This extended blog post was jointly authored by Ryan Bourne and Vuk Vokovic.
Reinhart and Rogoff's (RR) highly-cited paper on how high levels of gross government debt (over 90% of GDP) are associated with much slower real GDP growth, which allegedly became the justification for austerity, has been brought under scrutiny after the recalculation of their results by three economists from the University of Massachusetts, Thomas Herndon, Michael Ash and Robert Pollin (HAP).
The economists allege three charges against RR’s work: that there was a significant coding error in their original excel spreadsheet, that they selectively excluded some of the data, and that they used unconventional weighting of summary statistics in the original paper.
After fixing for these ‘errors’ (in inverted commas because the first charge is the only non-disputable mistake) they found that mean GDP growth for countries above 90% public debt-to-GDP was 2.1% per year, and not -0.1%, which was the initial Reinhart-Rogoff finding.
Those against public spending restraint have thus had a field day in light of these changes. Detailed critiques can be found on the Rotrybomb blog, or from Matt Yglesias, the FT, and some pretty harsh criticism from Paul Krugman. The take away being reported is that the intellectual case for restraining government spending has been somewhat debunked.
Having had chance to think about it for a week, the reaction seems pretty hysterical. Sure, the coding error was a dreadful mistake to make. Period. And some policymakers and journalists have been too keen in the past to take the stylised fact which RR presented as if it was a gospel truth that the world would end once the debt-to-GDP ratio approached a certain threshold, which was daft. And, certainly, more attention should be paid to the intricacies of different country-specific factors and effects.
But these points should not be used to re-write history, or in fact be used to rubbish the important work that RR have presented over the last few years on the fall-outs from financial crises and the historical impacts of debt burdens. So it’s worth spelling out a few key points:
1) RR’s work was not, by any means, the only justification for fiscal restraint
Indeed, it is far from the only paper which identifies a threshold over which a high debt burden tends to be associated with lower annual GDP growth, either. Cecchetti et al (2011), for example, identified a debt to GDP threshold of about 85% as a point beyond which tends to be associated with lower annual growth. Furthermore, there has been tonnes of literature on optimum fiscal consolidations, which suggest that those which focus on cutting current spending and welfare whilst protecting capital spending and keeping/lowering tax rates have the best effect on output and debt outcomes. These have been well-highlighted on this blog and elsewhere, as we have been critical of the over-reliance thus far of European countries significantly raising taxes and slashing investment spending while often protecting growth-retarding current expenditure. Finally, a point that has been made by both us and Andrew Lilico on numerous occasions is that looking beyond the short-term and into the medium-term, there is a huge range of literature which suggests that lower government spending and/or a lower tax burden, other country specific effects given, actually enhances medium-term growth. In other words, a smaller government is a supply-side enhancement for the economy. Thus, even if you were not worried about the level of debt per se, there is still a strong case for fiscal restraint and a medium-term deficit reduction plan based on spending cuts, which is what the Coalition originally outlined in 2010.
2) The critique by HAP doesn’t change RR’s essential insight, which is that higher debt-to-GDP above 90% is associated with lower growth
In comparison, the findings from the two papers look something like this (data from 1945 to 2009):
Reinhart & Rogoff
Herndon, Ash & Pollin
While a table with a longer time span, going back to 1800, looks like this (HAP paper didn't look into this data):
|Reinhart & Rogoff||Herndon, Ash & Pollin|
Both tables still yield the same conclusion: GDP growth tends to diminish as debt-to-GDP increases. The difference is that the effect isn't as severe as the initial *mean* results of RR seemed to suggest.
In op-ed pieces that RR wrote, however, they mainly emphasised their *median* results presented above, which is why in this op-ed, for example, they said:
“The average high-debt episodes since 1800 last 23 years and are associated with a growth rate more than one percentage point below the rate typical for periods of lower debt levels.” [our emphasis]
And here’s a key point – the mean result of HAP is remarkably similar to the median results of RR from 1945 to 2009, or indeed the mean results of RR for 1800-2009 (which HAP do not consider) – higher debt-to-GDP is associated with lower GDP growth.
3) The HAP result is being reported disingenuously
Reinhart and Rogoff have admitted to the coding error. But the coding error is not what significantly changes the results. Look at the table below, which highlights the correction for the coding error in isolation.
Ratio of public debt to GDP
Average GDP Growth, RR (2010)
Average GDP Growth, RR (2010), corrected
Would these revisions really have been enough to cause the storm we have seen in the past week? No. So what did? Well, HAP calculated different results using a different methodology and different data. They then wrote in the Financial Times:
““When we performed accurate recalculations using their dataset, we found that, when countries’ debt-to-GDP ratio exceeds 90 per cent, average growth is 2.2 per cent, not -0.1 per cent.”
But these different methodological and data issues are debatable. HAP appear to have used the fact that RR made a mistake in the coding to, as one blogger puts it, conflate “alternative computations”, “new data” and “a mistake” into just “a mistake”. This has seemingly been reported as they would have liked, which is unfair and disappointing.
4) But far from “debunking” the link association between debt and growth, the stylised fact of HAP is still extremely significant
But, for the sake of argument, let’s suppose all HAP’s changes are justified. So let’s suppose we live in a world where the original RR paper was never published and HAP was the first insight into this dataset. We are presented with the fact that the mean GDP growth rate is a percentage point lower for countries with debt-to-GDP ratios above 90 per cent compared to those with debt-to-GDP ratios in the range 60-90 per cent. Would we just shrug this off?
Of course we wouldn’t. As Reinhart and Rogoff’s analysis of historical debt episodes have shown, these tend to last for over two decades. Two decades of growth 1 percentage point a year slower is very, very significant. In fact, they estimate that output would be 25% lower under the average debt episode where debt is above 90% of GDP than it would have been if debt was 1 percentage point higher. It goes without saying that this is a big, big result, which should still be taken seriously.
5) Which brings us to the real point of contention – the direction of causality
Given that even the revised paper acknowledged the association, the real contention then is not whether association exists but what is the, or whether there is a, causation between the two. This has nothing to do with the past week’s events. Indeed, even many economists on the pro-spending cuts side of the original debate have been critical of the original RR paper’s reporting, when people seemingly asserted that high debts caused low growth without providing a causal mechanism.
This is a fair critique. But it works both ways. Several journalists have indicated that, contrary to the original reporting of RR, it is in fact slow growth which causes the high debts. But this doesn’t explain why the debt episodes tend to last for two decades. If there was a one-time exogenous shock to growth which led to an increase in the debt burden, then we might expect growth to then recover and the debt burden to gradually erode. RR’s work clearly shows that this is not the case – the average length of these high debt episodes is 23 years. Do “the growth causes debt crowd” have an explanation for this? What else, except for the boom in debt itself, may have caused the growth rate of the economy to fall so dramatically?
The financial crisis could be one explanation, but so far this crowd has not been able to explain ‘why’. Why would a one-time event permanently lead to two decades of slow-growth?
Is it, as John Cochrane has theorised, that: “governments go completely haywire and screw things up after financial crises. They bail out banks. They hike taxes on "the rich." They transfer wealth. They bail out borrowers. They stomp all over property rights (GM.) Thus, they kill capital markets for a generation. They clamp down on the financial system in horse-left-the-barn efforts to regulate "safety." They try big "stimulus" plans.”
Maybe, but many of the same people who object to fiscal constraint tend to be supportive of the damaging policies Cochrane cites.
In fact, there is good reason to expect a reinforcing mechanism through which higher public debt = lower growth. If the public debt stock is high, governments tend to hike taxes, expropriate wealth, inflate or financially repress – all of which will tend to be harmful to private investment. So whilst the initial collapse of growth might lead to a high debt stock, there’s good reason to think a high debt stock might result in lower growth rates from then on.
In the UK’s case, it was not just the financial crisis which caused the high debt jump. As Andrew Lilico has noted, lots of the additional spending wasn’t automatic stabilisers, but a discretionary decision to stimulate the economy. Thus, the inevitability of slow growth leading to high debt is baked in if you believe Keynesianism is necessary and desirable.
6) Of course, high debt situations aren’t always the same, and do not prevent all supply-side improvements/breakthroughs
In accordance to the review of the original findings, Martin Wolf in yesterday’s FT makes the following comment:
“In 1816, the net public debt of the UK reached 240 per cent of gross domestic product. This was the fiscal legacy of 125 years of war against France. What economic disaster followed this crushing burden of debt? The industrial revolution.”
His develops a point of reverse causality in the interpretations of the initial RR finding; it’s slow growth that causes high debt, not vice-versa. And it was the crisis that has caused high debts.
However, isn’t his initial claim a perfect example of reverse causality, or even better an omitted variable bias? Did high debt accumulated at the time create space for or in any other way favour the industrial revolution? Absolutely not! Acemoglu and Robinson explain what did. The industrial revolution was a critical juncture of history that resulted in the largest technological breakthrough the world has experienced so far. It happened despite the large debt.
And the big difference then to now is that we stopped spending on wars when wars were over. Perhaps Martin will tell us when he’d like us to stop borrowing today?
The original RR work threw up some huge economic debates, which deserve to be continued. Despite all the noise this week, their stylised fact that high public debt levels tend to be associated with lower GDP growth endures. Yes, we need more research on the mechanisms – why these debt episodes tend to lead to slower growth over a long period. But to suggest this past week ‘debunks’ the large economic literature which endorsed public spending consolidation is nonsense, and there are clear mechanisms through which we might expect high public debts to feed through into lower growth.