xiv
Proof.
The elasticity of log employment size,
l
with
respect to management quality,
m
is
𝜕𝑙
𝜕𝑚
𝑖
=
1
𝑑−𝛼
. Since
𝑑 ≥ 1
and
𝛼 < 1
, this is positive.
Proposition 2.
𝜕
2
𝑙
𝑖
𝜕𝑚
𝑖
𝜕𝑑
< 0
.
The impact of management quality on firm employment size is
decreasing in the degree of frictions.
Proof.
This can be directly seen from Proposition 1. The
magnitude of the employment-
management elasticity is decreasing with the size of frictions,
d
.
Corollary.
𝜕
2
𝑙
𝑖
𝜕𝑚
𝑖
𝜕𝜇
< 0
and
𝜕
2
𝑙
𝑖
𝜕𝑚𝜕𝜏
< 0
. Increases in firm market power (falls in
η
cause a rise
in margins
𝜇
) and increases in distortions (
𝜏
) reduce the elasticity of employment with respect
to managerial quality. This is the key idea: as frictions increase,
the impact of better
management on firm size, although remaining positive, will decline.
Mapping the model to the empirics
There is a straightforward mapping of this set-up to the empirics. Table 1 on the production
functions is the multi-factor extension to equation (E1) where we also allow for capital and
skills to be other factors of production (this is a trivial extension to the production function).
Propositions (1) – (2) are unaffected
by including extra factors, so long as they are all
statically optimized (see below for a discussion of adjustment costs and dynamic factors).
7
The positive relationship between employment and management is shown in all of the tables
as well as Figures 1 and 3. The intuition behind the stronger relationship between
employment and management in the US than Mexico is that competition is higher and market
distortions
lower in the US, as in Proposition 2. In this case,
𝜏
and
𝜇
have (implicitly)
country-specific subscripts. Similarly, the stronger relationship
between size and
management in the Mexican Manufacturing sector than in the Services sector (Table 2 and
Figure 3) is that competition is stronger (due to international trade) and distortions lower (due
to fewer regulations) in manufacturing. In this case,
𝜏
and
𝜇
have (implicitly) sector-
specific subscripts.
The bulk of the paper uses other observables
to shift
𝜏 and 𝜇
. In Table 3, we argue that the
drive time to the border is a municipality-specific indicator of competition. Firms located
closer to the US face effectively a greater degree of potential competition from US firms,
with a larger substitution possible for consumers (Proposition 2). Hence,
𝜇
is lower, for these
Mexican firms, so the relationship between employment size and management is stronger.
This is equivalent to introducing an area subscript, i.e.,
𝜏 = 𝜏
0
+ 𝜏𝐷𝑟𝑖𝑣𝑒
𝑚
where
𝐷𝑟𝑖𝑣𝑒
𝑚
is the drive time to the US border in municipality
m
. Similarly, the argument that a larger
market size in a city
c
will mean greater density and therefore more spatial competition in
7
There are analogous conditions for capital inputs and output. Capital is harder to measure of course as it the
volume of output as we do not have firm specific price deflators. This is why we prefer to focus on labor as our
key firm size measure.
xv
the Service sector (which, unlike manufacturing, is predominately locally traded), assumes
𝜇 = 𝜇
0
+ 𝜇𝑆𝑖𝑧𝑒
𝑐
. Finally, the frictions in Table 4 are also assuming that the distortions are
shifted by the institutional environment in a geographical area.
Some Theoretical Extensions
There are multiple extensions one could make to the baseline model.
First, the simplest approach to extending the model, is to consider a sunk cost to entry before
firms observe their realization of (stochastic) management as in the Melitz (2003) model. In
this way, we observe young firms for a period before they exit if they have a low draw of
management. The implication of this type of model is that (i) older surviving firms will have
on average higher management scores and (ii) the variance of management practices for a
cohort will shrink over time, as the lower tail of worst managed firms exits. The empirical
moments in Figure 4 are consistent with point (i) and those of Figure 5 with point (ii).
Second, note that the set-up in Bartelsman et al. (2013) is close to our approach here as it
emphasizes the robustness of the "Olley-Pakes moment" - the positive relationship between
relative size and productivity – as a measure of reallocation. This is the same as our approach,
except we have explicitly substituted in management rather than used productivity proxies
as they do. Their framework generalizes our approach as in addition to the sunk cost of entry
(as in the previous paragraph), they also allow for adjustment costs in capital. This creates a
dynamic optimization problem for capital investments. Since there is no closed-form
solution, they use numerical simulations to show similar results to our Proposition 2: in
environments
with greater distortions, there will be a weaker relationship between
management (TFP in their model) and firm size.
Bartelsman et al. (2013) keep TFP/management exogenous. Bloom et al. (2017) generalize
their approach even further by allowing management to be endogenously chosen with
adjustment costs (like capital investment). The dynamic optimization problem generates a
policy correspondence for the investment decisions of both dynamic factors.
The state
variables are managerial capital, non-managerial capital, and TFP (which is modelled as an
exogenous Markov process). Even in this much more complex set-up, they show that the key
intuition behind propositions (1) and (2) as well as the dynamic implications between firm
age and the level and variance of management in Figures 4 and 5.