MANAGEMENT AS A TECHNOLOGY?
Nick Bloom (Stanford), Raffaella Sadun (HBS) and
John Van Reenen (LSE)
Motivation: What is Management?
• Evidence of massive national and plant spread in TFP:
e.g. Hall and Jones (1999) Syverson (2004)
• Can management explain any of this: 1%, 10% or 100%?
• Two main theories take opposite positions:
– Management as Technology (MAT): management key
driver of performance gaps (Walker, 1887, Marshal 1887)
– Management as Design (MAD): management optimized
so less important (Woodward, 1958, core Org-Econ)
MAT v MAD – are management difference optimal?
Ohio, USA
Maharashtra, India
Summary of the paper
1. Management data from ≈ 15,000 firms in 30 countries
– US highest (unweighted) average management score
– US lead ≈50% larger if size weight (more reallocation)
2. Are these variations technology (MAT) or optimal (MAD)?
Develop three predictions on: performance, competition and
reallocation, and find best fit for MAT
3. Given MAT, then estimate management can account for
possibly ≈ 25% of plant and country spread in TFP
Management Data
Management Models
Testing the models
Management and cross-country and firm TFP
Survey methodology (following Bloom & Van Reenen (2007))
1) Developing management questions
• Scorecard for 18 monitoring and incentives practices in ≈45
minute phone interview of manufacturing plant managers
2) Getting firms to participate in the interview
• Introduced as “Lean-manufacturing” interview, no financials
• Official Endorsement: Bundesbank, RBI, PBC, World Bank etc.
3) Obtaining unbiased comparable responses, “Double-blind”
• Interviewers do not know the company’s performance
• Managers are not informed (in advance) they are scored
Example monitoring question, scored based on a number of
questions starting with “How is performance tracked?”
Score (1): Measures
tracked do not
indicate directly
if overall
business
objectives are
being met.
Certain
processes aren’t
tracked at all
(3): Most key
performance
indicators
are tracked
formally.
Tracking is
overseen by
senior
management
(5): Performance is
continuously
tracked and
communicated,
both formally and
informally, to all
staff using a range
of visual
management tools
Example incentives question, scored based on questions
starting with “How does the promotion system work?”
Score (1) People are
promoted
primarily upon
the basis of
tenure,
irrespective of
performance
(ability & effort)
(3) People
are promoted
primarily
upon the
basis of
performance
(5) We actively
identify, develop
and promote our
top performers
Survey randomly drawn firms from the population of
larger (50 to 5,000 employee) manufacturers across
countries (including public and private firms)
Locations of plants surveyed 2004-2008
Wide spread of management: US, Japan & Germany
leading, and Africa, South America and India trailing
United States
Japan
Germany
Sweden
Canada
Great Britain
France
Italy
Australia
Poland
Mexico
Singapore
New Zealand
Northern Ireland
Portugal
Republic of Ireland
Greece
Chile
China
Brazil
Argentina
India
Colombia
Kenya
Zambia
Nicaragua
Ethiopia
Ghana
Tanzania
N=1289
N=176
N=658
N=403
N=412
N=1208
N=632
N=313
N=454
N=364
N=515
N=306
N=150
N=136
N=307
N=160
N=269
N=581
N=755
N=1111
N=558
N=840
N=127
N=120
N=50
N=74
N=122
N=87
N=80
2
2.5
3
Average Management Scores, Manufacturing
Africa
Asia
Australasia
Europe
Latin America
North America
3.5
Note:survey
Firms between
50 of
and9/20/2013.
5000 employees,
Raw
datanot yet included in the paper
Data includes 2013
wave as
Africa
data
Some quotes illustrate the African management approach
Interviewer “What kind of Key Performance Indicators do
you use for performance tracking?”
Manager: “Performance tracking? That is the first I hear
of this. Why should we spend money to hire someone to
track our performance? It is a waste of money!”
Interviewer “How do you identify production problems?”
Production Manager: “With my own eyes. It is very easy”
3.5
Average management scores across countries are
strongly correlated with GDP
Africa
Australasia
Asia
United States
Europe
Japan
Germany
Sweden
Canada
Latin America
3
North America
management x log of GDP PPP per capita
United Kingdom
ItalyFrance
Australia
Poland
Mexico
China
Chile
Brazil
Singapore
New Zealand
Portugal
Republic of Ireland
Greece
Argentina
2.5
India
Colombia
Kenya
Zambia
Ethiopia
Ghana
Nicaragua
2
Tanzania
7
8
9
10
11
Log of 10-yr average GDP based on PPP per capita GDP(Current int'l $ - Billions)
Note: April 2013, World Economic Outlook (IMF) indicator
Data includes 2013 survey wave as of 9/20/2013. Africa data not yet included in the paper
Like TFP management also varies within countries
2 Japan
3 Germ any
4 Sweden
5 Canada
6 Great Britain
7 France
8 Italy
9 Aus tralia
10 Poland
11 Mexico
12 Singapore
13 New Zealand
14 Northern Ireland
15 Portugal
16 Republic of Ireland
17 Greece
18 Chile
19 China
20 Brazil
21 Argentina
22 India
23 Colombia
24 Kenya
0
.5
1
0
.5
1
0
.5
1
0
.5
1
1 United States
1
26 Nicaragua
27 Ethiopia
28 Ghana
3
4
5
29 Tanzania
0
.5
1
25 Zambia
2
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
Firm Average Management Score
Graphs by rank_cty
Data includes 2013 survey wave as of 9/20/2013. Africa data not yet included in the paper
Management Data
Management Models
Testing the models
Management and cross-country and firm TFP
Two perspectives on these management
differences in economics
• Management as a Technology (MAT)
– Management part of firm’s TFP, often said to explain
a firm’s TFP fixed effect (Mundlak, 1961)
– Think of improvements as “process innovations”,
hence the sense that management is a technology
Management as a technology: Francis Walker
“It is on account of the wide range [of ability] among the
employers of labor that we have the phenomenon in every
community and in every trade some employers realizing no
profits at all, while others are making fair profits; others, again,
large profits; others, still, colossal profits.”
Francis Walker (QJE, April 1887)
Walker ran the 1870 US Census, and was the
founding president of the AEA and president of MIT
Management as a technology: Alfred Marshall
“I am very nearly in agreement with General Walker’s Theory
of profits….the earnings of management of a manufacturer
represents the value of the addition which his work makes to
the total produce of capital and industry....”
Alfred Marshall (QJE, July 1887)
Two perspectives on these management
differences in economics
• Management as a Technology (MAT)
– Management part of firm’s TFP, often exampled as
explaining firm’s TFP fixed effects (Mundlak, 1961)
– Think of improvements as “process innovations”, hence
the sense that management is a technology
• Management as Design (MAD)
– Management is optimally chosen
– Popular in Organizational Economics (Gibbons and
Roberts HOE, 2013), and other fields like Strategy and
Management (Contingency theory, Woodward 1958)
So we define 2 highly stylized management models
Production Function: Y=AKαLβG(M), M = management
Flavor 1: Management as Technology (MAT)
G(M)=Mγ ; pay for M which depreciates (like R&D)
Flavor 2: Management as Design (MAD)
G(M) = 1/(1+|M*-M|); M* = “optimal” choice, M is free
• Otherwise common set of assumptions
– Changing M & K involves adjustment costs (L flexible)
– τ % of sales lost to distortions (bribes, regulations etc)
– Monopolistic competition (Isoelastic demand)
– Sunk entry cost (E) & fixed per period operating cost (F)
We simulate these two management models
1. Entrants pay a sunk cost E for a draw on (A,M,τ), with
rate of entry determined by the free entry condition
2. All firms get a shock to TFP: At=ρAt-1 + εt
3. Pay fixed operating cost F per period (or exit)
4. Invest in M & K, plus choose optimal labor
1
1
Result 1: Performance (size, TFP, profits) increase
in management with MAT but not MAD
.6
0
.2
.4
Sales
0
.2
.4
Sales
.6
.8
MAD: Y=AKαLβ/(1+|M-M*|)
.8
MAT: Y=AKαLβMγ
1
2
3
Management
4
5
1
2
3
Management
4
5
Notes: Results from simulating 2,500 firms per year in the steady state taking the last 5 years of data.
Simulations are identical except that the production functions differ as outlined above. Plots normalized
log(management) in the simulation data onto a 1 to 5 scale and log(sales) onto a 0 to 1 scale. Lowess
plots shown with Stata defaults (bandwidth of 0.8 and tricube weighting).
4
4
Result 2: Average Management increases with
competition (elasticity) with MAT but not MAD
MAD: Y=AKαLβ/(1+|M-M*|)
3
2.5
2
1
1
1.5
1.5
2
2.5
3
mean of ilogmanagement
3.5
3.5
MAT: Y=AKαLβMγ
2
4
6
8
10
Comp increasing
8
2
4
10
6
Comp increasing
Notes: Results from simulating 2,500 firms per year in the steady state taking the last 5 years of data for
each level of elasticity. Simulations are identical for MAT and MAD except that the production functions
differ as outlined above. Plots normalized ln(management) in the simulation data onto a 1 to 5 scale.
Result 3: Sales weighted management is higher in
less distorted economies with MAT but not MAD
MAD: Y=AKαLβ/(1+|M-M*|)
4.2
4.2
MAT: Y=AKαLβMγ
2
5
10
15
4.1
4
20
Distortions increasing
(% sales lost)
3.8
3.8
3.9
India
3.9
4
mean of ilogmanagement
4.1
US
2
5
10
15
20
Distortions increasing
(% of sales lost)
Notes: Plots log(management) scores weighted by firm sales. Results from simulating 2,500 firms per
year in the steady state taking the last 5 years of data for each level of elasticity. Simulations are identical
for MAT and MAD except that the production functions differ as outlined above. ln(management) in the
simulation data is normalized onto a 1 to 5 scale.
Very stylized models with obvious extensions
• Dynamics: maybe management also changes adjustment
costs, information (forecasting) and factor prices
• Multi-factor: currently 1-dimensional management, but
design view model could also be about sub-components
• Spillovers: Technology is (partially) non-rival so we should
see learning, information spillovers, clustering etc.
• Governance/ownership issues: family firms, FDI, etc.
• Co-ordination: e.g. Gibbons & Henderson (2012)
Measuring Data
Management Models
Testing the models
• Performance
• Competition
• Reallocation
Management and cross-country and firm TFP
TABLE 2: Performance is robustly correlated with
management (consistent with MAT)
Dependent
variable
Ln(sales) Ln(sales) Ln(employ Profit rate 5yr Sales
-ment)
ROCE
growth
Exit
Exit
OLS
Fixed
Effects
OLS
OLS
OLS
OLS
OLS
Firm sample
All
2+ surveys
All
All
Quoted
All
All
Management
0.150***
0.033**
0.338***
1.202***
0.039***
-0.006***
-0.002
Ln(emp)
0.645***
0.374***
Ln(capital)
0.307***
0.237***
Competition
0.094**
Management
*competition
-0.096**
Obs
8,314
6,364
15,608
9,163
8,365
7,532
7,532
M, Management Index is average of all 18 questions (sd=1). Other controls include % employees
with college degree, average hours worked, firm age, industry, country & time dummies & noise
(e.g. interviewer dummies). Standard errors clustered by firm.
TABLE 3: Performance: use the Great Recession as a natural
experiment and find again well managed firms perform better
Growth in firm sales
Dependent variable
Shock (Industry Sales)
-0.033**
(0.014)
-0.035**
(0.014)
0.027**
(0.011)
Management*SHOCK
-0.051***
(0.014)
Management
0.002
(0.006)
-0.014
(0.010)
0.001
(0.006)
-0.052***
(0.014)
0.018*
(0.010)
-0.008
(0.009)
Firms
Observations
1,567
1,653
1,567
1,653
1,599
1,685
1,599
1,685
Shock (Industry Trade)
Management*SHOCK
Notes: SHOCK is a binary indicator for a fall in exports or fall in sales in the SIC3 by
country cell from 2007 to 2009. All columns include controls for skills, firm and plant
size, noise, country and industry dummies. Management from 2004-2006
Table 4: Contingent performance: not consistent with MAD
(best to be at optimal industry/country point)
Dependent Variable
Ln(Sales) Ln(Sales) Ln(Sales)
Cell
SIC3
×Country
Country
0.236***
(0.032)
0.226***
(0.027)
Management
*I(M below cell ave)
Management
*I(M above cell ave)
F test of symmetry
above v below cell ave
Cell Clusters
Observations
ROCE
Growth
Exit
SIC3
SIC3
×Cty
SIC3
×Cty
SIC3
×Cty
0.219***
(0.034)
0.219***
(0.030)
0.226***
(0.039)
0.223***
(0.030)
1.681***
(0.528)
1.679***
(0.436)
0.038
(0.028)
0.044*
(0.024)
-0.018**
(0.005)
-0.014**
(0.004)
0.230
0.984
0.785
0.990
0.388
0.060
796
8,003
18
8,314
177
8,292
900
8,793
921
8,007
1,137
6,607
Notes: Regressions includes controls for country, SIC4 & year dummies, firm-age, average hours,
% with degrees, noise controls etc. SE clustered by firm. Productivity columns from regression of
ln(sales) as dependent variable with controls for ln(labor) and ln(capital). Only cells with 2+
observations used.
Measuring Data
Management Models
Testing the models
• Performance
• Competition
• Reallocation
Management and cross-country TFP
TABLE 5: Competition improves management
(consistent with MAT)
Competition
proxies
Import
penetration
Management
(estimated in levels)
0.805***
(0.236)
1- Lerner
Index1
0.608***
(0.230)
17.53*
(3.85)
# of
competitors
Balanced
panel
Obs
Management
(estimated in differences)
20.68**
(6.65)
0.120**
(0.052)
0.121***
(0.023)
No
No
No
Yes
Yes
Yes
2,657
2,819
2,789
412
429
432
Notes: Includes SIC-3 industry, country, firm-size, public and interview noise (interviewer, time,
date & manager characteristic) controls. Col 1,2, 4 & 5 clustered by industry*country, cols 3 & 6
by firm
Measuring Data
Management Models
Testing the models
• Performance
• Competition
• Reallocation
Management and cross-country TFP
Table 6: More reallocation to better managed firms in the US
where markets generally less distorted (consistent with MAT)
Dependent Variable
Management (MNG,US base)
MNG*Argentina
MNG*Australia
MNG*Brazil
MNG*Canada
MNG*Chile
MNG*China
MNG*France
MNG*Germany
MNG*Greece
MNG*India
MNG*Ireland
MNG*Italy
MNG*Mexico
MNG*New Zealand
MNG*Japan
MNG*Poland
MNG*Portugal
MNG*Sweden
MNG*UK
Observations
Employees
201.9***
5,842
Employees
359.7***
-270.9**
-259.3*
-211.7*
-169.3
-92.6
-84.9
-489.5**
-9.0
-355.9***
-145.4
-258.8**
-283.1***
-250.1*
-375.7*
-297.3**
-308.1***
-308.9***
-228.7*
-125.1
3,858
Sales Growth
0.092***
-0.134***
-0.145**
-0.101***
-0.131**
-0.150
-0.060
-0.085*
-0.080*
-0.089**
-0.066
-0.085
-0.092**
-0.075*
0.718***
-0.099**
-0.058
-0.109**
-0.068
-0.054
2,756
Notes: Includes year, country, three digit industry dummies, # management questions missing, firm age, skills and noise controls.
Table 7: Reallocation also stronger in countries with lower
labor and trade regulations (consistent with MAT)
Dependent Variable:
Management (MNG)
Management*Employment Protection
(World Bank Country Index)
Employment
231.46*** 356.73***
110.97*
(37.12)
(55.89)
(66.30)
-1.43**
(0.69)
Management*Trade Costs
(World Bank Country Index)
(0.69)
-0.18***
(0.05)
Tariffs
(country x country)
Management*Tariff
Observations
5,760
5,017
-4.96
(4.12)
-8.25**
(3.35)
1,559
Notes: OLS, clustered by firm; dependent variable is firm employment; Domestic firms only. Controls for
firm age, skills, noise, SIC3, country dummies, EPL(WB) is “difficulty of hiring” from World Bank (1=low,
100=high). Trade cost from World Bank (0=low,6=high). Last columns tariffs are in deviations from
industry and country specific mean, from Feenstra and Romalis (2012).
Measuring Data
Management Models
Testing the model
• Performance
• Competition
• Reallocation
• Extensions: skills
Management and cross-country TFP
Management and Education: use UNESCO World Higher
Education Database university locations (N=9,081)
Distance to the nearest university seems to matter for
firm skills and management
Dependent
Variable:
Drive time to nearest
university
Manage
Manage
ment
% firm
employees
with degree
ment
ment
OLS
OLS
OLS
IV
-0.049***
-1.534***
(0.019)
(0.423)
0.789***
3.190***
(0.082)
(1.113)
6,406
6,406
Manage
% employees with
degree in the firm
Observations
6,406
6,406
Notes: Clustered by 313 regions. In final column proportion skilled is instrumented with
distance to university. Include industry, regional and full set of firm and noise controls.
Measuring Data
Management Models
Testing the model
• Performance
• Competition
• Reallocation
• Extensions: skills
Management and cross firm and country TFP
Following MAT we can estimate rough contribution
of management to country TFP spread
1. Estimate country differences in size weighted management
2. Impute impact of this on differences in TFP
Requires many assumptions, so only magnitude calculation
First calculate the employment weighted difference
in management (from the US as baseline)
-.14
-.25
-.08
-.18
-.5
-.35
-.11
-.19
-.12
-.18
-.29
-.36
-.26
-.26
-.22
-.27
-.49
-.49
-.81
-.83
-1
-.74
-.98
-1.02
-1.18 -1.19
-1.2
-1.5
Management gap with the US
0
0
-2
-1.65
us
sw
jp
ge
ca
gb
po
Reallocation
mean of rel_OP
it
fr
br
cn
ar
pt
gr
Within
of rel_zman
mean Firm
Notes: Total weighted mean management deficit with the US is the number on top of bar. This is decomposed into (i) reallocation
effect (blue bar) and (ii) unweighted average management scores (red bar) . Domestic firms, scores corrected for sampling bias
First calculate the employment weighted difference
in management (from the US as baseline)
-.14
-.25
-.08
-.18
-.5
-.35
-.11
-.19
-.12
-.18
-.29
-.36
-.49
-.81
-1
-.74
-.26
-.26
-.22
-.27
30% of US-Greece
management gap
due to better US
-.83
reallocation
-.98
-.49
-1.02
-1.18 -1.19
-1.2
-1.5
Management gap with the US
0
0
-2
-1.65
us
sw
jp
ge
ca
gb
po
Reallocation
mean of rel_OP
it
fr
br
cn
ar
pt
gr
Within
of rel_zman
mean Firm
Notes: Total weighted mean management deficit with the US is the number on top of bar. This is decomposed into (i) reallocation
effect (blue bar) and (ii) unweighted average management scores (red bar) . Domestic firms, scores corrected for sampling bias
Second, estimate impact of management on TFP
using result from micro regressions and field
experiments result: ↑1 SD management ≈ ↑ 10% TFP
Country
US
Sweden
Japan
Canada
Great Britain
Italy
France
Brazil
China
Argentina
Portugal
Greece
Unweighted av.
Share-Weighted
Average Management
Deficit with US
0
-0.25
-0.35
-0.50
-0.74
-0.81
-0.82
-0.98
-1.01
-1.17
-1.18
-1.65
TFP GAP
with US
Proportion of TFP
gap due to
Management
32.2
33.6
22.3
20.3
17.2
25.3
59.6
78.3
57.3
24.9
51.0
7.8%
10.4%
22.4%
36.5%
47.7%
38.7%
16.9%
14.9%
20.6%
48.2%
32.4%
25%
Assume one sd increase in management increases TFP by 10%. Regressions suggest
about 5% to 15% depending on specification. TFP data from Jones and Romer (2010).
Interestingly, also get similar 25% share for cross firm
contribution of management to TFP spread
•
Typical 4 digit industry in US has TFP ratio of 2:1 for 90-10
(Syversson, 2004)
•
90-10 for management is about 2.6 standard deviations
•
Using a causal effect of 10% this implies management
causes ≈ ¼ of TFP dispersion
CONCLUSIONS
• Large spreads in size weighted management across firms
and countries, with about 1/3 of US lead due to reallocation
• Micro data consistent with a model in which this variation in
management is technology, with better & worse practices
• Estimate variations in management accounts for ≈25% of
plant and country variation in TFP
MY FAVOURITE QUOTES:
The difficulties of defining ownership in Europe
Production Manager: “We’re owned by the Mafia”
Interviewer: “I think that’s the “Other” category……..although I
guess I could put you down as an “Italian multinational” ?”
Americans on geography
Interviewer: “How many production sites do you have abroad?
Manager in Indiana, US: “Well…we have one in Texas…”
MY FAVOURITE QUOTES:
Don’t get sick in Britian
Interviewer : “Do staff sometimes end up doing the wrong sort
of work for their skills?
NHS Manager: “You mean like doctors doing nurses jobs, and
nurses doing porter jobs? Yeah, all the time. Last week, we had
to get the healthier patients to push around the beds for the
sicker patients”
Don’t do Business in Indian hospitals
Interviewer: “Is this hospital for profit or not for profit”
Hospital Manager: “Oh no, this hospital is only for loss making”
MY FAVOURITE QUOTES:
Don’t get sick in India
Interviewer : “Do you offer acute care?”
Switchboard: “Yes ma’am we do”
Interviewer : “Do you have an orthopeadic department?”
Switchboard: “Yes ma’am we do”
Interviewer : “What about a cardiology department?”
Switchboard: “Yes ma’am”
Interviewer : “Great – can you connect me to the ortho department”
Switchboard?: “Sorry ma’am – I’m a patient here”
“OLLEY PAKES” (OP) DECOMPOSITION OF WEIGHTED
AVERAGE MANAGEMENT SCORE (M) IN GIVEN COUNTRY
Overall management z-score of firm i
Size of firm i
M   M iYi
i
  [( M i  M i )(Yi  Yi )]  M
i
 OP  M
Covariance
(Olley-Pakes, 1996,
reallocation term)
Unweighted mean
of management score
DECOMPOSING THE RELATIVE MANAGERIAL DEFICIT
BETWEEN COUNTRY j AND THE US ECONOMY
M M
k
US
Difference in aggregate
share-weighted
Management scores
 (OP  OP )  ( M  M )
k
US
Difference in reallocation
(between firm)
k
US
Difference in unweighted
Means (within firm)
TAB 3: DECOMPOSING AGGREGATE MANAGEMENT GAP
INTO REALLOCATION & UNWEIGHTED MEAN DIFFERENCE
Country
US
Sweden
Japan
Germany
Canada
Great Britain
Poland
Italy
France
Brazil
China
Argentina
Portugal
Greece
ShareWeighted
Average
Managemen
t Score, M
Reallocation
effect
(Olley-Pakes,
OP)
Unweighted
Average
Management
Score
(1)=(2)+(3)
0.67
0.42
0.32
0.31
0.17
-0.07
-0.14
-0.15
-0.31
-0.34
-0.50
-0.51
-0.53
-0.98
(2)
0.36
0.22
0.18
0.28
0.25
0.17
0.18
0.07
0.10
0.24
0.10
0.14
0.09
-0.13
(3)
0.31
0.20
0.14
0.03
-0.07
-0.24
-0.32
-0.23
-0.41
-0.59
-0.61
-0.66
-0.62
-0.85
% of deficit
“Deficit” in “Deficit” in
in
ShareReallocatio manageme
weighted
n relative
nt score
Manageme
to US
due to
nt Score
worse
relative to
reallocation
US
(2)-0.36
(6)=(5)/(4)
(1)-0.67
0
0
-0.25
-0.14
56%
-0.35
-0.18
51%
-0.36
-0.08
22%
-0.50
-0.11
22%
-0.74
-0.19
26%
-0.81
-0.18
22%
-0.82
-0.29
35%
-0.98
-0.26
27%
-1.01
-0.12
12%
-1.17
-0.26
22%
-1.18
-0.22
19%
-1.20
-0.27
22%
-1.65
-0.49
30%
INFORMATION: ARE FIRMS AWARE OF THEIR
MANAGEMENT PRACTICES BEING GOOD/BAD?
We asked:
“Excluding yourself, how well managed would you say your
firm is on a scale of 1 to 10, where 1 is worst practice, 5 is
average and 10 is best practice”
We also asked them to give themselves scores on operations
and people management separately
-6
labp
Labor Productivity
-4
-2
0
2
SELF-SCORES UNCORRELATED WITH
PRODUCTIVITY
Lowess smoother
0
2
4
6
8
Their self-score:
(worst management
practice), 5 (average) to 10 (best practice)
Self 1scored
10
bandwidth = .8
* Insignificant 0.03 correlation with labor productivity, cf. management score has a 0.295
TABLE 11: COMPETITION AFFECTS FIRM’S SELFPERCEPTIONS OF MANAGEMENT QUALITY
Dep. Var.
Management
Self-score Management
FEs
SIC3
SIC3
Firm
SIC3
SIC3
Firm
Sample
Competition
%college
All
0.064***
(0.018)
0.115***
(0.008)
0.175***
(0.009)
2+ obs
0.082***
(0.031)
0.109***
(0.014)
0.157***
(0.017)
2+ obs
0.119**
(0.051)
All
-0.038*
(0.023)
0.040***
(0.011)
0.069***
(0.012)
2+ obs
-0.041
(0.039)
0.069***
(0.020)
0.060***
(0.022)
2+ obs
-0.046
(0.073)
8,776
3,276
3,349
7,960
2,934
3,007
Ln(emp)
Obs
Notes: Controls include country & year dummies, public & interview noise
(interviewer, time, date & manager characteristic). SEs clustered by firm.
ALGORITHM
• Discretise state space for M,K and A.
• Value functions for entrants, incumbents. Numerical
contraction mapping to obtain fixed point
• Investment policy correspondences (for M & K). L
defined optimally using first-order conditions
• Simulate data and focus on steady states after 50 years
for 10,000 firms
• Change parameters & examine different steady states
– Increase in competition
– Increase in distortions
• Compare actual and simulated data
PARAMETERS
Parameter
Value
Rationale
ρ
TFP AR(1)
0.95
Cooper and Haltiwanger (2006)
α
Capital
1/3
Capital share
β
Labor
1/2
Labor share
γ
Management (MAT) 1/6
Assumed management share (intangibles)
b
Management (MAD) 2
Assumed costs of non-optimal management
E
Sunk cost
100
Assumed 100 times unit price
F
Fixed cost
25
Assumed 25 times unit price
e
Price elasticity
4
Markup of 25%, midpoint of literature
QK Capital quad. Acs
100
ACs about 5% of revenue, Bloom (2009)
QM Manag. quad. Acs
100
ACs about 5% of revenue, Bloom (2009)
Dτ
10%
Hsieh and Klenow (2009)
Distortions upper
bound
TRANSITION MATRIX: MANAGEMENT, 2006-2009, 1600 firms
Quintile in 2009
Quintile in 2006
Bottom
Bottom
Second
Third
Fourth
Top
52
22
15
9
3
Second
23
25
25
8
10
Third
16
24
26
19
15
Fourth
7
16
26
26
24
Top
6
8
13
28
46
TRANSITION MATRIX: TFP 1972-1977, BAILEY ET AL (1992)
Quintile in 1977
Bottom
Second
Third
Fourth
Top
Bottom
36
18
11
18
16
Second
20
19
22
22
17
Third
16
24
22
24
13
Fourth
9
7
17
35
34
Top
5
6
7
16
65
Quintile in 1972
CONCERNS WITH THE CROSS-COUNTRY
DECOMPOSITION
•Re-weighting for sample responses. Use alternative
covariate sets (e.g. just emp in selection in Table B1)
•Using just labor as size measure (because cleanest). Use
capital as well to get “input index” (Table B2)
•Dropped multinationals because “size” unclear: what if we
include (Table B3)
•Only looking at firms between 50 to 5000 what about very
small and very large firms? (Table B4)
56
Figure B1: Relative Management (weighted by
employment shares; only emp in selection equation)
0
0
0
-.05
-.09
-.12
-.14
-.2
-.18
-.14
-.22
-.26
-.19
-.2
-.33
-.5
-.39
-.68
-.69
-.78
-1
-.92
-1.14
-1.5
-1.11
-1.54
us
sw
jp
ge
gb
mean of rel_man
it
po
fr
cn
pt
gr
mean of rel_OP
Notes: These are the differences relative to the US of (i) the weighted average management scores (sd=1, blue bar) and (ii) reallocation effect
(OP, light red bar). Domestic firms, . 2006 wave. Response bias corrections use country-specific employment only
Figure B2: Relative Management (weighted by labor and
capital inputs)
0
0 0
-.05
-.06
-.17
-.14
-.1
-.16
-.16
-.2
-.24
-.18
-.24
-.33
-.5
-.36
-.71
-.72
-.73
-1
-.88
-1.16
-1.5
-1.11
-1.52
us
jp
sw
ge
gb
mean of rel_man
po
it
fr
pt
cn
gr
mean of rel_OP
Notes: These are the differences relative to the US of (i) the weighted average management scores (sd=1, blue bar) and (ii) reallocation effect
(OP, light red bar). Domestic firms, . 2006 wave. Response bias corrections use country-specific employment only
Figure B3: Relative Management (weighted by
employment shares; multinationals included)
0
0 0
-.02
-.13
-.06
-.08
-.05
-.12
-.16
-.26
-.21
-.32
-.5
-.27
-.08
-.15
-.51
-.53
-.68
-.82
-1.27
-1.29
gr
cn
-1.5
-1
-.89
us
jp
ge
sw
fr
mean of rel_man
it
gb
po
pt
mean of rel_OP
Notes: These are the differences relative to the US of (i) the weighted average management scores (sd=1, blue bar) and (ii) reallocation effect
(OP, light red bar). Domestic firms, . 2006 wave. Response bias corrections use country-specific employment only
-2
-1.5
-1
-.5
0
Figure B4: Relative Management scores. Correcting for
missing very small and very large firms.
us
ge
sw
jp
gb
mean of rel_man
fr
po
it
pt
gr
mean of rel_WM
Notes: These are the differences relative to the US of the weighted average management scores. Response bias corrections. Blue bar is
baseline results and red bar corrects for missing firms with under 50 or over 5000 employees. The correlation between the two bars is 0.95.
MANAGEMENT AS DESIGN: STYLES DO DIFFER
SYSTEMATICALLY ACROSS FIRMS AND INDUSTRIES
Dependent
Variable
% Degree
(Firm)
% Degree
(Industry)
Ind Dums
Obs
Human
Capital
Managem
ent (zscore)
HC
1.193***
(0.072)
No
7,641
Fixed
Capital
Manage
ment (zscore)
FC
0.610***
(0.065)
No
7,641
Relative
style of
Managem
ent
Relative
style of
Managem
ent
HC-FC
0.583***
(0.085)
HC-FC
0.561***
(0.088)
No
7,641
Yes
7,641
Human
Capital
Managem
ent (zscore)
HC
Fixed
Capital
Manage
ment (zscore)
FC
Relative
style of
Manage
ment
0.743***
(0.229)
0.403*
(0.222)
0.341**
(0.167)
No
8,833
No
8,833
No
8,833
HC-FC
Note: OLS with standard errors clustered by 156 three digit industries below coefficients. “Human Capital Management” (HC) is the
average of the z-scores of questions 13,17 and 18 (and this overall average z-scored). “Fixed Capital Capital Management” (FC) is
the average of the z-scores of questions 1, 2 and 4 (and this overall average z-scored). “Relative style of management” is the simple
difference of HC and FC. All columns include year dummies, ln(firm size), ln(firm age) and a full set of country dummies. “% degree
(firm)” is the proportion of the firm’s employees with a college degree. “% degree (industry-level in US)” is the proportion of workers in
a three digit SIC with a college degree from the US Current Population Survey.
Management in the US vs developing countries
1
India
Nicaragua
Ethiopia
Ghana
Zambia
Kenya
Tanzania
.5
United States
Bottom 25% in US: 2.88
0
US
1
2
3
4
Firm Average Management Score
5
Note: Percentage of firms scoring within the range of the bottom quartile of US firms: 64% of IN
firms, 80% of NIC firms, 94% of ET firms, 95% of GH firms, 74% of KE firms, 90% of TZ firms,
77% of ZM firms
Data includes 2013 survey wave as of 9/20/2013. Africa data not yet included in the paper
PERFORMANCE REGRESSIONS
Performance
measure
ln Yit   M M it   L ln(nit )   K ln(kit )   X xit  uit
Management
(z-score each question,
average & z-score again)
Other controls
Labor
Capital
•
M, Management Index is average of all 18 questions (sd=1)
•
Other controls include: % employees with college degree, average hours
worked, firm age, industry, country & time dummies & noise (e.g.
interviewer dummies).
Education also important in accounting for
management differences
• Important in both MAD and MAT models:
– MAT: reduces the cost of good management
– MAD: likely to change the optimal choice of practices
• Empirically challenge is identifying causal effects, tried:
– Variation in universities across locations by country
– Using land-grant colleges in the US (Moretti, 2004)