A cityscape of Manchester at night.

City size and productivity.

Tom Forth, .

A calmer version of a tweet thread by me from 2023.

Bigger French cities have stronger economies.

Bigger French cities have stronger economies. OECD.

It’s not a perfect correlation and the methods for measuring the size of a city and the strength of its economy are varied and contested. But for a pattern in economics this is about as good as we get.

The best fit line on the graph above shows that doubling the size of a French city typically increases its productivity by 15%. Or in economics language, the elasticity of agglomeration in France is 15%.

Economics is almost always complex, causation almost always circulates rather than flows, and attempts to map and quantify relationships are always partial, poorly measured, and quickly out of date. It is often useful for avoiding low quality replies dressed up as profound thoughts to start pieces like this with that reminder. Appeals for more complexity rarely add value.

What are agglomeration benefits and why do they exist?

There are lots of reasons why larger cities generally have stronger economies.

A simple but important one is that cities with strong economies attract people to them, growing their populations. Cities with weak economies tend to lose people and shrink.

Slightly more complicatedly, cities increase the size of labour markets. Cities are attractive to employees because they offer a larger number and a greater variety of jobs to pick from. People can make better use of their skills and interests and get promoted more quickly in bigger cities. Relatedly, cities are attractive to employers because they offer a larger number of workers to pick from. Businesses can hire from a larger pool, pick better staff better suited to their culture or niche of operation, and grow more quickly in bigger cities. This feature of bigger cities tends to favour higher skilled people and the firms that employ them. Bigger cities tend to disproportionately attract higher skilled people and employers of such people correspondingly. This effect is particularly strong in high valued added service industries.

More complicated still, beyond just attracting people with more experience and skills, cities incentivise existing residents to acquire new skills. There is less reason to train to be a stock broker in Hull than there is in London or Amsterdam. Not only is the market value of stock broking skills higher in London and Amsterdam but it is also more visibly higher.

As a final example, larger cities, especially where they are dense, and especially where they are dense in specialised areas, lead to more and more valuable economic interactions, both planned and unplanned, between individuals, institutions, and businesses. These interactions lead to an increase in innovation and entrepreneurship which generates new faster growing businesses which demand high skilled workers in emerging economic sectors which skew towards higher productivity industries.

There are many more ways that city size can interact with determinants of city productivity over varying durations. We are just scratching the surface and already we are beyond our capacity to map and measure the economic system we’re describing with much certainty.

That’s enough theory, let’s look at some data.

Education, industry, and productivity.

While it is a deeply flawed proxy for skills, percentage of graduates in a city is one of the few widely available and comparable measures of the skills of a population.

French cities with more graduates have stronger economies.

French cities with more graduates have stronger economies. OECD, Eurostat.

It follows mathematically that bigger French cities have more graduates.

Now let’s look at economic sectors.

French cities with a higher share of employment in industry (and correspondingly lower shares of employment in services) have weaker economies, though this relation isn’t as strong as the previous relationships.

French cities with more industry have weaker economies. OECD, INSEE EMP T8.

Again, it follows mathematically that bigger French cities have a lower share of employment in industry. This is consistent with findings that service sector businesses typical benefit more from agglomeration benefits than manufacturing sector businesses which they then crowd out from large cities.

So we know that skill level of the population (measured via the proxy of graduate share), sectoral mix (measured via the proxy of industry share), and labour market size (measured via the proxy of population within the functional urban area) are all correlated with economic strength (measured via the proxy of GDP/capita).

It is understandable at this point that economists start asking how much each matters.

Specifically, as relates to France, it might seem reasonable to ask in what mix Paris has a very strong economy because it,

This question is the basis of a very large number of economic studies around the world. They typically find that the 15% agglomeration elasticity I showed for France earlier is about two thirds explained by skill levels, with other factors including the sectoral mix and the size of the city making up the remaining 5%.

A meta-analysis of 47 studies in 2018 left less than a third of the raw agglomeration benefit to city size, sectoral composition, and the rest — just 4.6%. Other similar analyses report agglomeration elasticities of between 2% and 5%.

The Power of Place.

The Power of Place, a recent paper for The Resolution Foundation uses much more and much better data combined with improved techniques to reinforce this finding. After controlling for skills, industry mix, and more, they report an agglomeration elasticity of just 3.9%. In their words, “a doubling of the size of the labour market boosts the pay of a typical worker by around 3.9 per cent”.

In writing up this work for the Financial Times, Chris Giles picked out the impressively underwhelming statistic from the report that “if good jobs moved to Birmingham and Manchester so that these cities eliminated their underperformance in pay and productivity, it would increase pay locally by 3%, boosting total productivity in England by about 0.1%”. He was right to note that “that is objectively small, providing half the boost to productivity that the Office for Budget Responsibility estimated would flow from the Labour government’s planning reforms alone”.

This led to much surprise online, including among some important policy making economists in Britain. Fixing the underperformance of Birmingham, Manchester, and by extension similar cities in Britain was not, as people like me claim, a path to British economic overperformance.

I am happy to bring much better news. I think the untapped potential of our large cities is about 100 times as large as that snippet suggests. I am confident that fixing the underperformance of Britain’s cities like Manchester and Birmingham would grow our economy by at least 7% above trend.

My argument begins with Chris’ own sentence — “one danger on display here is that the data is so interesting, and modern econometrics and computers so powerful, that we can generate huge quantities of statistics without gaining much insight”. I have little issue with the data or the mathematics of the paper’s work on agglomeration elasticities. I disagree that The Resolution Foundation paper measured the underperformance in pay and productivity of Birmingham and Manchester in a useful way, but within the context of the paper, that statement is justified.

My core issue is that I disagree with a widely shared assumption in the economics profession on how we should measure agglomeration elasticities. That assumption is summarised well in a line in Ahrend 2014 which warns that failing to control for skill levels in cities when calculating agglomeration elasticities “will lead to confounding agglomeration benefits with productivity increases from a more skilled workforce”.

I think that such confounding is both inescapable and essential. Doing so increases our estimates of the economic benefit we should expect from reducing the underperformance of Birmingham and Manchester by two orders of magnitude.

The peril of decomposing complex systems.

I started this piece by describing a small part of the complex system that makes larger cities more productive. In the simpler examples I argued that bigger cities, by offering a more liquid labour market, offer some productivity advantages through the better matching of a fixed set of employees with a fixed set of employers. This is the largest part of the agglomeration benefit left as a feature of a place’s size once confounding variables have been removed.

I am not at all surprised that this simple and largely fixed feature of a large city is only a small part of the benefit of its size. It is the dynamic effects that contribute much more to economic growth in the longer term.

What matters more is the ability of larger cities to offer better amenities to residents — larger football stadiums, a wider choice of schools and healthcare, better restaurants, more niche cultural activities — that retains skilled workers who would otherwise leave and to attract skilled workers from elsewhere.

What matters more is the ability of larger cities to increase interactions that create new innovations and businesses that create the employment demand in emerging sectors of the economy that encourage residents to acquire more skills.

What matters more is the ability of larger cities to enable these innovative firms to hire fast and grow quickly and change the sectoral mix of the city to a more productive one.

The greater ability of large cities to train, retain, and attract skilled workers and the firms that employ skilled workers is a core input to their greater productivity and a largely inseparable result of their size. I don’t think that we can understand the dynamism of cities without confounding skills, sectoral composition, and much else into a measurement of the effective size of a city.

I insist that both the best and the most useful measure of the French agglomeration elasticity is 15%, not 3%.

But this disagreement is only part of why my interpretation of the excellent work in the Resolution Foundation paper is so different to that of the authors and the Financial Times.

The curious lack of agglomeration benefits in Britain.

Agglomeration benefits are common. The USA has a very similar curve relating city size to GDP/capita as France, giving an estimated agglomeration elasticity of 11%.

Larger US cities have stronger economies. OECD.

A combination of the cities of the Low Countries and the Nordic Countries (no country on its own has enough cities to generate a useful estimate) gives an estimated agglomeration elasticity of 17%.

Larger Nordic and Low Country cities have stronger economies. OECD.

By contrast, the UK’s cities show an estimated agglomeration elasticity of just 4%. This number is not statistically significant and if London is removed from the regression it becomes negative.

Bigger British cities barely have stronger economies. If London is removed from the regression the relationship is negative. OECD.

The Resolution Foundation does something more complicated than I do here, but it’s not that different.

They take an estimated agglomeration elasticity very similar to the one I have drawn here and they ask what would happen if the UK’s second and third largest cities increased their productivity to match the trend. In the graph I’ve drawn below, Manchester’s GDP/capita would have to grow 6% to match the productivity expected by the UK’s tiny and statistically insignificant agglomeration elasticity estimate. Birmingham’s GDP/capita would have to increase by 22%. If this happened, in my model where I don’t reduce the agglomeration elasticity by controlling for skills, UK GDP/capita would increase by 1.2%.

Still small.

If the six largest British cities outside of London enjoyed agglomeration benefits the same size as French cities the UK would have 7% higher.

But this is neither a statistically sound ambition (the UK’s agglomeration elasticity is not statistically significant) nor anywhere near an ambitious enough one.

The UK’s large cities should expect to enjoy similar agglomeration benefits to similar sized US and French cities.

Sheffield should be like Pittsburgh or Bordeaux. Newcastle and Liverpool should be like Nantes or Baltimore. These are not ridiculous ambitions. East Germany’s cities have achieved about the same catch up growth in the last 25 years as would be required of British cities. I don’t accept that Dresden, Liepzig or Chemnitz have some special character that Sheffield, Leeds, or Manchester lack.

If we could achieve these agglomeration benefits in Britain for our six largest cities outside of London, they would grow by between 33% for Manchester and 52% for Birmingham. In total, since these six cities are together home to 15 million people, the UK economy would grow by 7%. This is nearly one hundred times larger than the suggestions in the Resolution Foundation paper.

If we set our sights as high as achieving the agglomeration benefits we see in Belgium, The Netherlands, Denmark, Sweden, Norway, and Finland we could double all of these numbers again. It’s hard to imagine now, but it’s not ridiculous.

But how could we do it?

Achieving agglomeration in Britain

I think we know the answers. We also know that we have never tried them.

It’s notable that compared to the French, Dutch, Belgian, Nordic cities that outperform them, British cities,

We know from comparisons with East Germany that rapid catch up growth is possible. The approach which worked there was to focus on physical infrastructure, R&D, and local power, not direct investment in skills as has been tried with very disappointing results in the UK. We should try that.

 

blog comments powered by Disqus