Despite an extensive literature on the productivity gains associated with large cities in developed economies, little is known about this issue in the context of developing countries. Investigating agglomeration economies in the rest of the world is crucial for several reasons. First, the growth rate of the world’s urban population is being driven by urbanisation outside the developed world (OECD and EU 2020). Given the importance of cities to the national economies of these countries, understanding the drivers of agglomeration is relevant for boosting national productivity through government policy, reducing regional disparities, and improving the wellbeing of millions of urban dwellers.
Second, one cannot assume that the existing models for the developed world can be applied to the agglomeration economies of the developing world (Chauvin et al. 2017). For instance, the rapid urbanisation observed in developing countries in the second half of the 20th century stands in stark contrast to the stability seen in the developed world (Glaeser and Henderson 2017). High rates of urbanisation, differences in the growth of cities, and their institutional or infrastructural qualities may affect the size and extent of the gains associated with larger and denser cities.
In recent work (Özgüzel 2022), I shed light on these critical differences by providing evidence from Turkey, a highly urbanised developing country with significant spatial inequalities. Among the OECD countries, Turkey has the highest regional disparity in terms of GDP per capita, although substantial inequalities also exist across almost all metrics, including life expectancy, broadband connection, health facilities, and housing quality (OECD 2020, Karakoç et al. 2020) (Figure 1).
Turkey’s rapid urbanisation experience since the 1950s and significant spatial differences in productivity make it an excellent case study for understanding the sources and consequences of agglomeration economies in highly urbanised developing countries.
I examine the determinants of wage disparities – a key measure of productivity – across Turkish provinces and regions over the period 2008-16. I follow the two-step estimation strategy proposed by Combes et al. (2008) that simultaneously considers a broad set of local factors that influences local productivity (e.g. skill composition of workers, Marshallian externalities, local non-human factors). I use social security records, a new administrative dataset that has become available to researchers recently and thus has never been used in research before. I address the endogeneity bias due to the reverse causality by using historical instruments based on census data from the Ottoman Empire and the early years of the Turkish Republic.
Figure 1 GDP per capita (top) and productivity differences (bottom) in Turkish provinces, 2016
Note: GDP per capita in US dollars (Panel A) is based on statistics provided by the Turkish Statistical Institute. Productivity differences (Panel B) correspond to province-year fixed effects estimated in the first-step and measure the productivity differences from the national average net of sectoral composition of the province. Provinces correspond to NUTS3 regions.
The analysis leads to several findings. First, I find the elasticity of wages with respect to employment density to be around 0.056–0.06 when controlling for local characteristics and addressing reverse causality. This means that doubling the employment density in a province increases the average wages (or productivity) of workers by 3.8–4.2%.
These numbers are much higher than the estimates in developed economies, for which elasticities range between 0.01 and 0.03 (Combes et al. 2008, Ahrend et al. 2017, De la Roca and Puga 2017). Compared with other developing countries, this elasticity is lower than that estimated for Chinese cities (Combes et al. 2015, Chauvin et al. 2017) and Indian districts (Chauvin et al. 2017), similar to that estimated for Colombian cities (Duranton 2016), and higher than that for Ecuadorian cities (Matano et al. 2020) and Brazilian microregions (Chauvin et al. 2017).
Second, I find a positive and strong effect of domestic market potential on labour productivity. The estimated elasticity is around 0.076–0.089, which implies that if the market potential of a province doubles, wages will increase by 5.3–6.4%. This result corroborates findings from other developing countries, which show that productivity gains associated with a larger market potential are stronger than for developed countries and that they matter for explaining differences in spatial productivity within countries.
Third, using the individual-level regional data, I find weak sorting effects across regions of workers according to their observable and unobservable skills. This result is in sharp contrast to what is usually observed in developed countries, where a large fraction of the explanatory power of city effects arises from the sorting of workers (Combes and Gobillon 2015). It is, however, very much in line with the results for China, where no sorting effects are evident (Combes et al. 2015, 2020), suggesting that urbanisation patterns may operate differently in developing countries and reinforcing the need for further evidence from such countries (Glaeser and Henderson 2017).
Overall, these results indicate significant differences between developed and developing countries in terms of drivers and consequences of agglomeration effects. These findings corroborate recent evidence that the main mechanisms of urban economies established in developed economies are also present in developing countries.
But they also show that the current models need to be extended to capture differences observed in developing countries, especially those which have experienced rapid urbanisation. The factors that differentiate the sources of urbanisation and gains associated therewith in developed countries compared with developing countries are significantly absent from the literature and thus remain high on my research agenda.
This blog post is based on data from the recently published Voxeu.
