the variables selected by Boruta.
Potential applications
Debt-at-risk provides a tractable way to evaluate debt risks and has several potential applications. The framework can be flexibly expanded to include other conditioning variables or explore how debt risks could evolve under different macro-financial scenarios—for example, during a recession or credit crunch. With coverage for nearly every economy in the world, the framework is a useful tool for cross-country empirical research.
Source : VOXeu
