The New Consensus

Brad DeLong sends us to a fantastic paper from University of Massachusetts-Amherst economist Arindrajit Dube. This, in the wake of a powerful essay from Miles Kimball and (undergraduate!) Yichuan Wang, we learn two things:

  • Quartz publishes important empirical findings. (!)
  • There is not “even a shred of evidence […] for a negative effect of government debt on growth”.

We are approaching a new consensus, among serious people and clowns alike. The narrative – whether Keynesian or not – for the past two years has been a reluctant understanding that debt can have a adverse effect on level of GDP. The argument for immediate stimulus derived from the idea of hysteresis and liquidity traps rather than a rejection of long-term consequences from deficits, per se. Indeed, in my most popular blog post, I maintain that “It is of unimpeachable importance that the USA pay down its debt in the long-run.” I’m still wedded to this idea, but I would replace “unimpeachable” with “probable”. I am less confident that debt, even by itself, negatively effects growth. Indeed, in the long-run, we must all be agnostic. From where does this new consensus emerge?

Enter Dube. The central argument submitted by Keynesian critics such as Paul Krugman regarding the Reinhart-Rogoff result was the chance, indeed likelihood, of reverse causality. That is, R-R all but explicitly argued that high levels of debt are causally-linked with lower growth rates. The theoretical provisions included:

  • The idea that as the debt-to-annual-GDP ratio increases, interest payments will eat a greater portion of national output, and hence constrain investment. This is an argument I’ve empirically-questioned before.
  • The idea that as the debt-to-annual-GDP ratio increases, bond markets will perceive a positive feedback loop between further deficits and growing interest rates and will hence be unwilling to lend at any but exorbitant yields. The empirical doubts here are self-evident.

It is curious that no one until now has so thoroughly thrashed this idea, empirically, as Dube has just done. His insight was to delineate the empirical presentation of the two causal levers: (A) higher debt causes lower GDP and (B) lower GDP causes higher debt. The mathematical tautology that RR casually ignore is that, as Brad DeLong and Larry Summers to not tire of noting, ratios must have both a numerator (debt) and denominator (GDP).

Dube suggests, rather convincingly, that if (A) is true, current debt levels will be correlated with future growth rates and that if (B) is true, current debt levels will be correlated with past growth rates. And yet:

ImageWe do not see any  association – let alone threshold! – with regard to the debt-to-annual GDP ratio and the 5-year forward growth rate (light bar). Or take:


The correlation with past growth (i.e. slow growth causes debt) is far tighter than the corollary. It is noted several times in the paper that the very tight association for past, present, and future growth to debt correlations exist only at very low levels of debt. Perhaps it is because the sample for such is far smaller, and hence less diverse across economic structures, when debt is so low, indeed such is the exception not the rule.

Nagging at me, however, is that the empirics in this study look forward and backward only by 5 years. That is, I am curious how the relative correlation between past, present, and future growth with debt changes if the increment were changed to 1 year. To 10. And 50. The statistical and intellectual validity of very high increments are limited by the fact that:

  • The international economic structure can change wildly over long times.
  • There is very little theoretical reason to believe that debt levels in 1900 affect us today.
  • There is even less reason to believe growth rates in 1900 affect debt today.
  • Path dependency and serial correlation of GDP will make this a very noisy set.

In other words, the economic “long run” is not readily defined.

But “5” is still an arbitrary number. The bond vigilante theory would be quick-acting, suggesting we should employ a smaller increment. To put it crudely, the causality may be “diluted” by other complex factors with a large window. Further, low growth today effects the debt-to-annual-GDP ratio most importantly (by tautology) today, and decreases in causal importance – ignoring serial correlation – over a larger window.

Therefore, it might be interesting to see a correlation that defines the window by a weighted average of the association over a set of years, decreasing by some “discount” rate by year.

Here’s why this is important. There may be a certain “discount rate” – for wont of a less confusing term – or simple window period at which the relative correlation between lever (A) and lever (B) are very different. There would be substantial econometric and theoretical value in understanding the parameters for which (A) is most easily defended and, too, rejected.

A note of caution. There is a big obstacle in understanding the mechanism of causation between debt and GDP that is too little discussed. See:


I’m sure you’ve seen this graph a million times. Note how slow-changing it is. Both Kimball-Wang and Dube note this as a forgone conclusion, rather than as a warning sticker on their results. Unlike unemployment or GDP growth rates, which vary as a nice and bumpy business cycle, the above ratio hardly changes, and does so slowly. There are periods of secular decline (1950-1980) and secular ascent (1980-2000). There are no trends.

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.

Therefore the ultimate lesson from any of the RR-smashers cannot be that debt does not hinder growth. Rather, it reinforces the econometric difficulty of answering such a question. It is often said that macroeconomists are deprived of a scientific lab. However, those studying business cycles and Okun’s Law can may control for factors within the relatively short 15-year time period, wherein the grand structure of the economy is unchanged.

RR, Kimball-Wang, and Dube face the more Sisyphean task of comparing economic structure over the longest of runs. Even the most sophisticated statistical techniques cannot erase the ferment economies undergo.

But I said a new consensus was emerging. No longer can even serious conservative commenters so easily claim to care about “growth” and simultaneously extoll the importance of balanced-budgets. RR, to their silent chagrin, have served, serve, and will continue to serve as the intellectual firepower for causal lever (A). But liberals no longer need to uncomfortably point to hysteresis and the liquidity trap.

It is no longer self-evident based on theoretical whims that debt causes low growth. As a profession, economists will emerge as formally agnostic on this question. And it will be their responsibility to inform politicians of this evolution and hence to propose an economic vision that is not held captive by one little ratio, but perhaps more ambitious goals: to remain the world’s richest and most innovative economy in the face of incredible competition.

1 comment

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s