Management and Misallocation in Mexico

 
Exposure to trade within the Manufacturing sector 
The contrast of the proximity results across manufacturing and services is interesting, but of 
course, there are many other differences between these broad sectors that could be generating 
the heterogeneity of the drive time coefficient that we have not controlled for. To better test 
our hypothesis, we build a NAICS-6 digit industry-level indicator of the export share of sales. 
If our hypothesis that proximity to the US reflects the stronger effects of competition from a 
bigger market, then the proximity coefficient should be stronger for firms in industries that 
are more open to international trade.
Adding an interaction between the industries' export intensity and drive time (along with 
interactions with other observable covariates) shows that US proximity has a significantly 
stronger association with management for the more export-intensive sectors. We observe the 
same relationships for TFP as for management. 
We tested the robustness of all these results by re-estimating this equation using a ten-year 
average sectoral share of exports with information from the 2004 and 2014 Economic 
Censuses as well as export averages using the 2004 and the 2009 Economic Censuses. The 
results are robust to these changes in the measure of sectoral export orientation. 
2
We also tested the robustness of our result using a different definition of regions, and we observe that the 
results do not change. 

---

xi 
An additional concern is that the results might be driven by the importance of two main states: 
Mexico City, which is located in the central region of the country, and Nuevo Leon, which 
is located near the Northern border of the country. According to the 2014 Economic Census, 
these two states account for 17% of Manufacturing Gross Product and 34% of Services Gross 
Product. The results do not change significantly when we estimate the same equations but 
exclude these states. 
 
City Density and Reallocation in the Service Sector: Robustness Tests 
Testing the robustness of our results on market size, we estimate the relationship between 
management and market size in a regression framework. The coefficient on the linear city 
size variable is small, negative, and insignificant, whereas the interaction between the 
services dummy and market size is positive, large, and statistically significant. A similar 
pattern holds when we condition on other covariates or split the sample by manufacturing 
and service sectors. We further assess the robustness of these results by using alternative 
proxies for market size based on population size. We use a dummy that indicates whether the 
municipalities are small urban areas, medium, metropolitan areas, or large metropolitan 
areas. As the municipalities grow in market size, the management score for the services sector 
improves, but the same is not observed for manufacturing.
In an alternative specification using the TFP index as the dependent variable, we find that 
local city size matters for productivity in services, but not in manufacturing. The absence of 
a city-size effect on productivity may appear surprising, as there is a vast economic 
geography literature that argues for higher productivity effects in large cities. It is worth 
noting, however, that most individuals and firms are not in manufacturing, so some of the 
existing empirical studies are likely driven by the services sector.
3
Furthermore, most of the 
studies are in high-wage countries where agglomeration effects may be stronger than in a 
middle-income country like Mexico. 
One concern with our results is that the average income could reflect the presence of more 
skilled potential employees. We disaggregate our market size measures by including income 
and population density as separate variables to address this. We find that our local size effects 
are driven by population density that has a similar statistically significant coefficient, but 
income is insignificant.
4
It is possible that these results are driven by reverse causality or omitted location-specific 
confounders. To partially address this, we use population density and income in the 1990 
Population and Housing Census as an instrument for the 2010 market size data. The results 
do not change much. 
Finally, we estimated the same equations again, excluding Mexico City and Nuevo Leon, 
and confirm that our results are robust to excluding these two main cities. 
3
A smaller literature uses plant and firm level data in manufacturing. Some of these studies do find significant 
and positive city size effects, but all the ones that we know of are in high wage countries (e.g., Combes et al., 
2012). 
4
As before, the variables are insignificant in the manufacturing sector
.

---

xii 
Institutional Frictions and Misallocation: Robustness Tests 
We analyze the robustness of the results presented in Figure 12 and Table 4, by using a 
different threshold (5%) to define a high level of contract enforcement problems, kidnapping, 
corruption, and our business crime composite index. Our results are robust to using this 
alternative threshold.