In 1896, he wrote this book (An Essay on the Present Distribution of Wealth), which was arguably the first attempt at estimating U.S.-wide inequality and national income at the same time—what we would now call a social table. It was total income and inequality in 1890.

The book was poorly received. Really poorly. It got thrashed in Political Science Quarterly and the American Journal of Sociology. It had tons of flaws.

His estimates relied largely on monetary income thresholds rather than fractiles, used family income rather than individual income, and were built from totals that appeared too low and too heavily rounded to inspire much confidence. More seriously, both ends of the distribution are likely distorted. Capital income at the top is overstated by implausibly high interest rates (7% and 8% whereas the literature points to 4% to 5%), while labor income at the bottom is understated by wage figures below contemporary estimates of actual labor earnings (at $380 rather than a range from $408 (farm labor) to $524 (manufacturing wages) and without accounting for in-kind payments) .

But at the same time, you have to admit it is an awesome piece of work. When it was written—Spahr was a socialist sympathizer which may be in part why he may have been marginalized—it was nevertheless a pretty powerful effort at pioneering. And when you look through what he did, a lot of it is still being done today in these kinds of approaches. True, there are tons of refinements that have been added. But the core is the same. In fact, you can see that the works that followed Spahr in the U.S.—notably Willford I. King—referred to him with some respect and copied parts of his approach. Others followed from Spahr in the 1890s and 1900s. For example, A.J. Ferris improved the level of disaggregation (see below), but did not really solve the underlying problem because his estimates were not independent of Spahr’s. He explicitly used Spahr’s categories and evidence, consulted Spahr directly, and allocated individuals across narrower classes partly on the basis of probability where direct evidence was unavailable.

But interestingly, even modest adjustments for capital income (i.e., via interest rates) and missing labor income brings you within the range of estimates constructed with far more advanced equipment, sources and methods (see table below). Thus, it wasn’t that bad.

In fact, the aforementioned Ferris and King were joined by others like Maurice Leuven and Charles Merwin. They built on Spahr and referenced him. They also improved and refined the basics. And then came Simon Kuznets who referred to this second wave that followed from Spahr. Then the next wave after Kuznets copied from Kuznets and refined him and they added their own stuff. Spahr faded in the rearview. By the 1950s, he had been forgotten entirely.

And in this, there is a lesson: pioneering research is done a bit for fame, a bit for money, and in large part out of the sheer desire to conquer an intellectual frontier. The problem is that the payoffs are uncertain. But down the line, its still pioneering work.

One reason is that the counterfactual for women’s labour-force participation is built using all of Canada except Quebec. The problem is that two simultaneous—but unrelated—reforms occurred in Quebec at roughly the same time: universal childcare and employment insurance reform. Both reforms worked in the same direction by increasing labour-force participation. The problem is that employment insurance reform mattered far more for Quebec than for the rest of Canada, with the exception of Atlantic Canada—New Brunswick, Nova Scotia, Newfoundland and Labrador, and Prince Edward Island.

The counterfactual is thus polluted by a major violation of the basic requirement that control units should be similarly exposed to other important shocks. In a clean research design, the comparison provinces must not receive Quebec’s childcare treatment. But they should be exposed in approximately the same way to other simultaneous shocks. Provinces outside Quebec were not treated by the childcare reform, which is good. Yet many were much less exposed than Quebec to the employment insurance reform, which is bad.

Restricting the comparison to the Atlantic provinces greatly reduces that problem, since they and Quebec were the regions most affected by employment insurance reform. This comparison allows for a better assessment of the effect of childcare policy on labour-force participation.

This can be seen in the Figures below. In the first figure, we should imagine that if universal childcare were the sole driver of the increase in female employment, Quebec should have diverged sharply from the rest of Canada while ALL other regions should have followed similar paths. This is true only if ALL is meant to include of Canada west of Quebec. However, Quebec and everything to the east (i.e., Atlantic Canada) display remarkably similar trajectories beginning in the late 1990s, with both regions experiencing much larger employment gains than Ontario, the Prairies, and British Columbia. This is exactly what one would expect given that Quebec and the Atlantic provinces were disproportionately affected by the 1996 Employment Insurance reform.

We can see something similar by looking for the women aged 55 + which should not be affected by the reform on childcare (because…you know … biology). Their labor force participation surged massively in Quebec suggesting that the employment insurance reform mattered to them alot. It probably does not matter as much to prime age women, but its enough to give massive pause.

The cleaner comparison is therefore between Quebec and Atlantic Canada. Those provinces did not adopt Quebec’s childcare policy, but they were much more similarly exposed to employment insurance reform. In early work, the data used made it really hard to solve that problem though. And this is probably true for the survey (the Labour Force Survey LFS) used in the new working paper. The problem is that you get an insanely small sample once you restrict it to working-age mothers, then to mothers with children under six, and then to the Atlantic provinces. Whatever covariates you add soaks up alot of degrees of freedom. It is simply way too small to do much that would be serious.

Honestly, I don’t know if there is a solution. In unpublished work with Israt Jahan, we tried to address this problem directly. We restricted the donor pool to the four Atlantic provinces and used synthetic control methods to construct a counterfactual for Quebec. I never published the results because the exercise has an obvious limitation: four donor provinces are far too few for strong synthetic-control inference. However, once Quebec was compared only with the Atlantic provinces, the estimated childcare treatment effect fell by between 50 and 70 percent. But, again, the available data are simply not good enough to identify it with great precision. But the direction is clear: comparisons with the entire rest of Canada assign too much of Quebec’s employment increase to childcare.

Despite the results showing what I explained above, I didn’t think our result rose high enough quality-wise and I have been giving the matter alot of thought as to how to solve (but also have a million other papers to finish). Nevertheless, I wish someone would take that measurement point seriously and try to find a novel solution to assess it.

If I am wrong, I wont mind. But what is above should indicate that Quebec’s universal childcare policy did do the things people say it did. However, it did not do as much as people say it did. And it definitely did not pay for itself.

There are notably debates that suggest that living standards for the bottom 90% in the 1920s did not increase much (see notably the work of Gene Smiley in the Journal of Economic History in 2000). But that could simply (but only partly) be a fluke of when Prohibition entered in effect (in 1920).

Consider a few facts I accumulated when writing my forthcoming book on inequality. First, Rachel Soloveichik in a paper for the Bureau of Economic Analysis found that illegal drugs (of which alcohol was included) represented 2.02% of national output in 1933, 0.59% in 1934, 0.25% in 1935 and 0.02% in 1936. As prohibition officially ended in the United States on December 5, 1933, we can take the different between the two numbers to ballpark the share of national output in alcohol at roughly 2%. Thus, already, we know that living standards are going to be “artificially” (in the statistics, not in facts) boosted upward in 1934 and “artificially” deflated starting in 1920. This makes the whole 1920s higher in truth than they look.

Second, consider the implications for comparisons of living standards across countries without prohibition. In those countries, living standards relative to the USA will appear higher than they really were by 2%.

Third, consider that the timing of Prohibition coincides nearly perfectly with the 1921 recession. Thus, if one fits a trend of economic growth from, say, 1870 to 1920—the year before the recession—and extrapolates it to observe the deviation caused by the 1921 recession, some of the apparent depth of the downturn will be artificial and attributable to Prohibition. Moreover, the recovery will appear more modest since the level is “prohibited” (pun intended) to return to the one that includes alcohol and which was part of the 1870-1920 trend. Thus, deviations will appear more persistent than they are.

Is this a huge deal? Depends on the question really! Some facts of economic history are robust enough to not be affected by these statistical and measurement issues. But some are more fragile and may be affected by these simple issues. More importantly, even “strong” facts can be “moderated” which amplifies the relative importance of other factors at play — thus changing the interpretation.

These ads can be collected every year from the beginning through the tools of the archives. And, when you use the Quebec Mercury and Quebec Gazette, you end up with the total services of shipping space offered by shipowners to other merchants. The graph below (only for Quebec Gazette since it was published for more years than the Mercury which started only in 1805 compared to 1764) shows the number of ships entering port (orange line). It starts at relatively low levels in the late 1760s and 1770s, often around a few dozen to a little over one hundred ships. It then rises sharply after the late 1790s and especially after 1810. By the 1830s and 1840s, entries are generally around or above 1,000 ships per year, suggesting a major expansion of Quebec City’s maritime economy. The blue line is the number of ships advertised for charter or rent. It is plotted on the right-hand axis, also in logarithmic scale. Early on, the number of ads is low and very volatile, often around 10 to 50. It rises strongly after 1800, especially in the 1810s–1830s, reaching several hundred and sometimes around 1,000 ads. This suggests that the market for hiring or chartering ships became much thicker over time.

In this graph, I see twopotential papers on economic history that would fill holes in the literature.

The first is that, in other works, I have estimates of ships arriving in ballast. Ships entering in ballast are underloaded which means unused capacity and thus lost revenues. Ads for charter/rent may reveal attempts to remedy that loss. The question becomes: did the charter market improve ship utilization? This is, obviously, a simple enough question but it matters a lot. Capacity issues are basically bottle necks in production and can affect the ability to connect hinterland producers with foreign markets. If you can find ways to create contracts or find an industrial organization form that limits the capacity problem, you access untapped exchanges.

Now, here is the interesting thing for the first paper idea. Below is the two series from the first graph divided by one another (multiplied by 100). Essentially, this the number of adverts as percentage of the number of ships. As one can see, the market is getting smaller. Why?

Well, one thing that happened is that the market became thicker and it was easier to find clients. But there is also a Coasean story in the background. Earlier, newspaper advertisements served as a public matching device for ships and cargo. Shipowners owned a few vessels and that was it. However, over time, owners started to own far more substantial fleets which is why some historians speak of “River Barons” to reflect the move from a small number of owners of a small number of vessels to a larger number of owners and an even larger number of vessels such that there was rising concentration. This is a transaction cost story. The fall in advertisements per entering ship is consistent with a transition from a fragmented maritime market, where small operators relied heavily on public search, toward a more concentrated port economy in which larger firms coordinated shipping through internal fleets, standing contracts, and regular agents.

Writing the story of this transition speaks to market integration across the Atlantic in the first half of the 19th century. Doing so makes for a nice article that could be submitted to European Review of Economic History, Business History, Business History Review or European Journal of Law and Economics.

The second paper idea is tied to the role of information costs. The ads are not just a count of shipping services. They are a measure of information costs. In a thin market, buyers and sellers have to rely on personal networks. In a thicker market, newspapers become matching platforms. Ads lower search costs by making idle capacity visible. This can be combined with ads for other services tied to maritime safety such as insurance ads and auction ads to create a “maritime service index”. Anyone familiar with the works of Douglas North on shipping productivity (who emphasized the role of organization rather than technological innovations) and Knick Harley (who argued for the reverse will see that we now obtain a measure of a “non-technological factor” behind shipping productivity growth. Redoing estimates of TFP (and decomposition) will help reveal if this really mattered to growth thus providing a serious revisitation of a big debate in economic history. This is something that could get into Cliometrica, Explorations in Economic History, Economic History Review, Journal of Economic History or European Review of Economic History.

Both ideas would provide useful progress in economic history and, if someone is interested, I’d be thrilled to collaborate or share the data (depending on the conversation).

Robert Higgs’s ratchet effect (expressed in Crisis & Leviathan) is the idea that government power expands sharply during crises—wars, depressions, emergencies—and then only partially retreats afterward. Like a ratchet! Even when the crisis ends, the state rarely returns to its previous size because new agencies, taxes, regulations, and expectations become politically entrenched. So each crisis pushes government upward like a ratchet: temporary emergency powers become the new baseline.

In Higgs’s model, crises are exogenous. Randomly distributed events whose timing cannot be predicted. But what if the Leviathan doesn’t just exploit crises—but helps bring them about (at the very least without meaning to do it). This is akin to stating whether there is a process whereby crisis in t affects the probability of crisis in t+1. In other words, a probability that has memory.

Why would there be memory? Because government actions in a war crisis can potentially destabilize markets causing a financial crisis in ten or fifteen years. That would be something akin to Mises’ dynamics of interventionism.

To see how, let X ~ Poisson(λh) where h is an interval and λ is what we will call the “exogenous” expected number of events per unit of time. The probability mass function is thus the conventional:

which is a memoryless process and where k is the number of events over the interval. This is a usual property of the exponential part of the distribution. Because intervals are independent, the occurrence of an event in one interval does not affect the probability of future events in subsequent intervals. Here, “events” should be converted in the reader’s mind as “crises”. They should be, for simplicity, assumed to be randomly distributed over time. Normally, the relative likelihood of having to wait exactly t units of time for the next crisis is given by λe^(-tλ). The expected waiting time is 1/λ which gives you the number of time-units before the next crisis.

The ratchet effects of Higgs potentially creates a memory process via the following definition of “relative state”:

where G_t is the size of government at a time after a crisis and G_t- is the size of the government just before the last crisis. This modifies our first equation in the following way:

where λ_t is given by λ φ(R_t) with λ being the baseline crisis rate (as in the classical Poisson distribution) and φ(R_t) is the memory adjustment function that is increasing in R_t. What this latter term captures is whether, after a crisis, the state has been permanently increased in size. In Higgs’ formulation, R_t > 1 should generally be the case if there is no ideological resistance to state growth, such that every crisis leads to a permanently larger government. In other words, R_t introduces path dependence into what is normally a memoryless process.

In Table 1 below, we show an illustration of our mechanism by setting h = 1 year and λ_0 = 0.2. This is the baseline scenario of a memoryless Poisson process. In that case, the expected waiting time would be 5 years with a probability of 18.13% of at least one event in the year. Imagine that once an event-crisis occurs, the government responds to deal with it by temporarily increasing its size. Once the crisis recedes, it remains 10% (i.e., R_t = 1.1) larger than it was before the crisis. This scales λ_0 and we can expect to wait 4.55 years before the next crisis, and there is now a 19.75% chance of at least one crisis-event in the next year. Each of the rows where R_t ≠ 1 in Table 1 shows how much memory is being introduced from the previous crisis via the permanent growth of government.

Another way of formulating this would be to point out that the Gamma distribution (waiting time until the k-th event) would differ across countries in proportion to R_t. Remember that the Gamma distribution scaling of time is λt. The ordering is still the same, but the pace is faster.

In other words, there is a way to think of the Ratchet Effect as being a process with a degree of memory rather that a series of exposure to random shocks whose process is independent from government responses to shocks.

The second most interesting thing in the paper is this graph showing the evolution of sugarcane yields since 1911. Notice that there is an incredible increase in land productivity over time.

The most interesting graph in it is this one showing the evolution of each crop type such that it seems that the huge gains in land productivity come from improved crops rather than better techniques (biological innovations).

Economists familiar with Zvi Griliches’ 1957 paper in Econometrica will recognize the point immediately: this is the sugarcane version of hybrid corn. Griliches’s classic work on hybrid corn showed that some of the biggest gains in agricultural productivity came not from farmers simply “working harder” or applying more capital inputs, but from biological innovation embodied in the crop itself. The seed was the technology and it followed a logistic growth curve (fast initial adoption).

The difference here is that the crop types evolve as a sequence of overlapping waves (notice for example the green , yellow and black coexisted but shifted). This is not one variety taking over once and then staying dominant forever. It is a rolling process: CP 36-105 gives way to CP 44-101, which gives way to CP 52-68, then CP 65-357, then CP 70-321, then CP 85-384, and so on. Each dominant variety eventually declines and is replaced by another.

That actually makes the Griliches link stronger. In hybrid corn, Griliches emphasized diffusion: the new biological technology spreads because it raises yields where it is well adapted. Here, the figure suggests not just one diffusion curve but repeated diffusion curves. Louisiana sugarcane productivity is being pushed forward by a succession of embodied biological innovations.

This makes for a cool question: why is adoption not automatic. The better crop should, logically, be adopted immediately and the inferior being abandoned. Why are some farmers using CP-36-105 while the better CP-44-101 is available? Here, I can see two papers or two into one paper.

Take the data from that paper and its sources. Match with census data on farming and test out the determinants of adoption. I expect that “larger farms” adopt faster while "more market-access” areas (i.e., more connected to infrastructure) access faster. But, most obviously since this is Louisiana, one could look at the historical legacy of lynchings. I expect that in high-violence districts, risk-aversion is greater (or maybe he is risk-neutral but there is the extra cost of his investment being destroyed by racist thugs before it pays). Thus, paying the higher fixed cost of a new and more expensive seed (which is to be expected) is shunned until prices fall more to make adoption profitable for a risk averse Black person.

This is essentially asking a cool question about “technological adoption”. There are lots of papers in economic history and the economics of discrimination on how racism and racist policies affected innovation but innovation is not exactly identical to adoption. New ideas can be made (innovation) but adoption is delayed until profitable or suitable.

Now, because the data is continuous over time, you could also check whether the legacy of racism has a short or a long-half life. A bit like in this Maseland paper on the half-life of colonial legacies published in the Journal of Institutional Economics (see graph below showing the logic).

A paper like this would be immensely valuable for all the reasons above. They speak of: a) the determinants of adoption in a settings where technologies rapidly become obsolete (because of fast innovation); b) the potential role of firm size, market access, infrastructure access; c) the role of racist policies on technological adoption; d) the length half-life of racist policies.

The setting is France’s taxe sur les logements vacants introduced in 1999 (because of course, it has to be in France). The tax applied only to some municipalities in large and tight housing markets. Basically, the ideal setup for a quasi-experiment: some comparable cities were taxed, others were not. The approach used is the pretty usual difference-in-differences approach which (ahead of regression discontinuities and synthetic controls) is the main workhorse for causal inference work.

But the problem is not the method but the variable. The outcome is the private vacancy rate, defined as vacant private units divided by the total housing stock. This is super problematic. The analogy of relevance here is the minimum wage literature where the debate is "should we use employment/population or just employment”. This matters because the minimum wage could reduce employment but it could also driver workers out of the market in question thus reducing population. If both things happen, using employment rate (emp/pop) will be likelier to show fuck all changes (if things move together). The graph below shows this (thanks to Jonathan Roth).

So, going back to the vacancy tax idea, what if the tax reduced vacancies and housing units? What if vacancies increased modestly but housing stock decreased massively? The author of the paper tries to deal with this by looking at the primary residence ratio. It rose by about 0.76 points by 2001 and 1.23 points by 2005. The secondary residence ratio does not rise in a statistically significant way. So the author interprets this as evidence that most vacant units became primary residences, not second homes or fake reclassifications. She also goes at looking at construction which appears unaffected. But this is weak tea because it does not tell us all of the ways that the tax could have affected the housing stocks. Were residential units converted to commercial ones? Were some destroyed? Were there subdivisions of existing units? All of these other sources of change to the housing stocks are not accounted for.

This is problematic because the coefficients found for what she did control for (addressing partly my concern) actually have the sign one would expect given economic theory. But they are statistically imprecise (larger standard errors). Accounting for the full housing stock changes could possibly improve precision and size of coefficients.

Moreover, there is a connected issue in the control variables — the use of the change in the area of housing units (how big units are square feet-wise). This is an instance of bad control. If the tax itself caused units to be subdivided, then average surface area is a post-treatment mediator. Controlling for it is not innocuous — hence why its a bad control. If subdivision is one channel through which the tax works, it estimates something closer to the “effect of the tax holding average unit size constant,” not the full effect of the tax.

As such, I think the case for a vacancy rate is still pretty weak empirically. Even if the French study in Journal of Public Economics is highly interesting and valuable.

That is not hyper controversial. There is a good literature on forest transitions (with debates on exact shape of the relationship and the determinants of when it exists or fails to exist). The interesting part is that they looked at whether “institutional freedom” (polity scores and Freedom house scores) determines the peak point for income. And the answer is that, yes, it does modulate the relationship.

But here is the question for that paper: why political freedoms? I am not saying they dont matter but the bigger question is whether strong property rights modulate the relationship. Strong property rights mean that forest farming is more possible. It also means that negligent forest management comes with higher insurance premiums and tort suits — thus internalizing some externalities. It also means that there are strong abilities to appropriate the profits from innovations that reduce forest (or, indirectly via land use changes and density). Why not property rights? But also, why not open trade variables (tariffs, NTBs, etc.)?

This seems like a low-hanging fruit paper especially since carbon sinks in forestry imply that the results could speak to the question of GHG concentration in the atmosphere.

Agersnap, Jensen, and Kleven’s (AJK) famous Danish “welfare magnet” reports a migration elasticity of 1.3, implying that non-EU immigration responds strongly to welfare benefit levels. The paper focuses on Denmark’s Start Aid policy, introduced in 2002, repealed in 2012, and reintroduced in 2015. The policy reduced welfare benefits for non-EU immigrants during their first seven years in Denmark. AJK used these changes to test whether generous welfare states attract migrants.

The question is essentially how sensitive immigrants are to welfare. The replication focuses on three criticisms but one of them catches my heart more. The AJK estimate is an estimate of net migration which is a highly sensitive measure.

In work with Gary Wagner, Alicia Plemmons and Justin Callais, we looked at internal migration between metropolitan areas in the USA. We were trying to build on the earlier works of others who had migration but we had a hard time understanding the main paper before us. It used the natural log of the net migration rate as dependent variable. This was puzzling for the simple reason that you cannot take the log of a negative number — which net migration can be.

We eventually figured it out, but in the process, we had to think deeply about migration variables and elasticities. The obvious question was: how should we test migration rates? Using the rate simply meant that we were assuming a constant coefficient, not a constant elasticity. Net migration is not a simple flow variable. Because it is made of arrivals minus departures, a small change in either component can produce a large percentage change in the net number, especially when net migration is close to zero. These are serious limitations. In our paper, we opted for the inverse hyperbolic sine transformation, which is well defined for all real numbers. But we still tested with untransformed rates and many other things just to be safe.

But now, reading the replication of AJK, I wonder what would have happened if we had looked at the natural log of gross inflows and gross outflows. Adema’s replication moves in that direction. He looks at gross arrivals, departures, stocks, and asylum applications separately, and asks how each responds. Essentially, he replaced the net-migration object with separate outcomes: arrivals, departures, stocks, and asylum applications. Normally, decomposing what is a “net migration recipe” (all of these variables combined give you net migration) should provide the same result if that result is robust and valid. If not, then the heavy lifting of the result was done by the choice of approach. The point is that the choice of the migration variable is not innocent. It can impose a subtle econometric form that helps drive the result.

This first issue was compounded by a second. The AJK measure of net migration was restricted to the change in the stock of migrants aged 30 and above. This mechanically includes existing migrant residents simply turning 30. Honestly, I had to go back to the AJK paper to look at this, and I didn’t find much of a convincing defense for the 30+ choice. The figure below shows both changes together.

Notice that the differences in raw numbers are not exactly gigantic. Nor are the movements dramatically out of sync. Simply different levels with strong co-movements. This should not yield massive differences. Yet this was sufficient to reduce the estimated elasticity of migration to welfare benefits by roughly three quarters of what AJK found (from 1.3 to 0.28 with the flow variables decomposed). This is a good lesson in how to think about your choice of specification and variables.

AI disclaimer: I used GPT to edit writing.

The point he makes is that looking at the income of survivalist entrepreneurs will overstate their wellbeing. This is because it was often a family strategy whereby a displaced worker opens a small retail business, while spouses or children worked in it without wages, receiving subsistence indirectly through the household.

Proprietor income and labor have been counted. But the unpaid family labor behind the business have not been. So how much does this matter and how much are we missing?

Fortunately, in the 1930 and 1940 census, unpaid workers who were relatives helping in a family business or on a farm are reported. Boyd uses Census data from 1930 and 1940 to examine whether the contraction of the depression distressed demographic groups differently into family enterprise. The analysis compares Black women, white women, Black men, and white men across 22 large cities, with retail proprietorship and unpaid family work as the outcomes of interest and joblessness as the key explanatory variable.

For white women, joblessness was associated with both unpaid family work and proprietorship. A 10 percentage point increase in white women’s joblessness is associated with about 9 additional unpaid family workers and 7.9 additional proprietors per 1000 white women.

For Blacks (both gender), things are mostly insignificant statistically speaking. This is interpreted as evidence that low human and financial capital limited the ability to use survivalist entrepreneurship and use family resources to deal with the downturn in the economy.

Overall, a pretty nice paper with tons of nice possibilities regarding the Great Depression’s study. I am not exactly fond of the empirical approach since the full count census or samples are available to play and something more could have been done. Right now, there are eight regressions of 22 observations. But as proofs of concept goes, this is interesting and worth pushing.

One cool way forward would be to use this paper as motivation, use the microdata and combine it with anything related to labor laws, occupational (Carollo et al. has a nice dataset) licensing or regulation (Rebecca Holmes has a dataset too).

Note: I edited my writing with GPT to remove typos.

I just finished revising a new drafts on this paper which initially started with Carlos Ramirez but now includes Zachary Allen. I asked GPT to summarize this paper in three sentences and it gave me this:

This paper argues that peace was a necessary condition for sustained productivity growth in shipping (more than 1.2% per annum). Using new data from Quebec’s shipping industry between 1763 and 1860, it shows that wartime conditions favored smaller, faster vessels and limited the exploitation of economies of scale, whereas the return of peace after 1815 allowed shippers to adopt larger ships, reorganize trade networks, and dramatically increase productivity especially as larger ships became faster and smaller vessel types became heavier. Technological improvements contributed to this process, but their impact was magnified because peace created the conditions needed for specialization, scale, and organizational innovation to flourish.

To which I would add: had permanent peace happened in 1763 (after the Seven Years War), the first era of globalization could have started earlier. The paper cannot really say that in itself. But its pretty much what I have been thinking about for two years now since I published this chapter with the Cato Institute.

Wars and protectionism pushed the frontier of “economically” feasible technologies for trade — like larger vessels, bigger ports, more capital intensive infrastructure, more complex shipping firms and shipping insurers — inwards. However, the “technologically” feasible frontier never changed. All these techniques were possible and known to shippers. Each time long episodes of peace occurred, such as 1713 to 1740, they were adopted only to be snuffed out by war (see an example from my paper regarding vessel size) as shippers had to shift to smaller (faster) vessels and reallocating vessels depending on trade routes (improper ship-to-route matching).

In practice this means that the “big boom” in shipping productivity that I observe before technologies like steam boosted shipping productivity were just “catch-up growth”. Even before technology conquered nature’s constraints (distance) with the steamship, the productivity was feasible. Implying that the natural barriers were not as big as we think. Manmade barriers (wars and protectionnism) mattered more than meets the eyes. The productivity boom in shipping could have occurred much earlier had there been peace. Ergo my point which may be my future book: did wars and protectionism delay the First Era of Globalization by a Century?

I add that I do think that wealth buys influence. The question is “how much more wealth” do you need to “buy this much influence”.

I think you tax 9/10 of the wealth of the richest and still leave them with 9/10 of their political influence. That is my point. Once wealth exists, it has a strong effect passed a point. You would need to destroy so much to eliminate even a small share of it.

The point, rather, is the “existence” of influence. The more you have political competition, federalism, checks and balance, strong constitutions etc., the less you make it for wealth.

I drew this graph below right now to explain my point. Its simplified, but its sufficient for our purposes here.

The graph says: yes, wealth buys influence — but the real question is the conversion rate from wealth into influence.

On the x-axis is wealth inequality. On the y-axis is the political influence of the wealthiest. The gray Piketty-Saez-Zucman-Cagé line assumes that as wealth inequality rises, political influence rises more and more rapidly. In that view, reducing wealth inequality directly and strongly reduces plutocratic influence. This is the relationship which (if you believe it is accurate) underlies the case for wealth taxes.

The orange Geloso-Kenworthy line says something different: once the rich have enough wealth to enter politics, fund campaigns, lobby, buy access, and shape institutions, the extensive margin already exists. Past that threshold, more wealth still buys more influence, but the marginal effect is much smaller. So taxing away 90% of the wealth (to exaggerate here) of the richest may still leave them with most of their political influence.

The blue Geloso-about-institutions line makes the institutional point. In a system with political competition, federalism, checks and balances, strong constitutional limits, secure property rights, and fewer privileges to hand out, wealth is much harder to convert into political power. That means “political efforts” by the richest are costlier (and offer fewer rewards). Costlier things (with fewer smaller rewards and benefits) means fewer of those things. And that means less influence for the rich. The issue is not only “how much wealth exists,” but whether the state has rents, privileges, and discretionary power that wealth can buy.

So the graph is not denying that wealth matters. It says wealth matters a lot once it crosses the threshold of political access, but the deeper problem is the existence of a political market for influence.

Wealth taxes move you left on the graph; better institutions push the whole relationship downward. That is the public choice point of people like Buchanan, Tullock, Holcombe and others.

But notice the problem for the wealth-tax argument. The usual claim is: “We need wealth taxes because wealth inequality allows the rich to dominate democracy.” But wealth inequality fell in the UK while Cagé and Dewitte find that "the correlation between candidate spending and votes has strongly increased since the 1880s, peaking in the last quarter of the twentieth century". Cagé is also a big advocate of wealth taxes in the French-speaking world (a world where economists who come from it often say things very differently than they do when they speak of the same things in the English-speaking world).

You can reduce measured wealth inequality and still have donor influence, political rents, party machines, and insider access. Say you could cut wealth inequality by 9/10 and still have 9/10 of the political influence of the rich. This is something Lane Kenworthy points out in Is Inequality the Problem? Its a cool book by the way since Kenworthy is actually a social-democrat who wants a bigger welfare and more redistribution but he refuses many instrumental arguments made such as the democracy one. He wants redistribution only out of the need to help the poor economically (a purely utilitarian argument in my view). But back to the margins: the intensive margin changes little with wealth taxes, income taxes. In fact, you could make things worse in some ways. Once the extensive margin exists, it is there and it is done. Wealth taxes have no effect on the extensive margin

The extensive margin exists because the state can confer privileges unconstrained by electoral competition, constitutions, checks and balances, secure property rights, etc. It is not about taxing wealth, it is about making it impossible to use wealth to buy influence.

The case for using wealth taxes to save democracy is really dubious.

Cheers,

V