- Reinhart and Rogoff may understate correlation between high debt and low growth.
- Heavy selection bias in Reinhart and Rogoff, Herndon, Ash and Pollin, Dube, Kimball and Wang.
- The most pernicious effects of high debt ignored.
- Attempts to determine causation flawed due to arbitrary assumptions and selection bias.
- Policy makers should be very wary in using any of these studies to inform their intuitions.
Does High Debt Cause Low Growth?
Back in 2010 Reinhart and Rogoff demonstrated a correlation between high debt and low growth across 40 or so economies. They did not demonstrate (and did not attempt to demonstrate) causation, although they and many others have implied causation, partly on the basis of their study.
Unfortunately some errors crept in to Reinhart and Rogoff’s calculations – these were uncovered by Herndon, Ash and Pollin. After correction, the correlation is still there, and still significant, albeit smaller. Be that as it may many opponents have opportunistically attempted to discredit the entire study. This is also unfortunate.
What has not been highlighted though is that the Reinhart and Rogoff correlation as it stands now is potentially massively understated. Why? Due to selection bias, and the lack of a proper treatment of the nastiest effects of high debt: debt defaults and currency crises.
The Reinhart and Rogoff correlation is potentially artificially low due to selection bias. The core of their study focuses on 20 or so of the most healthy economies the world has ever seen. A random sampling of all economies would produce a more realistic correlation. Even this would entail a significant selection bias as there is likely to be a high correlation between countries who default on their debt and countries who fail to keep proper statistics.
Furthermore Reinhart and Rogoff’s study does not contain adjustments for debt defaults or currency crises. Any examples of debt defaults just show in the data as reductions in debt. So, if a country ran up massive debt, could’t pay it back, and defaulted, no problem! Debt goes to a lower figure, the ruinous effects of the run-up in debt is ignored. Any low growth ensuing from the default doesn’t look like it was caused by debt, because the debt no longer exists!
Since Herndon, Ash and Pollin there have been further critiques of Reinhart and Rogoff. Dube’s study focuses on the idea of “reverse causation” and contains results that indicate a correlation between low GDP growth in the past and current debt to GDP ratio. The implication one might draw is that “High Debt doesn’t cause Low Growth, it’s the other way around”. Dube stops short of explicitly saying this, but others have jumped on his results, such as Matthew Yglesias here: “there’s some empirical confirmation of the basic theory that high-debt episodes are largely caused by slow-growth episodes.”
Dube has confirmed nothing of the sort. There are a number of issues with interpreting his calculations in this way. Firstly, it only shows a correlation between low GDP growth in the past and current debt to GDP ratio, over the 20 most advanced economies, in their most prosperous time in history – from WW2 to 2009. It also includes some arbitrary assumptions – most notably that three years forward and three years back are somehow the key time periods to consider. Why three? No reason is given.
The primary problem with interpreting Dube’s paper as having anything to say about causation is that it compounds the selection bias first introduced by Reinhart and Rogoff. If Dube wants to say anything about general causation he needs to look much more broadly, account for debt defaults and currency crises, and look at many more leads and lags than just three years. It’s unlikely that the Russian debt default of 1998 was caused solely by borrowings from 1995, 96 and 97. These things take decades to build up.
Kimball and Wang follow Dube down the same path and appear to generate more compelling evidence, but on the whole they repeat the same errors. They are merely attempting to refute Reinhart and Rogoff based on the same biased set of data, using some unjustified assumptions about leads and lags.
Let’s take a step back and think about how causation might logically happen.
The first concept to be aware of is that these things are not usually a simple matter of one factor causing another. In fact, feedback loops are much more common in economic systems than simple cause and effect. Is it not possible that lower growth causes more debt, which causes lower growth and so on in a feedback loop? Why must we assume simply that one causes the other?
Is it not also possible that more debt causes more debt? If a government can borrow more to make its interest payments, and also borrow more to roll over debt so that it doesn’t need to pay it back for a very long time, won’t it tend to do so? In fact, haven’t governments often tended to do this?
And if governments do this, isn’t it possible that the effects of borrowing now may not affect growth significantly until a very long time into the future? Surely to get the full picture we need to look at countries that have been through full debt cycles – i.e. increases to very high levels followed by reduction back to low levels? The period since WWII is not a long enough period to capture the full dynamic of all the countries involved.
I can think of at least six effects of an increase in government debt on GDP: a) a short term boost to GDP as the funds are spent; b) a lagged reduction in GDP due to interest payments for the term of the debt; c) a boost to GDP due to mitigation of b) from previous debt; d) a lagged sudden reduction in GDP due to need to pay back principal; e) a boost to GDP due to mitigation of d) from previous debt (i.e. a rolling over of debt); f) a long term boost to GDP via investments funded by debt that generate more value than the value of interest payments. There are many more indirect effects, such as effects via inflation, pressure on central bank to purchase debt and reduce interest burden and so on.
These effects are complex, and affect GDP over differing time periods. The positive effects tend to be short term. The negative effects can be very long term. Contemporary correlation captures only a sliver of the story. A three year backward or forward correlation is also wholly inadequate. The whole debt cycle needs to be taken into account.
In addition, it is quite possible that increases in debt when debt is low, are efficient and profitable, but at the same time increases in debt when debt is already high are not. None of the causation studies try to account for this – all data points are aggregated together.
The academics discussed above seem to be so enamoured of their statistical armory, and so determined to make political points one way or the other, that they forget to step back and critically examine their own processes. Their papers are littered with unstated assumptions and arbitrary parameters. Their correlations are heavily biased toward healthy nations, and a search for contemporary causation in the face of logic that suggests long term influences of past policy.
Economists face a difficult task. They cannot run controlled experiments in the same way as scientists. Rather than accept this, and therefore accept that their ability to know is inherently limited, they clutch at straws and pretend that the data we have describes all there is. They pretend that there is simple cause and effect, and that this only happens over the short term. Policy makers need to be continually on their guard against acceptance of recommendations based on studies such as these.
Update June 2 2013
In any broader study of the effects of high debt on GDP, another factor that must be included is bail-outs. If a country is in financial trouble but is saved from default via a bail-out, the resultant lowering of debt must be accounted for as if it was a default. Without this the negative effects of high debt are understated.
Ideally policy makers would also keep in mind that there are many other factors. For example not all growth is equal. If a country has high debt, it will likely have a class of powerful bond-holders, to whom the remainder of the taxpayer population is indebted. Therefore high debt may be correlated with a high wealth divide. If growth of the bond-holding sector’s income is high, and this contributes significantly to GDP, then the aggregate GDP figure may be masking an unsustainable underlying wealth gap dynamic.