Gender board diversity and firm performance

We study the effects of gender board diversity on firm performance. We use novel and rich firm-level data covering over seven million private and public firms spanning the years 1995-2020 in Europe. We augment a standard TFP estimation with a shift-share instrument for gender board diversity. We find that increasing the share of women in the boardroom is conducive to better economic performance. The results prove robust in a variety of subsamples, and to a variety of sensitivity analyses. This outcome is driven primarily by firms from the service sector. The positive impact was stronger during the more recent years of our sample that is a period with relatively more board diversity.

1. GBDD - Gender Board Diversity Database
Hubert Drażkowski(FAME|GRAPE)
Joanna Tyrowicz(University of Warsaw, FAME|GRAPE, University of Regensburg, and IZA)
Sebastian Zalas(FAME|GRAPE)
OECD, Paris, February 2025
1

2. New source of data
•Public firms = public scrutiny⇒reputation costs, etc.
•Systematic evidence on/from private firms scarce
What we do?
1.Go beyond stock-listed firms (novel data)
2.Explore data from 1985 onwards (novel gender identification)
2

3. GBDD: overview
Orbis: 28+ mln firms
•(Quasi-)Administrativedata: 1985 - 2020 for 44 European countries
•All kinds of firms: subject to public eye scrutiny + under veil of silence
•Board members: one-tier and two-tier systems
Sample properties
•Average # of years in the sample per firm: 6 (mean), 5 (median), 35 (max)
•Average # of people per firm: 5+
3

4. GBDD: overview
•Harmonizing the firm-level Orbis data.
•We concatenate 11 waves of the Orbis
2000, 2002, 2003, 2004, 2006, 2008, 2010, 2012, 2014, 2016, and the wave of Orbis Historical Data from 2020 (OHD)
•Orbis data is notoriously difficult to process→new processing procedures
1.Legal form:identify who should have at least a management board
2.Industry:new crosswalks
to assign firm-year observations with a NACE Rev. 2 classification (97.24%)
4

5. GBDD: private vs public
5

6. GBDD: board assignment
•Orbis itself not reliable, hence own coding
•One-tier, two-tier systems and mixed systems
•Some position names are not sufficiently informative
(e.g. “board member” or “president of the board”)
6

7. GBDD: board assignment
•Parse the words→morphemes consistent withspecificboard positions
•non-executive/supervisory
•executive/management
•ambiguous
7

8. GBDD: gender assignment
•Gender assignmentbased on linguistic rules
•In some languages: name or surname is unequivocal (parse names to first names & surnames)
•In other languages: book of names (from first names)
•In case of conflict (e.g. expats), check on a case-by-case basis
•Overall:<1%of cases was not resolved
•Manipulation checks: accurate in 99%
8

9. GBDD in all its glory
# of unique obs. # of obs.
# of observations in
management supervisoryambiguous
Firms 28,140,575147,923,404 110,209,358 1,680,70049,724,960
Listed 22,205 177,287 89,121 47,151 132,378
People 58,921,483239,151,923 148,201,928 4,474,35386,585,070
Men 43,675,552180,887,653 113,260,406 3,428,02864,288,678
Women 14,363,672 55,953,919 33,801,047 960,97021,211,696
Total attributed 58,039,224236,841,572 147,061,453 4,388,99885,500,374
Women % in total attributed 24.75 23.63 22.98 21.89 24.81
Unattributed 882,259 2,310,351 1,140,475 85,355 1,084,696
9

10. GBDD in all its glory
10

11. Potential uses of GBDD
Use GBDD both as outcome measures and as correlates / drivers
•Three indicators by country & time
•Correlates in politics & social life
•Institutional determinants
•Enrollment in education / STEM
•Wage gaps and employment gaps
•Measuresalsoby industry (so within country & time)
•Participation in GVCs, openness to trade
•Relative innovativeness of business, etc.
•We can also produce other measures, if you are interested
11

12. What do we do with the individual data
Several research agendas
•Board diversity spillovers
•w/ Drazkowski & Timmermans: horizontal spillovers only work in public eye
•w/ Contreras & Drazkowski: ownership network spillovers actually work (but are small)
•Effects of gender diversity (causal)
•w/ Zalas: TFP
•w/ Timmermans & van der Velde: gender wage gaps
•Measuring/understanding diversity
•w/ Contreras & Drazkowski: implicit quota
•w/ Drazkowski: tokenism
12

13. Revisiting gender board diversity and firm efficiency
13

14. Motivation: generating costs or benefits?
•Majority of firms have no women in leadership
•GBD is increasing (ILO, 2015) (mandated quotas & voluntary practices)
•Bertrand (2018):an economy that is tapping into a limited pool (men) to find its leaders must be
operating inside the efficiency frontier
•The theory provides justification for both+and−, and empirical evidence mixed
1.much of the literature relies on correlations→causality (?)
Post & Byron 2015, Terjesen et al 2015
2.many causal studies use the case of Norway→external validity
Bertrand et al 2003, Dezso-Ross 2012, Dale et al 2013, Bertrand et al 2019
3.women do no harm in listed companies
Sieweke et al 2023
Goal
estimateimpactof the diversity of company boards on their performance
14

15. Our model relative to the literature
•IV strategy similar to Flabbi et al (2020) & Sieweke et al. (2023)
•Other IV strategies:
→the proportion of men who are on different boards with women (Adams and Ferreira, 2009); percentage of
women among lower-level managers (Low et al., 2015; education of the CEO’s spouse (Smith et al., 2006);
cultural revolution in China and its impact on participation of women managers and employment of women
(Liu et al., 2014)
•We estimate:log(yi,t) =β0+γGBDi,t+βklog(ki,t) +βllog(li,t) +ηi+ηt+εi,t
•To address the endogeneityGBDi,twe instrument:
zi,t=sharei,t0·gk,t=
#womeni,t0
boardsizei,t0
·
(
#womenk∈c,t
boardsizek∈c,t
/
#womenk∈c,t0
boardsizek∈c,t0
)
,
wheret0 is the first year in which the companyifrom industryk(2-digit NACE) in countrycis
observed in our sample
•The common shocks need to be exogenous to firm-level decisions:gk,texogenous wrt. time varying
heterogeneity
FS binscatter
15

16. Estimation sample
•IV sample: all available 1̃00 mln of observations
•Estimation sample:
•30 mio observations: 7 mio unique firms, 37 countries, years 1995-2020
•OPRE - output measure, VA problematic (availability of VA predicts lower output,ceteris paribus)
Value added
•employment and assets are input measures
•Weights to match the representation of country in our sample to the relative size of that economy in
Europe
Coverage by country
•share of population in our group of countries
Descriptive statistics
16

17. Baseline results
Dependent variable: full sample manufacturing services
log(y
i,t
) OLS IV IV OLS IV IV OLS IV IV
(1) (2) (3) (4) (5) (6) (7) (8) (9)
GBD
i,t
-0.008
∗∗
0.256
∗∗∗
0.370
∗∗∗
-0.013
∗∗
0.124
∗
0.324
∗∗∗
-0.005 0.201
∗∗∗
0.364
∗∗∗
(0.003) (0.040) (0.047) (0.006) (0.071) (0.123) (0.004) (0.045) (0.050)
log(a
i,t
) 0.477
∗∗∗
0.477
∗∗∗
0.477
∗∗∗
0.478
∗∗∗
0.479
∗∗∗
0.479
∗∗∗
0.473
∗∗∗
0.473
∗∗∗
0.474
∗∗∗
(0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
log(l
i,t
) 0.515
∗∗∗
0.515
∗∗∗
0.515
∗∗∗
0.540
∗∗∗
0.540
∗∗∗
0.541
∗∗∗
0.496
∗∗∗
0.496
∗∗∗
0.496
∗∗∗
(0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)
Instruments F test 1156.316 2840.238 224.918 229.790 1071.880 1339.431
[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]
Endogeneity test 38.438 66.315 1.486 7.759 23.870 55.458
[0.000] [0.000] [0.223] [0.005] [0.000] [0.000]
First Stage F stat. 719.64 989.78 148.17 97.73 658.86 468.01
No. of companies 5,535,498 5,535,498 5,535,498 1,731,123 1,731,123 1,731,123 3,895,114 3,895,114 3,895,114
No. of observations32,707,48732,707,48732,707,48710,412,71010,412,71010,412,71022,237,10822,237,10822,237,108
17

18. Reduced form
Dependent variable: full sample manufacturing services
GBD
i,t
(1) (2) (3) (4) (5) (6)
IV=z
i,t
-0.123
∗∗∗
-0.060
∗∗∗
-0.251
∗∗∗
(0.017) (0.021) (0.045)
IV=z
2
i,t
0.028
∗∗∗
0.016
∗∗
0.105
∗∗∗
(0.007) (0.007) (0.031)
IV=z
3
i,t
-0.001
∗∗∗
-0.001
∗∗
-0.015
∗∗∗
(0.000) (0.000) (0.005)
IV=ˆz
i,t
-0.128
∗∗∗
-0.062
∗∗∗
-0.146
∗∗∗
(0.016) (0.023) (0.020)
log(a
i,t
) 0.477
∗∗∗
0.477
∗∗∗
0.478
∗∗∗
0.478
∗∗∗
0.473
∗∗∗
0.473
∗∗∗
(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)
log(l
i,t
) 0.515
∗∗∗
0.515
∗∗∗
0.540
∗∗∗
0.540
∗∗∗
0.496
∗∗∗
0.496
∗∗∗
(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)
Instruments F test 21.956 63.430 2.777 7.140 25.264 51.536
[0.000] [0.000] [0.040] [0.008] [0.000] [0.000]
No. of companies 5,535,498 5,535,498 1,731,123 1,731,123 3,895,114 3,895,114
No. of observations32,707,48732,707,48710,412,71010,412,71022,237,10822,237,108
18

19. Time trends
19

20. These results are robust
•weights
No weights
•alternative production functions: country-specific slopes & industry-specific slopes, translog
•alternative instrument definitions
•redefinition of the instrument to the 3-digit NACE
3-digit NACE
•leave-one-out modification ofg
k,t
L1O
•numerous sample restrictions:
Restrictions
•excluding firms which never appoint women or always appoint only women
•excluding companies with boardsize<3
•focus on management roles
•subsamples of countries
•timing: lagging GBD by 1 or 2 periods
Lags
•imputations
Imputed values
•nonlinear first stage
Non-linear first stage
20

21. Thank you for your attention!
Questions?
w: grape.org.pl
t: grape_org
f: grape.org
e: [email protected]

22. Descriptive statistics
full samplemanufacturing services
GBD 0.254 0.182 0.283
(0.364) (0.309) (0.380)
log(Y) 5.536 5.967 5.363
(2.177) (2.147) (2.165)
log(K) 5.226 5.620 5.067
(2.242) (2.253) (2.218)
log(L) 1.468 1.854 1.312
(1.364) (1.490) (1.277)
Board size 1.831 1.892 1.807
(1.578) (1.520) (1.601)
% manufacturing 28.68 100.00 0.00
% services 71.32 0.00 100.00
% 49+ 8.25 13.25 6.24
% 49- 91.75 86.75 93.76
# of countries 37 37 37
# of firms 7,053,476 2,084,854 4,968,622
# of observations34,236,398 10,845,793 23,390,605
Years 1996 - 2020 1996 - 2020 1996 - 2020
Back
21

23. Coverage by country
Country # of observations # of unique companies Country # of observations # of unique companies
Austria 218,714 54,851 Belgium 375,646 69,677
Bosnia 11,360 2,641 Bulgaria 1,222,843 228,482
Croatia 73,295 16,232 Cyprus 7,798 2,384
Czech Republic 895,898 157,769 Denmark 123,546 31,525
Estonia 571,358 91,503 Finland 912,365 148,258
France 4,309,024 924,482 Germany 2,235,611 573,165
Greece 114,905 27,238 Hungary 529,556 190,900
Iceland 33,615 11,032 Ireland 70,414 18,331
Italy 1,997,453 461,214 Latvia 585,972 114,247
Lithuania 34,929 63,55 Luxembourg 7,066 2,028
Moldova 7,212 4,611 Norway 498,058 121,648
Poland 305,193 86,556 Portugal 1,765,599 279,143
Romania 11,341 5,067 Russia 6,339,823 1,744,953
Serbia 128,368 40,950 Slovakia 498,067 110,018
Slovenia 12,987 2,283 Spain 6,821,490 1,007,522
Sweden 2,957,499 382,722 Switzerland 4,990 900
United Kingdom 521,976 113,187 Ukraine 80,370 23,851
Total 34,241,670 7,053,985
Back

24. First Stage is nonlinear
Back

25. First Stage results
Dependent variable: full sample manufacturing services
GBD
i,t
OLS IV OLS IV OLS IV
(1) (2) (3) (4) (5) (6)
IV=z
i,t
-0.521 -0.602 -0.748
[0.015] [0.022] [0.018]
IV=z
2
i,t
0.040 0.037 0.125
[0.006] [0.004] [0.010]
IV=z
3
i,t
-0.001 -0.001 -0.005
[0.000] [0.000] [0.001]
Instruments F test 837.558 309.092 1730.876
[0.000] [0.000] [0.000]
Endogeneity test 8.589 1.448 10.404
[0.003] [0.229] [0.001]
First Stage F stat. 518.88 192.27 1044.26
No. of companies 5,535,883 5,535,883 1,731,334 1,731,334 3,895,316 3,895,316
No. of observations32,710,61832,710,61810,414,29510,414,29522,238,63522,238,635
NotesEstimates of equations obtained for the full sample and the subsamples. The specification includes firm fixed effects, hence roughly 1.5 million singleton observations was dropped.
We estimate a two-stage linear model. Corrected p-values in brackets for the first-stage estimates. Standard errors are clustered at the firm level.
Back

26. Robustness checks: imputations
Dependent variable: full sample manufacturing services
log(y
i,t
) OLS IV OLS IV OLS IV
(1) (2) (3) (4) (5) (6)
GBD
i,t
-0.006
∗
0.287
∗∗∗
-0.011
∗∗
0.119 -0.003 0.261
∗∗∗
(0.003) (0.042) (0.006) (0.075) (0.004) (0.046)
log(a
i,t
) 0.464
∗∗∗
0.464
∗∗∗
0.461
∗∗∗
0.461
∗∗∗
0.462
∗∗∗
0.462
∗∗∗
(0.001) (0.001) (0.002) (0.002) (0.001) (0.001)
log(l
i,t
) 0.529
∗∗∗
0.529
∗∗∗
0.557
∗∗∗
0.557
∗∗∗
0.509
∗∗∗
0.509
∗∗∗
(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)
Instruments F test 1087.534 178.225 1059.503
[0.000] [0.000] [0.000]
Endogeneity test 44.546 0.947 35.454
[0.000] [0.330] [0.000]
First Stage F stat. 675.58 118.98 649.54
Summary statistics
log(y
i,t
) 5.25 5.74 5.07
(2.27) (2.25) (2.25)
GBD
i,t
0.25 0.18 0.28
(0.36) (0.31) (0.38)
No. of companies 5,639,143 5,639,143 1,766,577 1,766,577 3,974,824 3,974,824
No. of observations34,291,94734,291,94710,975,90910,975,90923,255,30323,255,303

27. Robustness checks: restrictions
Dependent variable: full sample 3 digit NACE GBD∈(0,1) board size>3 management roles
log(y
i,t
) OLS IV OLS IV OLS IV OLS IV OLS IV
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Second stage
GBD -0.008
∗∗
0.097
∗∗∗
-0.009
∗∗
0.075
∗
-0.013
∗∗∗
0.143
∗∗∗
0.014 0.598
∗∗∗
0.019
∗∗∗
0.028
(0.003) (0.033) (0.003) (0.040) (0.003) (0.023) (0.009) (0.096) (0.003) (0.021)
log(a
i,t
) 0.477
∗∗∗
0.477
∗∗∗
0.476
∗∗∗
0.476
∗∗∗
0.501
∗∗∗
0.502
∗∗∗
0.503
∗∗∗
0.503
∗∗∗
0.499
∗∗∗
0.499
∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.004) (0.004) (0.001) (0.001)
log(l
i,t
) 0.515
∗∗∗
0.515
∗∗∗
0.514
∗∗∗
0.514
∗∗∗
0.533
∗∗∗
0.533
∗∗∗
0.493
∗∗∗
0.492
∗∗∗
0.334
∗∗∗
0.334
∗∗∗
(0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.005) (0.005) (0.001) (0.001)
Dependent variable: GBD (first stage, instrumental variables)
IV=z
i,t
-0.521 -0.367 -0.987 -0.560 -0.834
[0.015] [0.011] [0.030] [0.034] [0.019]
IV=z
2
i,t
0.040 0.035 0.071 0.059 0.085
[0.006] [0.003] [0.010] [0.014] [0.006]
IV=z
3
i,t
-0.001 -0.001 -0.001 -0.001 -0.001
[0.000] [0.000] [0.000] [0.000] [0.000]
Instruments F test 837.558 625.505 634.401 166.963 3912.569
[0.000] [0.000] [0.000] [0.000] [0.000]
Endogeneity test 8.589 5.126 53.341 39.700 3.102
[0.003] [0.024] [0.000] [0.000] [0.078]
First Stage F stat. 518.88 394.76 402.30 100.73 2350.39
No. of companies 5,535,883 5,535,883 5,524,221 5,524,221 1,704,444 1,704,444 500,020 500,020 4,052,611 4,052,611
No. of observations32,710,618 32,710,618 32,528,534 32,528,534 11,744,444 11,744,444 3,551,279 3,551,279 21,553,203 21,553,203

28. Using VA as outcome variable
Dependent variable:Value added indicator
Percentiles of revenue 0.0171
∗∗∗
0.0264
∗∗∗
(0.0000) (0.0001)
Percentiles of revenue by year & country 0.0142
∗∗∗
0.0236
∗∗∗
(0.0000) (0.0001)
Percentiles of revenue
2
-0.0005
∗∗∗
(0.0000)
Percentiles of revenue by year & country
2
-0.0004
∗∗∗
(0.0000)
Observations 32771255 32771255 32771255 32771255
Back