Ratings, debt, and deficits: An exploration

What matters more for sovereign credit ratings – today’s debt or tomorrow’s deficits? This column explores how ratings weigh the two and finds that ratings place much more emphasis on current debt than on projected fiscal deficits. Compared with the implications of a simple model, ratings differ in three important ways: they overweight debt relative to forecast deficits, they understate the effects of the interest rate-growth differential, and they are heavily influenced by other country-specific factors. Countries with the same debt and deficit forecasts can receive extremely different ratings, implying widely different levels of debt needed to reach a given rating.

The explicit goal of sovereign debt ratings is to assess the probability of default on sovereign bonds. To quote from Standard & Poor’s (S&P): “[Ratings reflect] the ability and willingness to service financial obligations to non-official (commercial) creditors […] Our sovereign rating criteria incorporate the factors that we believe affect a sovereign’s willingness and ability to service its financial obligations to non-official creditors on time and in full” (S&P Global 2017).

Ratings matter for governments as they influence the premium that governments have to offer on their debt. This raises an obvious question: What weight do the rating agencies put on debt, on primary deficits, and on other factors? Put another way: How much does the past (embodied in the current debt level) matter? How much does the anticipated future (embodied in the forecast distribution of future primary balances) matter? How much do other factors matter? 

A large literature estimates the determinants of ratings (Gaillard 2012, Jinjaraket al. 2013, Ardagna 2018, Hadzi-Vaskov and Ricci 2020). The general approach is to regress ratings on several macroeconomic variables, such as GDP per capita, growth, inflation, the current account, debt, and so on. 

In a recent paper (Blanchard et al. 2026), we differ from this reduced-form approach by looking at the evidence through the lens of a structural model that is simple but yields conclusions we believe to be more general. Using the debt accumulation equation, and taking account of uncertainty, we show how current debt, forecast primary balances, and forecast interest rates and growth rates affect the distribution of debt over time. We then assume that the probability of default is an increasing function of the likelihood that debt growth exceeds a threshold over a given horizon. Through calibrations, we show how the probability of default – or, equivalently, the probability of no-default (the rating) – depends on the level of debt, the distribution of future primary balances, interest rates, and growth rates. This gives us a benchmark against which to compare our empirical estimates. 

We then examine ratings from S&P, Moody’s, and Fitch for 35 advanced economies and 97 emerging market and developing economies, annually from 1995 to 2025. We use IMF data on debt and five-year-ahead forecasts of primary balances, interest rates, and growth rates. As ratings come in ordinal categories, we use ordered probit regressions. Our baseline specification regresses ratings on debt, forecast primary balances, and year- and country-fixed effects. 

We obtain three initial findings: First, debt is a highly significant determinant of ratings. The effect of forecast primary balances is positive, but less statistically significant. Importantly, the effect of forecast primary balances relative to debt is substantially smaller than our model suggests. Second, estimated year-fixed effects are rather flat, a somewhat surprising finding given the substantial decrease in interest rate-growth differentials (“r – g”) over the sample, at least in advanced economies. Third, estimated country-fixed effects are surprisingly large. 

One plausible explanation for the weak relative weight on forecast primary balances versus debt is that our data on forecasts, which come from the IMF World Economic Outlook, are noisy proxies for the true expectations of rating agencies – an ‘errors-in-variables’ issue. Indeed, by regressing actual outcomes of primary balances on the forecasts, we find that forecasts tend to be very noisy compared to rational expectations, implying that our estimates of the effect of expected primary balances on ratings are downward biased. Correcting for these biases, we find that the gap between the implications of the model and empirical estimates narrows significantly but remains wide. In short, relative to our model, ratings appear to put too little weight on primary balances relative to debt. 

We additionally find that the flat year effects we estimate hide a more complex story: more favourable ratings over time given debt and expected primary balances, but tougher ratings on debt relative to expected primary balances. The latter effect more than offsets the favourable effect of the decrease in r – g observed in many economies in recent decades. Importantly, the effect of r – g on ratings appears much smaller than implied by our model. 

Finally, we find that country-specific effects – that is, the effect of fixed country characteristics on ratings for a given level of debt and primary balance forecast – are the central drivers of ratings. Our estimates imply that the level of debt required to achieve a given rating with a given probability varies very widely across countries. The size of these differences appears larger than could be reconciled solely with variables we have left out of our specification, such as the nature of the investors holding the debt, political dynamics, or the degree of uncertainty associated with the forecasts. 

Figure 1 illustrates this finding for advanced economies. It plots the debt-to-GDP ratio above which a country is predicted to lose an S&P rating of at least AA with 50% probability, given zero forecast primary balances and the year 2020 estimated time effect – an illustrative ‘debt tolerance’ ratio. At one end, the highest illustrative debt ratio tolerance levels are for Singapore (272%) and Germany (248%). In the middle of the range are countries such as the US (161%), France (135%), and Italy (89%). At the low end are the Baltic states, with negative ratios: even with zero debt, and with zero primary balances, the estimates imply that rating agencies would grant them a rating of AA or more with a probability of less than 50%. While the levels of the debt ratios would be different, the differences between them would be the same if we chose a different rating, a different probability, or a different year. We find similarly striking differences in required debt ratios for emerging market and developing economies.

Figure 1 Illustrative debt tolerance ratio for S&P rating (percent of GDP)

Note: Figure plots the estimated debt-to-GDP ratio above which a country is predicted to lose an S&P rating of at least AA with 50% probability, given zero forecast primary balances and the year 2020 estimated time effect.

Overall, there is a clear gap between the implications of our model and the rules implicitly used by rating agencies. It may be that our model is missing some essential aspect of how debt and deficit forecasts determine default. Alternatively, it may be that rating agencies should revisit their rules. This is why we have chosen “An Exploration” as part of the title of our paper. 

There are different ways to explore the results further. The first is to look at the same relation, but in terms of spreads, which reflect investors’ rather than rating agencies’ perceptions. In preliminary work, we find that spreads put relatively more weight on forecast primary balances relative to debt than ratings do, thus closer to the implications of our model, although a gap remains. A second way is to explore how much of the country effects can be explained by differences in political variables, the role of the central bank, and so on. A third is to look at actual defaults, and their relation to debt and primary balances. This is easier to do for emerging markets, with low average ratings and a higher proportion of actual defaults than for advanced economies with much higher ratings, and very few actual defaults. 

Source : VOXeu