52nd Congress of the European Regional Science Association: "Regions in Motion - Breaking the Path", 21-25 August 2012, Bratislava, Slovakia
In this paper I measure the impact of urbanisation and localisation economies on commercial rents (offices and industrial buildings) using unique micro-data on both commercial property transactions, as well as all firm (establishment) locations. I add to the literature in three important ways. First, I use measures of urbanisation and localisation that are continuous over space to overcome the modifiable areal unit problem. More specifically, I estimate kernel employment densities that are continuous over space. Second, to distinguish agglomeration economies from unobserved endowments, I include postcode fixed effects (about the size of a census block) and use temporal variation in densities to measure the impact of urbanisation and localisation economies on commercial property values, assuming that unobserved endowments are time-invariant. To relax the latter assumption I also test robustness by including time-variant endowments (e.g. new station openings, changes in land use) and region-year fixed effects. This is arguably preferable over standard IV approaches using on long-lagged instruments, which may be invalid. Third, I determine the spatial extent of agglomeration economies. Using a cross-validation procedure, the spatial decay parameter is estimated within the model. The results show that urbanisation economies, rather than localisation economies, have a considerable impact on rents: a standard deviation increase in employment density leads to an increase in rents of about 10 percent. This effect is larger in magnitude than estimates of agglomeration economies using wages, suggesting that agglomeration economies mainly capitalise in rents. The geographical extent of these benefits is about 15 kilometres and very robust across different specifications. So, urbanisation economies seem to mainly take place at a city or municipality-level, rather than at a more aggregate level. I furthermore show that the bias of ignoring unobserved endowments seems to be rather small.