11
These relationships are confirmed within the Mexican data in a firm-level regression with
ln(employment) as the dependent variable in Table 2. Consistent with Figure 3, column (1)
shows that the coefficient on management in services is 1.6, while for manufacturing, it is
significantly higher at 2.752 (2.752 = 1.622 + 1.131). Column (2) controls for education,
region, industry, and time dummies and again finds the same ranking. Columns (3) through
(6) repeat the analysis but split the samples into manufacturing and
services and confirm the
result of a stronger management-size relationship in manufacturing.
Another piece of evidence pointing towards the existence of misallocation in Mexico,
especially in the services sector, is the relationship between management and firm age that
we again show separately for US manufacturing,
Mexican manufacturing, and Mexican
services in Figure 4. Here, each age bin plots the average management score in deviations
from each sample mean. Dynamic reallocation implies older firms should have higher
management scores as the poorly managed firms should have been
selected out at a younger
age (or improved their management quality as time passes). Consistent with this, we see in
Panel (a) that older American manufacturing firms have higher management scores, with the
implied “shakeout” being particularly strong over the first five years. Similarly, there is an
upwards slope in Panel (b) between management and age in Mexican manufacturing firms,
albeit slightly weaker than in the US. In principle, this could also be due to learning as firms
get better with age. What is striking is that in Panel (c), there is actually a negative gradient
between management scores and age in Mexican firms in the service sector. Since it is
unlikely that service firms have nothing to learn, this seems more likely to be signaling the
absence of strong selection effects in Mexican services.
This is consistent with Hsieh and Klenow (2014), who argue that the size-age relationship is
an
indicator of misallocation, as in services, clearly, market selection is not effectively
leading inefficiently managed (smaller) firms to exit across their life cycle.
23
23
This is also consistent with the observation that Levy (2018) makes about observing higher misallocation in
the services sector as the productivity gap for firms within this sector is higher than in manufacturing.
Regressing the TFP index over age groups, productivity appears to decrease over age groups, especially in the
case of services, starting at 16 years of age. In the case of manufacturing, mostly non-significant coefficients
are observed.
12
A further piece of evidence is shown in Figure 5. If selection is important, we would expect
that the variance of management across firms falls in older cohorts as the worst managed tail
is selected out. This is certainly the case in US manufacturing in Panel (a) where the standard
deviation falls sharply in older age bins. By contrast, in Panel (b) for Mexican manufacturing,
the spread of practices within each age bin is
not
decreasing. Finally, in Panel (c) for Mexican
services, the spread of management
practices is actually
rising
with age.
To summarize, we have shown that despite a similarly strong association of productivity with
management across countries (US vs. Mexico) and sectors (manufacturing vs. services), the
reallocation of economic activity to better managed firms appears (i) weaker in Mexico than
in the US and (ii) weaker in Mexican services than in Mexican manufacturing. We see this
when looking at the size-management correlation and the relationship between firm age and
the first moment (mean) and second moment (variance) of management.
We also consider some more dynamic indicators of misallocation in Online Appendix Table
A4.
24
In the first two columns of Panel (a), we see that employment growth is higher for firms
with greater management scores for both manufacturing and services, but it is significantly
higher for manufacturing (test of the difference p-value = 0.008). Similarly, the last two
columns indicate that better-managed firms are significantly less likely to exit, but this
relationship is stronger for manufacturing than services (p-value = 0.028). Panel (b) shows
that there are similar patterns when using productivity
instead of management, but the
relationship is weaker (e.g., the difference is not statistically significant for employment
growth).
25
These dynamic effects corroborate the selection
mechanisms underlying our
interpretation of Figures 4 and 5.
Dostları ilə paylaş: