Skyscrapers and the economy in Latin America

Abstract

Skyscrapers are an intellectual challenge for urban analysis because of their imposing visual presence in the city landscape, and because of their environmental and real estate impacts. These characteristics however have not received much attention in the literature, particularly in analyses about Latin American cities. In this paper, we describe and test four theories about record-breaking buildings height: traditional microeconomic theory, game theory, business cycle, and global cities. We use a 2000–2012 panel database of 29 cities from 10 different Latin American countries, in order to contrast the contesting explanations about buildings’ height. We design a baseline model and then, using four different estimation techniques and diverse specifications, find that traditional theory and more strongly, global cities are good predictors of buildings’ height.

Keywords:

• Urban economics
• Latin America
• skyscrapers
• urban studies
• global cities

Notes

1. And, in similar fashion, households trade off space against commuting costs to the city centre, higher density of use (and hence building height) compensating for higher land costs in the centre.

2. That is, average building heights increase with maximum building height.

3. That is a finding in line with the ideas about cycles and construction expressed by Shiller (2005).

4. Barclays Capital – Equity Research. Skyscraper Index: Bubble building. 10 January 2012.

5. The time trend is used as a proxy because more direct technology variables such as Total Factor Productivity or United Nations Industrial Development Office (UNIDO) manufacturing complexity index were never significant in any estimation framework.

6. The full list of cities and top buildings height are reported in the Appendix 1 to this paper. There were nine cases where the record holder in 2000 was less than 100 m. We have added these buildings to the database so that we can produce complete series of tallest building replacements per city for the full period 2000–2012.

7. We need to rely on the structural height in metres, because it was impossible to construct comparable databases of more precise variables like the average height in the high buildings district per city, or the Floor-to-Area-Ratio in the plot of land with the tallest building.

8. The H–H (Herfindahl–Hirschman) indicator is defined in our case as ${\sum }_{i=1}^n{\left(\frac{X_i}{X}\right)}^2$, where X_i is the population of each administrative spatial unit, and X is the population of the entire metropolitan area.

9. Most of the record-breakers are located in the central municipalities of their corresponding metropolitan areas, because of their economic potential or global character. As we already include those variables in the regression, the central municipality dummy proxies a pure location effect due to regulation.

10. Results are available from the authors on request.

11. Or at least, we expect it to be systematically associated. A measure of its strength as a predictor of the heights.

12. Global and World Cities Research Centre – University of Loughborough. We are conscious of some criticisms to this classification system, although we have not found an alternative source with a panel structure and a comparable calculation structure. In addition, the GaWC model explicitly identifies the connectivity of cities in the world economic system, which is our hypothesised predictor of heights.

13. We have however run the same specifications when replacing Area with Density (its semi-panel derived variable), and when not excluding Unesco and Population H–H in these regressions. The signs and significance of the estimated parameters do not change. We use instrumental variables in order to be reasonably sure that there are no indirect causation effects between the explanatory variables.

14. Greene (2004) detects that truncation will produce biases in the estimation of the β vector and the fixed effects; however, in our simple pooled estimation we have only one of these two sources of bias, as we do not have fixed effects.

15. Quantile regressions (results available from the authors) allowed us to detect that our estimates are also max-consistent in terms of signs and significance of the parameters, although there is a significant change of the parameters from the median to the 5th quintile.

16. Albeit one strongly influenced by its offshore financial role.