The Economist's Case for Reaction: is Democracy Good for Growth?
“More political rights do not have an effect on growth… The first lesson is that democracy is not the key to economic growth” - Robert J. Barro
“I’m a monarchist because we don’t have to live like this.” - Curtis Yarvin
This blog post is lengthy and divided into three sections, so feel free to skip to whatever interests you. The introduction summarises the recent history of anti-democratic thought on the Right and why we should empirically analyse the success of democracy. In the second section, 'Democracy and Economic Growth', I perform a short review of the economic literature on whether or not democracy improves economic growth. In the third section, 'Reaction and the Problem of Variance', I look at why dictatorships are more varied in their growth rates and whether or not this is a reason to prefer democracy.
Introduction
Away from academia and mainstream media, on the internet where actual thinking occurs, the online right has progressively shunned democracy.
The neoreactionary movement has for the last 15 years revived arguments against democracy and developed them for contemporary circumstances. Curtis Yarvin, the movement’s key thinker, makes two important arguments in my opinion. 1. That absolute power is essential for effectively wielding increasingly large states towards effective outcomes and that 2. Liberal democracies that encourage input from voters and non-profit civil society suffer a fatal ‘power leakage’. When power depends on opinion, the public debate becomes a breeding ground for info viruses/memes that best evolve to rent seek and control elites, opinion and institutions through social desirability bias, virtue signalling and censorship.
Reactionaries such as Curtis are often ex-libertarians and these intellectual cousins agree. Libertarians have always thought (classical) liberal democracy is a contradiction in terms for which classical liberalism should take precedent. Hayek was ready to support capitalist dictators over democratic socialists, whilst Friedman and his Chicago Boys advised the dictator General Pinochet. In recent years these ideas have been formalised in an array of books and papers, such as Bryan Caplan’s ‘The Myth of the Rational Voter’ arguing that voters are systematically biased against good policies. He suggests a radical restructuring of the franchise with especially ‘wrong’ groups such as immigrants being barred from voting (see his book ‘Open Borders’).
The third part of this online coalition against democracy is the more general ‘alternative right’ (for the lack of a better term) such as Bronze Age Pervert. In fascist-lite sympathy, this group adores the effective strong man and correctly recognises that the democratic system they live under suppresses their ideas and persecutes their supporters.
Whilst a lot of ink, and even more text, has been spilt outlining the theoretical arguments of this coalition (and responding to it), this coalition has notably avoided empirically testing their theory that democracy is bad. Curtis Yarvin’s writings are filled with interesting theory and impressive historical analysis, but anecdotes can be cherry-picked.
In this essay, I review the empirical evidence on whether there is an economic argument for reactionary regimes. I first perform a literature review of the effects of democracy on economic growth and conclude there is no effect on average. After that, I show that dictatorships have a higher variance in growth rates suggesting that Democracy is a safer option. Whilst reactionary regimes may be the ideal form of governance, being undemocratic is risky - a roll of the dice only advised only under desperate circumstances.
Before I begin with my discussion of the literature I should stress this argument from the online right is not merely academic. With deep polarisation and political discontent in western societies we are in fertile conditions for political actors to consider undemocratic reforms and actions - either to prevent populist right takeovers or because the right must take authoritarian action to achieve their goals.
After all, dissatisfaction with democracy has been rapidly increasing for over a decade in Western countries. In America, dissatisfaction has more than doubled since the dawn of the new Millenia reaching, from 25% to 50%. Perhaps it should not have been a surprise when French generals recently wrote a letter to Macron threatening a military coup, with no less than 58% of the French people supporting the letter.
Democracy and Economic Growth
Many of the richest countries are Democratic, but is this really causal? When we plot how Democratic a country is against their GDP per capita there is a small positive relationship, but there are plenty of undemocratic countries that are doing quite well. Moreover, the relationship is U-shaped - the richest countries are either very democratic or extremely authoritarian. When we plot economic growth against how democratic a country is now we find no relationship! Note the non-democratic countries that genuinely perform badly are in Africa, which has many problems besides being undemocratic. From a superficial look at the data, the argument that democracy is good for economic growth looks tenuous.
The earliest empirical research on democracy and economic growth typically modelled economic growth for a given timespan using explanatory variables from a starting year eg. Enrollment in schools, climate and measures of democracy. In the 1990s these cross country growth regression became very fashionable and widely published. However, these papers could not find a robust effect for democracy whether positive or negative. To get published with original findings the papers would slightly change their methodologies, time frames, country samples and explanatory variables, resulting in a constant back and forth between economists who found slightly positive effects of democracy and others who found slightly negative effects. By 1993 economists Przeworski and Limongi wrote the first review of this empirical literature - just as many papers found negative effects of democracy as positive effects.
Despite this inconclusive result the economists kept publishing these regression papers, using slightly different sets of explanatory variables to get slightly different results. Eventually, a new more efficient methodology was used to study economic growth. Instead of publishing a paper with a few different models in, economists used Bayesian Modelling Averaging - running millions of models with different explanatory variables then taking a kind of weighted average of the results. The idea behind this is that it solves the problem of model uncertainty. Without knowledge of which variables should be in the ideal model economists could choose just the right combination of variables that lead to results supporting their hypotheses - be it pro-democracy or against. In Sala-i-Martin’s paper “I just ran four million regressions" he found the coefficient for his democracy index to be negative and only statistically significant in 17% of regressions. In other words, after millions of models created by the supercomputers of the day, economists found that democracy doesn’t affect growth.
For comparison, in 2005 Garett Jones used the same dataset but also included National IQ. He found IQ to have a larger ‘posterior inclusion probability’ than any of the 59 variables employed by Sala-i-Martin. That is the results suggest National IQ had the highest probability of being in the ‘ideal model’. Despite these results, few economists are willing to consider that IQ affects economic growth but, as we shall see, many still believe that democracy is good for growth.
In 2008 and 2019 economists published meta-analyses on democracy and economic growth - studies that combine all the results of previous papers in their analysis. In the 2008 paper 84 studies were used and 188 studies were used in the 2019 paper. The first meta-analysis’s results were as expected, showing zero effect of being democratic on economic growth. Despite this, the authors claim their results only show democracy has no ‘direct effect’ on economic growth but is likely to have indirect effects through other variables democracy correlates with such as human capital, lower inflation, lower political instability, and higher levels of economic freedom. Whilst this interpretation is not impossible it is very unlikely - we have no reason to suppose that the correlation between these factors and democracy is causal or that causation goes from democracy to these factors. Besides, in the cross-sectional studies controlling for democracy in the starting year would also account for its effects on human capital in subsequent years (assuming the ‘indirect effects’ of democracy are continuous and not just a ‘one off’ benefit). The estimated effect of democracy should partially include its indirect effect already!
The second meta-analysis, from 2019, does find a small positive effect of democracy - about a third the effect size of human capital. Below on the left, the distribution of effect sizes for human capital and democracy are compared. The authors find democracy on average has a partial correlation of 0.04 with economic growth. The effect is extremely small! Moreover, this result is not robust. When the authors only use models that control for human capital they find no effect of democracy on economic growth. These results are in the below plot on the right. As human capital and institutional theories of economic growth are chief rivals in the literature it is inappropriate not to control for human capital. After all, we should expect better-educated people to prefer democracy or be better at sustaining it causing.
Despite Bayesian model averaging and reviews of the literature suggesting democracy has no effect on economic growth, the consensus amongst economists may be changing towards a positive view of democracy. In 2019 Acemgolu et al. published a paper titled frankly “Democracy does cause growth”. It was a big deal - the authors estimated that becoming democratic increases your GDP by 20% over 25 years equating to an increase in the growth rate by 0.7%. This was a far cry from the previous view of 0%.
The authors are something of rock stars in the economic world, giving the paper substantial clout in changing the consensus. The lead author, Daron Acemoglu, is an MIT economist with 176,000 citations and multiple popular books. As of writing, the paper already has over 1,000 citations, roughly double that of the 2008 meta-analysis showing zero effect. If citations are a reasonable measure of influence this one study showing a large positive effect of democracy is liable to change the prior consensus on democracy.
So how did Acemgolu et al. find such strong results for democracy? They used panel data from countries modelling log GDP per capita with lags of GDP, a fixed effect dummy for each country and a dichotomous democracy variable. A problem for modelling the effect of democracy is that being a democracy is itself endogenous, caused by GDP. For example, dictatorships that experience an economic crisis may be liable to be overthrown. This means countries that undergo a democratic transition may appear to have become poorer but that is only because of a prior crisis. Thus by using lags of GDP Acemoglu et al. tried to account for this endogeneity, which they think masks the actual effect of being democratic.
Previous papers using time series or panel data approaches had found smaller effect sizes and sometimes even negative effects. Acemgolu et al. give a couple of reasons why they think these results were so poor. One reason they suggest is that prior papers use continuous democracy indexes (eg. 1-10) which are rife with measurement error, whilst a simple dichotomous variable (democratic or undemocratic) will track whether nations are democratic in a clear way. The crucial reason Acemoglu gives is that the prior papers do not systematically control for lags of GDP, meaning that the endogeneity of democracy masks its true positive effect. Whilst these are plausible reasons for explaining why Acemoglu’s results differ from other panel studies, it does not obviously explain why cross-sectional models of growth do not find a strong positive effect of democracy. This is for three reasons: 1. Dichotomous variables have been used in cross-section studies and in the 2019 meta-analysis they were not associated with more positive estimates of democracy’s effect. 2. The cross-section approach controls for endogeneity in a similar fashion to using GDP lags by controlling for GDP in the starting year. 3. Most democratic countries in the starting year of a cross-section study will have been democratic for a long time by the end of the growth period meaning the effect of a temporary crisis during democratic transitions should not bias the results.
I think the most plausible reason why Acemoglu’s approach finds a strong effect is that his lags of GDP bias the result. The problem is that the lags of GDP before a democratic transition are qualitatively different from other lags of GDP. In general, a country’s GDP will be similar to its GDP in the year before, albeit a little larger. As such the first lag of GDP in Acemoglu’s models has a coefficient very close to 1. However, after going through a temporary crisis a nation’s GDP is going to be much larger than the prior year as it catches up to its pre-crisis level. The model cannot distinguish between permanent falls in GDP and temporary falls in GDP. As such when Acemoglu’s model sees a country that has been driven by a temporary economic crisis to democratize, the GDP lags predicts GDP is going to remain low when the country was inevitably going to grow fast anyway.
If you look at the graph above from Acemoglu et al.’s paper my suspicion looks like it might be right. The graph shows the GDP of the average country transitioning to a democracy relative to similar countries that do not become democracies. Acemoglu states the graph shows that democracy does improve GDP because GDP goes up after a democratic transition. On the other hand, the increase in GDP could just as easily be due to catch up growth after the prior economic crisis. Even on Acemoglu’s optimistic interpretation things look bad for democracy. A country takes 10-15 years to return to its pre-democratic relative GDP level.
Before Acemoglu et al. was even published, economists at the Inter-American Development Bank came up with the same line critique which they laid out in a working paper. If some democratic transitions were caused by temporary economic crises, Acemoglu’s estimate of democracy’s effect would be upwards biased. The Development Bank economists made a survey of 165 country-specific democracy experts asking whether the democratic transition in their studied country was caused by an economic shock (an endogenous democratic transition) or another reason (an exogenous democratic transition). They ran models similar to Acemoglu using all democratic transitions and others excluding transitions caused by an economic crisis. Whilst the former models showed a positive effect of democracy, the latter showed no effect. Democracy is only associated with economic growth when it is preceded by a temporary slump in the business cycle - suggesting it is not actually increasing growth.
Unfortunately, this critique of Acemoglu et al.’s research has gained no traction. It has gained 15 citations whilst Acemoglu’s paper has over 1,000. It seems academics would much rather cite studies showing democracy is good for growth than those showing scepticism. To make matters worse after writing their working paper, Acemoglu et al. had three years to consider the critic's paper yet neither responded to it nor even mentioned it in their final paper - purposefully hiding the problems with their own research. Poor incentives and political bias appear to have severely distorted the economists’ perspective on democracy. Alas poor incentive structures are not unique to economics - a new paper shows that replicated findings are cited less often than findings that don’t replicate
Overall, democracy seems to have no effect on economic growth. Bayesian modelling averaging and meta-analyses don’t support democracy, whilst the popular recent work by Acemoglu et al. is not only weak but appears to actively ignore criticism.
Reaction and the Problem of Variance
On average democracies have no greater economic benefit than the alternative, so is this a draw between the democrats and the reactionaries? No. Whilst on average democracies perform no better, the variances of outcomes for democracies and dictatorships may differ.
One old paper (Almeida and Ferreira, 2003) suggested that the growth rates of democracies have lower variance, making democracy a safe bet. Their theory was that authoritarian Governments give their leaders a lot of power and keep their leaders in power for longer, meaning the results are often highly dependent on getting lucky with the leader. In a democracy, bad leaders can be voted out. At its worst you can get a Mao or the inbred Charles II of Spain as the leader of an authoritarian regime, making democracy the safer bet.
To make sure this effect was real I replicated the result below for different time periods. I use the PolityIV measure for categorising countries as democracies and dictatorships. And the effect remains!
An important possibility is that democracies are more homogenous (typically Western, European countries) whilst dictatorships vary in many traits be they communist, capitalist, Asian or African. Perhaps the variance in dictatorships growth rates is not because the countries are dictatorships? To test for this possibility I compare the 1960-2010 growth rates of democracies and dictatorships after controlling for GDP in the starting year and human capital (National IQ scores). In other words, how do democracies and dictatorships compare when you look at the difference between their actual growth and their predicted growth (as expected from human capital differences)?
Still, the higher variance of dictatorship growth is very clear. Moreover, there is a large negative tail - many non-democracies completely screw up. In his book, 10% Less Democracy, Garett Jones whilst critical of democracy takes a similar position arguing it is a wise safe bet against major government failure. He cites Armatya Sen who has argued that no democracy has suffered from famine and also the literature on the ‘Democratic Peace’ suggesting democracies practically never go to war with each other. On the other hand, the Irish Potato Famine still happened and before the 20th century democracies were *more* likely to go to war.
In a democracy the people have the power to kick out a terrible leader preventing mass tragedy, even if the system doesn’t always choose the best leader or create the best outcomes. More simply the evidence supports the hypothesis of this meme:
Whilst risk aversion makes Democracy useful, it need not be the perfect system. Almeida and Ferreira think dictatorships have a higher variance in growth because their dependence on good leaders makes them unstable. But this is not the only explanation - it could be due to heterogeneity in types of dictatorships. Despite the set of dictatorships including failed states and communist experiments gone wrong, their growth rate is not lower than democracies - meaning there are plenty of dictatorships that do better than the democracies! For every Casto or Chavez, there is a Pinochet or a Hans-Adam II of Liechtenstein.
There is an easy way to test whether dictatorships are more unstable or whether they are just more varied in nature. We can look at whether the performance of individual dictatorships correlates over time as much as democracies do. To do this I used national IQ and GDP per capita in the starting year to model growth in four time periods of 20 years (eg. 1960-1980, 1970-1990…). Then I noted the error terms or ‘standardised residuals’ of democracies and dictatorships in each time period. I took the first principal components of each of these sets of residuals. The principal component is the ‘latent variable’ that best explains the variance in the residuals, it does this by reducing the residuals of each time period into one variable - the ‘prinicpal component’. What I found was that the principal component for dictatorships explained 54% of the variance in the residuals compared with 51% for the democracy principal component.
In other words, the growth rate of dictatorships was slightly more stable than that of democracies! As such, the higher variance of dictatorship growth rates indicates that there are just ‘good’ and ‘bad’ dictatorships, not that individual dictatorships are inherently risky.
Furthermore, the list of reactionary regime types need not be limited to the antiquated dictatorships of today practised outside of the West. For example, Moldbug has suggested the idea of neocameralism - a joint-stock company/state that would use effective mechanisms from corporate governance fuelled by the natural self-interest of the free market. This would not be too different from the very successful East India Company, which through mercantile prudence and cooperation with Indian Princes came to rule over and unify the entire Indian Subcontinent.
Many types of dictatorships have been lost to the past. With all the technology of modernity perhaps we could upgrade reactionary regimes for the modern age. For example, absolute monarchy has untapped synergies with 21st century capitalism. Today we could genetically engineer or clone great leaders to give power to. We also have a wealth of trained social scientists, effective bureaucracies and state capacity which could leverage the wisdom of a philosopher king.
“Real reactionary dictatorship has never been tried” might be the appropriate response here. If the reactionary claims the higher variance of non-democratic growth suggests there can be hyper-successful nondemocratic regimes it needs to be clear what makes these regimes different from the extremely unsuccessful dictatorships. As Scott Alexander has pointed out it’s not obvious why North Korea is not an absolute hereditary monarchy. So whilst we know dictatorships can be better than democracies we need to figure out *which* dictatorships are better.
Even if reactionary regimes can be superior to democracy there is another risk in implementing them. Without popular support, strikes, sanctions and foreign intervention can end the new regime quickly. Rhodesia and South Africa come to mind. An interesting example is the Kapp Putsch of 1919. The German military tried to reestablish autocratic rule only to be overpowered by striking workers and antagonistic left wing elites. The civil servants stopped working making the nation ungovernable. The newspapers and media refused to cooperate so the government could not relay their commands and propaganda. It was so bad the military had to communicate via courier.
What would happen if the French military started a coup? All multinational corporations would shut down their communications, whilst using Twitter and Facebook to organise resistance and strikes. Within days the coup would fail and the military would be purged of its talent. It would be a disaster.
On average, neither democracies nor dictatorships are superior to the other. Nonetheless whilst our democracies are safer because we know how they perform, a dictatorship is an enticing gamble. Dictatorships have more varied growth rates, not because they are unstable, but because there is a wide variety of dictatorships to choose from. A practical neoreactionary agenda needs to be clear on what types of dictatorship are the good ones and how do we get one of these regimes safely? Ditching democracy at best grants you a philosopher king, at worst it brings you communism. Perhaps we can test the waters and evolve towards a less democratic regime? Perhaps when the old regime has collapsed or the nation is in mortal danger, ending democracy could be justified?
After hundreds of years of pleasant democracy, crises forced the Romans to reconsider their Republic. In 49 BC Julius Caesar crossed the Rubicon and uttered "alea iacta est" - the die is cast. Reaction is a gamble, are you willing to roll the dice?