In two very influential papers, Reinhart and Rogoff (2010) and Reinhart et al. (2012) investigated the relationship between public debt and economic growth. By classifying the annual observations of their data set into public debt categories (low debt, medium debt, high debt, very high debt) and identifying public debt overhang episodes, they indicated that higher public debt-to-GDP ratios are related to lower economic growth. They also emphasised that this relationship is non-linear: although the debt-to-growth correlation is weak below the 90 per cent debt-to-GDP threshold, it becomes much stronger above it. As is well-known, these results were used by many policy makers in support of the austerity policies that have been implemented over the last years in various countries.
In their popular critique Herndon et al. (2013, 2014) called the results of Reinhart and Rogoff into question. They pointed out three problems: (i) coding errors; (ii) selective exclusion of available data; and (iii) inappropriate weighting of summary statistics. They showed that when these problems are tackled, economic growth does not dramatically reduce when the public debt-to-GDP ratio passes the 90 per cent threshold. Reinhart and Rogoff (2013) responded by acknowledging the coding errors in their estimations; however, they disagreed that their weighting method is inappropriate and that they made selective exclusion of data. They themselves presented some corrected estimations according to which the negative relationship between growth and debt remains, but ceases to become stronger above the 90 per cent threshold.
An interesting perspective to this debate is that the whole discussion about the relationship between public debt and economic growth would have been completely different if Reinhart and Rogoff had decided to focus on the adverse effects of low growth on public indebtedness rather than on the adverse effects of high public indebtedness on growth; in other words, if they had analysed their data set using a more Keynesian perspective that emphasises the role of automatic stabilisers and the direct favourable impact of a higher GDP on the debt-to-GDP ratio. In a note that I recently published (Dafermos, 2015) I show what their results would be in that case. Using the same descriptive statistics techniques that Reinhart and Rogoff utilised in their papers, I classify the annual observations of their data set into economic growth categories (low growth, medium growth, high growth, very high growth) and I indicate that the public debt-to-GDP ratio increases as economic growth declines. I also identify low growth episodes and I show that in most countries these episodes are characterised by higher public indebtedness. Therefore, if Reinhart and Rogoff had decided to present their data in this way, the main implication of their analysis would have been that growth policies need to be adopted by policy makers in order to avoid high public indebtedness; and not that policy makers need to focus on the reduction of public debt in order to avoid low growth.
Of course, Reinhart and Rogoff are careful about this issue: they clearly state that their analysis does not capture causality. However, by classifying their data set into public debt categories and identifying debt overhang episodes they unavoidably concentrated on the growth-reducing effects of high debt, relegating the debt-increasing effects of low growth to the sidelines. On the contrary, if they had adopted a more Keynesian perspective, they could have focused on the debt-increasing effects of low growth. In that case, their conclusions, which informed the policy debate, would have been completely different.
It is also important that the econometric research that followed the publication of their papers was substantially affected by the decision of Reinhart and Rogoff to focus on the growth-reducing effects of high public debt: most researchers have paid attention to the adverse effects of high debt on growth and not the other way round. Interestingly, the literature has not so far provided strong support to the causality from public debt to economic growth (see footnote 1 in my note). This implies that the empirical research needs to investigate the debt-increasing effects of low growth in greater depth; as would have probably been the case if Reinhart and Rogoff had decided to analyse their dataset using a more Keynesian perspective, or if they had explicitly presented both ‘halves’ of the public debt-economic growth relationship.