Reflections on Reinhart and Rogoff I: Why Time Series Aren’t Enough
After crunching Reinhart and Rogoff’s data, we’ve concluded that high debt does not slow growth. That was the conclusion of a joint Quartz article with Miles Kimball in which we looked at post war data on sovereign debt and growth in advanced economies. Miles’ recent foray back into the debate made me reflect upon our previous work together.
Upon further reflection, I’m not sure I still believe in that title. Not because I now believe that debt to GDP ratios above 90% slow growth, but rather because I’m unsure if long time series provide meaningful information about the effect of debt on growth at all. Given the rarity of high debt episodes, time series data is either too sparse or takes data from drastically different policy regimes to be reliable.
Below is a plot of the overall paths of debt/GDP ratios in countries that passed the 90% threshold after winding down war debts. There are only 5 countries: Belgium, Greece, Ireland, Italy, and Japan (Note that the US is not included as the sample ends in 2009).
Therefore the sentence “whenever a country has hit 90% of debt to GDP” can easily be replaced by the phrase “when these handful of countries hit 90% of GDP in these specific years”.
This does not bode well for inference. One could look at each of these countries and provide a highly idiosyncratic story for why debt blew up and growth lagged. Japan was a story of fiscal policy in a liquidity trap. Greece was a story of “macroeconomic populism” in a time of stagflation. You might have specific beliefs about the governments in each of these countries, above and beyond their attitudes towards debt. These idiosyncratic components matter a lot, and as such there is no power (in the statistical sense) to detect a negative association between debt and growth. Indeed, growth was all over the place. Ireland boomed, Japan stalled, and the others were in the middle.
The fact that debt is very slow moving also means that when you compare different levels of debt and subsequent growth rates, you’re looking at very different time periods. Ashok states this forcefully after my article with Miles:
This means comparisons on different levels of debt take place on very different economic structures. When we talk about the America with 30% debt-to-annual-GDP (the first bin in R-R), we’re talking about an America before modern floating exchange rates, strong unions, extremely high taxes, and declining inequality — each of which adjusts the causal mechanism in question.
This problems caused by looking at such diverse time periods get worse if you try to look at even older data. To deal with the rarity of high debt episodes, Reinhart and Rogoff look all the way back to the 1800's. But macroeconomic institutions were very different back then! For starters, countercyclical fiscal policy and fiat currencies were not nearly as prominent. Yet these are two critical issues that would change the way debt affects predictions of growth.
In a world of countercyclical fiscal policy (i.e. stimulus packages), rapid onsets of debt coincide with policies that induce more growth in the future. If instead debt is not associated with fiscal policy, then increases in debt are likely from exogenous factors that are worse for future growth. Recent experience with the Eurozone also suggests that having your own currency has tremendous effects on whether high debt levels are associated with financial distress. If there’s no monetary authority to backstop the debt, then increases in debt truly might have large negative effects on growth.
So if you have a strong prior that fiscal and monetary policies matter for the relationship between debt and growth (as I do), then thousands of observations from the Gold Standard era are nonetheless uninformative about how much current observed levels of debt matter for future growth.
But the entire question of causality remains unresolved! If governments are incompetent along many dimensions, debt being only one, then the association between debt and growth might just be a matter of generally incompetent governance. Without a proper instrument for debt, these long time series say nothing about how much high debt levels actually hurt growth. The time series evidence provides no guidance on how to weigh the cost of debt against the benefits of countercyclical fiscal policies.
So is there any way out? I think the answer has to lie not with bigger data sets, but rather theory and smaller case studies that help us shed light on potential mechanisms between debt and growth. For example, Krugman emphasized in his 2013 IMF seminar that the effect of debt depends crucially on whether the sovereign is borrowing in its own currency. Externally denominated debt puts a constraint on monetary policy. Therefore the effects of debt on growth might depend on whether monetary policy is still able to operate. This played out dramatically in the 1997 East Asian debt crisis. Crises near the end of the interwar Gold Standard also highlighted what can go wrong when central banks are forced to allow deflation that raises the real value of debts.
These examples suggest that debt buildups in countries such as the US or the UK are unlikely to pose major risks in the short run while debt buildups in countries without independent monetary policies — such as Greece, Italy, or Spain — are. With a more in depth analysis of theory and mechanisms, more recent examples from modern economic regimes can provide better guidance for policy.