Two month ago there was an interesting debate on the effects of inequality on economic growth in the otherwise boring Spanish blogosphere. Here is my contribution to the debate, translated from Spanish.
First, if you can read Spanish, and if you want to read the entire debate, read the posts linked here.
If not, the summary goes like this: There are recent studies by the OECD (Cingano, 2014) and the IMF (Ostry et al. 2014) that are cited to support the thesis that inequality does harm growth (And, it is then argued, it must be fought as a separate thing from poverty). But then, a meta-analysis (Neves et al. , 2016) shows that the the impact of inequality on growth, while negative and statistically significant, it is so small that it is economically meaningless (And, it is then argued, there is no need to worry about fighting inequality as a thing in itself)
Indeed in January, Sam Bowman pointed out in this ASI post that those two papers are cross-sectional, they compare countries that are very different from each other. (In opposition to studies within countries, or perhaps studies that consider many countries and fixed effects. He cites Forbes (2000), altough there is much recent work available (including papers that directly criticise Forbes like Banerjee and Duflo, 2003), so arguably that post doesn’t settle the debate (And Sam should have taken the later literature into account!).
Back to our debate, how comes those two papers reach conclusion at odds with the meta-analysis? Because one has to carefully differentiate between inequality of income, inequality of wealth, measures of inequality, source of inequality, and in which set of countries inequality occurs.
For example, another paper cited (Voitchovsky, 2005) argued that inequality caused by the poor becoming poorer is bad and harms growth, but that inequality caused by the rich getting richer is good and increases growth. And also, yet another paper argues that inequality in developing countries is negatively correlated with GDP growth, but that relationship does not obtain for developed countries, where it is neutral or even slightly positive (Kolev and Niehues, 2016)
And finally, another paper that tries to separate the individual channels that cause inequality reaches the conclusion that while most inequality in the aggregate is bad, if one takes only the subset of inequality that happens because of market processes, that one is actually good for growth. (Castells-Quintana y Royuela 2017).
What followed after this was a methodological debate on sample sizes, and on what the meta-analysis really said, to the point where I had to email one of the authors to settle the debate (And it was settled in favour of the interpretation that inequality was indeed economically insignificant)
What do economists think?
What is the consensus in economics about this issue? The ones arguing for a negative relationship between GDP growth and inequality pointed to a vast literature that showed the opposite result, but if one actually looks at the recent literature, what one finds is quite different:
However, if there is anything we can take away from the existing literature then, it is the fact that there is no consensus on the question of whether inequality affects growth positively, negatively, or at all (Herzer and Vollmer, 2012)
The empirical literature on the relationship between inequality and growth is large, and there is no consensus on the question of whether inequality affects growth positively, negatively, or at all. (Klasen et al. 2016)
Summing up, although theoretically the relationship between inequality and growth works through different channels, with inequality potentially having at the same time a positive and a negative effect on economic growth, empirical evidence in this sense remains scarce. Castells-Quintana and Royuela (2017)
The large empirical literature attempting to establish the direction in which inequality affects growth is summarized in the literature review Table A2.1 (see Annex 2 ). That survey highlights that there is no consensus on the sign and strength of the relationship; furthermore, few works seek to identify which of the possible theoretical effects is at work.This is partly traceable to the multiple empirical challenges this literature faces, ranging from the poor quality of available data to the limited possibilities of capturing changes in the shape of income distribution and an estimation approach reflecting a lack of time series variation. Cingano (2014)
This work builds on a tentative consensus in the growth literature that inequality can undermine progress in health and education, cause investment-reducing political and economic instability, and undercut the social consensus required to adjust in the face of major shocks, and thus that it tends to reduce the pace and durability of growth (Persson and Tabellini, 1994; Easterly, 2007; Berg, Ostry and Zettelmeyer, 2012)
But even they don’t go as far as saying that the consensus is settled or solid.
Interestingly, Paul Krugman said two years ago that people who want to see a negative relation between GDP growth and inequality might be falling prey to wishful thinking. For a progressive it would be convenient that the efficiency-equity tradeoff is actually not even a tradeoff, at least for the level of redistribution currently existent. Still, he says, if there is no relation (as he thinks, and as I think that is consistent with the data, one can still do redistribution without harming growth (But then one loses the redistribution-is-efficient argument)
The methodological dispute
Given that there is a relatively recent meta-analysis, the discussion centered around it quite heavily, specifically on Table 4. The meta-analysis tries to explain why the papers reach different conclusions, and controls for many different things that I mentioned earlier and more.
This seems to be, at a quick glance, a table saying that if you set your dummy variables to 0, the effect of inequality on growth is -0.0327, each point increase in Gini reduces growth by -0.03 every year (From 2% to 1.97%, for example). Increasing Gini by 10 points would however decrease growth by -0.3 points every year, quite more noticeable. But if you set to 1 the relevant dummies for developed countries (and OECD, income inequality rather than wealth inequality, etc) you get a range of estimates between –0.0116 and +0.008. Hence one could claim that inequality has no effect on growth (because it is quite small, and the results depend on how you measure it).
But but then, that regression says that it was done on the t-statistics, not on the effect of inequality itself(!), so maybe nothing can be said about the effect from that table, only which variables significatively explain the different findings that are presented in the literature.
But but but then if one looks at what the authors are doing, they are normalising both their explanatory variables and the dependent variable by their standard deviation as a method to control for heterokedasticity, so it really is a measure of the effect of inequality on growth, directly.
One of the authors of the paper confirmed that this table does measure the effect of income-Gini on yearly GDP growth, and thus ended the debate.
I am sure there will be more refinements of this literature: the meta-analysis didn’t account for the type of inequality (generated by impoverishment vs enrichment, or generated by rent-seeking vs generated by natural market forces). If those factors were included, the estimates might be even lower, or perhaps slighly positive. In any case, it is plainly wrong to claim that inequality in developed countries harms growth. Institutional quality would be the moderating factor.
How inequality affects growth
One thing is to study the magnitude of the impact, another is to study why one thing has an impact on the other.
Neves at al. (2013) and more recently Klasen et al. (2016). did that Six specific channels are proposed in the literature, remarking that the evidence behind these channels is limited, except perhaps for the instability channel.
- Credit market imperfection: If they are present, it will be difficult for the poor to invest or acquire education. They then can’t use their full potential and this hampers growth. The last paper about this topic cited in the revision is from 1999.
- Fiscal policy: According to the median voter theorem, the lower inequality is, the median voter will want and will get less taxes. In the limit, if everyone has the same income, it doesn’t make sense to redistribute. That is, more inequality would cause more state spending and taxes, and this would reduce growth
- Corruption: More inequality can affect growth negatively by giving the rich more power (lobbying, political campaign finance, bribes, etc), and that somehow is bad for growth (They perhaps would vote for policies that are good for them but bad in aggregate or something like that)
- Sociopolitical instability: A two staged mechanism. First, inequality generates political instability (Ranging from social unrest to civil war), then those things cause lower growth due to lower investment and increased uncertainty.
- Savings: While one might think that the rich, as they save more, would generate more economic growth if they control a larger share of the income, some studies suggest that there is no relationship at all.
- Economic instability (from the IMF paper): Drawing from Rajan (2010) and Stiglitz (2012), inequality could increase debt among the poor (to “keep up with the Joneses”), and then, politicians would react to this increase in debt by enacting policies that may then lead to a boom-cycle (In this case, to make homeownership easier). Stiglitz in particular points to the rich as a major cause, as they would co-opt the government and use it for their own ends.
Given that most of the channels are not particularly well studied, lets focus on the promising ones: the one that Ostry et al. mention, and the instability channel.
Let us study first the Stiglitzian part of the channel: the influence of the rich.
If there is a developed country with a high degree of inequality and an open debate about the impact of the rich in its political system, that is the United States. The thesis that the rich have a massive influence in US politics is a popular one, and while in most cases general belief is a marker of truth, it is not in this case. The go-to paper for this has been, until recetly Ansolabehere (2002).
But more recently, Gilens and Page (2014) claimed that the rich do influence poitics in the US, lending more support to the popular claim. This need not be bad in the aggregate, given your own values. But three years after the paper, those conclusions are unwarranted for reasons detailed here. And if this is so in the US, the claim might be extensible to countries where is far less money in politics.
Rajan’s hypothesis can be split in two pars: first, that inequality increases family debt, and second, that family debt increases the risk of a financial crisis.
For the first part, there is mixed evidence (Yamarik et al., 2016, Coibion et al. 2017). In addition, Rajan’s mechanism doesn’t seem to have happened outside of the US – if it even has happened. For example, in the UK, inequality was constant in the period 1990-2008, and still there was a housing bubble.
I couldn’t find evidence for the family debt-crisis relation, but I found papers testing directly the relation between inequality and financial crisis.
Bordo and Meissner (2012) reject the hypothesis using OCDE data. Gu and Huang (2014), using a similar sample, conclude that sometimes the relation does exist, but they acknowledge that “IIt is natural for our result to differ from that of Bordo and Meissner (2012) because of the adoption of different subsets of the data and different models of regressions. Indeed, different estimators or different specifications may produce differing results in terms of estimated values and their significance. Therefore, the results from their work and ours should be interpreted with caution, indicating that we are still unsure whether rising inequality leads to a financial crisis in general.”
Morelli and Atkinson (2015) say that the evidence does not back a relation between inequality and economic growth in general, although it cannot be ruled out for the US and the UK. At this point, they conclude, evidence remains inconclusive.
Going back to Ostry et al. and some papers cited therein, Persson and Tabellini (1994) present a possible mechanism through which inequality can negatively impact growth. This would be that an increase in inequality would produce social unrest, and this would cause voters to vote for redistributive policies that would in turn decrease growth. They say, however, that the evidence backing that channel is meagre, and they themselves focus on the inequality-growth relationship directly.
Easterly (2007) doesn’t study the effect of instability directly, but mentions the channel in his literature review. He finds that inequality predicts worse institutions. (An increase of 9 points in the Gini index reduces institutional quality by a standard deviation, and income by a standard deviation). Easterly does not measure inequality directly, but uses the surface dedicated to wheat and sugar as a proxy, as he argues that past literature shows that it is a powerful determinant of inequality. More recent work casts doubt on this particular paper (Dutt and Tsetlin, 2016), where they compare the use of OLS and LASSO for selecting predictors, and they find that poverty is the most powerful predictor for underdevelopment, inequality being only predictive insofar as it is associated with poverty.
Berg, Ostry, and Zettelmeyer (2012) find that income negatively affects growth, but they don’t study whether inequality has an impact on social instability.
Other papers that featured during the blog-debate were:
Alessina and Perotti (1996) build an index of aggregate political instability: political assassinations, number of people assassinated in conjunction with outbursts of large scale violence, number of successful coup de états, and whether the State is or not a dictatorship. The reader might – correctly – think that this is kinda irrelevant for developed countries -. They conclude that the kind of instability that affects growth is that which manifests itself through violence.
Barro (2000) concludes that inequality has negative effects on GDP growth for poorer countries, and that it has a positive effect on rich countries. He doesn’t directly analyse this channel, but he mentions it as a possible channel, deferring to the literature.
Ehrart (2009) says that the political instability channel has empirical support from the literature, citing Alesina, Peroti and Gupta (the latter of which considers together violence against the regime, violence committed by the regime, or violence committed within the regime). Property rights insecurity, electoral system manipulation, and crime would also count as instability for him.
Most recent work, however, finds somewhat different results. Comeo and Nehrer (2012), studying 24 OECD cuntries for 30 years do not find effects of inequality on honesty, altruism, sense of civic duty, a small negative effect on obedience and tolerance, and a positive effect on work ethic. Caveat: the study is based on self-reports, which are not exactly 1:1 correlated with external measures of those traits…
Similar results are obtained by Fairbrother and Martin (2012), in that while cross-sectionally it holds in the US that higher inequality states have less trust, over time increases in inequality do not cause lower trust. For more fine grained (county level) data, even the cross-sectional relationship disappears.
Nivette (2011) did a meta-analysis of the possible determinants of crime at a national scale. Inequality turns out to be a powerful predictor, only behind being a Latinamerican nation.
However, Pridemore (2011) and Pare and Felson (2014) argue that if one controls by poverty, the effects of inequality on crime disappear, arguing that for that reason, correlations found previously may have been spurious.
For homicide, Oulmet (2011) does conclude that it is inequality, not economic development or poverty that drives the correlation. In countries with a high HDI, though, while the correlation is still positive, it is not statistically significant. (It is in the whole sample, and in the set of countries with average development, suggesting that the negative effects wane with economic development).
Trent and Pridemore (2012) did a literature review of the social determinants of homicide, and say that as of 2012, the inequality-homicide relationship has becoma a a generally accepted “stylised fact”, but that there is preliminary evidence that once one controls for poverty (though not many studies do), inequality stops being a powerful predictor. They do accept, tentatively, the inequality-homicide relationship, as also do Koeppel, Rhineberger-Dunn and Mack (2013).
Another meta-analysis (Rufrancos et al. 2013) finds that there is an inequality-crime against property relation, but that the relationship for crimes against the person is more complex and depend on the sort of crime considered. Many of the articles studied do not control for poverty, which surprises me.
A recent line of research in the literature suggests that what matters is not inequality, but how visible it is (Hicks and Hicks, 2014), which makes sense, given that people do not have much of an idea about how unequal their societies are (Gimpelson and Treisman, 2015)
And finally, a paper I found after I wrote the original post: Santos, Testa, and Weiss, who argue, replying to Pridemore et al., that even when controlling for absolute poverty, inequality remains a predictor of homicide. 1 point increase in Gini increases homicide by aound 2%. Poverty is measures using infant mortality as a proxy, and has an effect of 0.5. The most powerful predictor is the male/female ratio, altough it is not statistically significant in all the quantiles they study. But its influence ranges from 2 to almost 6%. Oddly, but in line with other papers (Rogers and Pridemore, 2015) , they find no effect of the share of young people on crime. On a footnote, they replicate the study of Pridemore et al., who only considered developed countries, and so for developed countries, inequality does not seem to matter as much as absolute deprivation.
Chong and Gradstein (2007) study the causal relations between institutional quality and income inequality, concluding that there is bidirectional causality, with the inequality->worse institutions being the strongest.
Savoia, Easaw, and McKay (2010) review the literature on the impact of politic-economic inequality in institutions. They emphasise the need to do more research given that the relationships between economic institutions, democracy, and economic inequality are not well documented. With that caveat, they say that economic and political inequality can affect institutions through rent-seeking bhaviour, where elites take control of the State apparatus and design policies that benefit themselves at the expense of everyone else. As an example of this mechanism they cite post-transition Russia. Another mechanism for this is the exacerbation of distributive conflicts with growing inequality, although more recent research goes against the latter part: more inequality apparently does not drive people to demand more redistribution (Wright, 2017)
Bagchi and Svejnar (2015) find a negative relationship between inequality and growth. But when they control for “politically connected” inequality, the relationship disappears.
See also the Easterly and Dutt and Tsetlin papers above for more on the institutional side.
After the discussion above, what should one think about the relationship between inequality and growth?
For starters, that the consensus of the literature points to our lack of knowledge, and the need to be very careful when studying these phenomena. As of today there is no solid consensus on the effects of inequality on growth. Tentatively, on the grounds of Neves et al.’s meta-analysis, we can conclude that the impact of inequality on developed countries is economically insignificant. This means that one can claim that inequality is good, bad, or neutral for growth as long as the effects claimed are small and one talks about developed countries. For developing countries, the relationships are more negative.
On the channels that are supposed to drive the relationship, most channels do not have solid evidence backing them.The evidence is somewhat better for crime and for institutional deterioration if one considers developing countries.
A poor understanding of the effects of inequality on growth and the channels that tie both variables can lead to policy proposals that may suppose a waste of resources (a waste given the values of those that espouse them, that is). As an example, one might think that the more schooling the better for economic growth. It sounds totally plausible, but the reality is that once a basic level of education has been achieved, it does very little. Other famous example is that of quality pre-school education, once defended by most policy wonks as a panacea with very high rates of return on investment. But as of today, the effects of this intervention are far from being proven with more recent studies finding much lower effects.
To finish, some considerations on the IMF and OECD papers around which the present debate revolves. Do their numbers actually make sense? The effects that they posit are quite big.
In the last 60 years, the countries that have reduced their inequality the most (Around 10 points, measured by income-Gini) have been countries like France or Italy, and they haven’t seen an important increase in growth (If anything, the opposite is true). It could be argued, surely, that perhaps other factors have counterbalanced this underlying increase. But this is hard to believe. If we believe Ostry’s or Cingano’s papers (Or, rather, if we read them naïvely), growth would have been progressively increasing until reaching 0.1*10=1% extra per year. That is, instead of growing at 2% they would be growing at 3%. This effect is 10 times larger than what Neves found (Table 3, model 4 in Ostry’s paper). What other economic force of similar magnitude could have counterbalanced that? In the case of France, perhaps an increase in the size of the State.
On the other hand we have Sweden, that increased their Gini between 1980 and 2012 by 12 points. This would have led, conversely, to a decrease in yearly GDP growth of 1.2%. But again there is no substantial trend change… perhaps because of a reduction in the size of the State in the same period? Maybe we will notice it in the future? Or maybe because the effect is, like the meta-analysis sais, economically insignificant?
It is often said that what matters and should be remedied is not inequality, but poverty. People with an egalitarian disposition would be predisposed not to accept this, and so it is natural that they want to find that inequality also matters somehow in terms of economic variables. But no, it mostly doesn’t. The evidence here, however, also suggests that they might be able to get their equality if so they want it, for free. The so called equality-efficiency tradeoff doesn’t seem to be a very strong one. The efficient frontier appears to contain one point for each level of equality one might desire. This could be an artifact, however, of the different ways that inequality affects growth. If one lives in a institutionally robust country, combating inequality could reduce growth (from the incentive-reducing effect), and viceversa for countries with high corruption.
If one still desires to reduce inequality, and one decides to do this via increasing the size of the State, one is faced again with the tradeoff: an increase in State size of 10 points (Making the US into Sweden) is associated with a 0.5-1 decrease in GDP growth, per year. (Bergh and Henrikson 2011, Bergh and Henrikson 2015) (Maybe! But Maybe it is the same story with inequality and there is little that can be done to affect growth by shifting resources around)