Gender-Segmented Labor Markets and the Effects of Local Demand - - PowerPoint PPT Presentation

Gender-Segmented Labor Markets and the Effects of Local Demand Shocks

Juan Pablo Chauvin Research Department, Inter-American Development Bank

6th Urbanization and Poverty Reduction Research Conference

Washington DC, Sept. 2019

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 1 / 17

People, markets, cities... and policies

Cities are engines of growth in developing-world countries But the economic opportunities are not equally accessible to everyone Policies can open the access to oportunities for some groups and reduce it for others Distributional effects are not always the intended

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 2 / 17

Place-making policies

Large investments in “place-making” policies around the world

Literature on who benefits from these policies (Bartik 1991, and many that followed)

Existing studies largely assume away gender

But gender differences in the labor market are large, likely to matter

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 3 / 17

This paper

How do local labor and housing markets react to:

Shocks to labor demand that favor male employment, vs. Shocks to labor demand that favor female employment

Context:

Brazil 1990s y 2000s Unit of observation: “microregions” Primarily decennial census data

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 4 / 17

Intuition of the model

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 5 / 17

Intuition of the model

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 5 / 17

Intuition of the model

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 5 / 17

Intuition of the model

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 5 / 17

Intuition of the model

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 5 / 17

Male-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 6 / 17

Male-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 6 / 17

Male-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 6 / 17

Female-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 7 / 17

Female-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 7 / 17

Female-leaning shocks to labor demand

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 7 / 17

Reduced-form relationship of interest

∆t−toOutcomej = α + β ∆t−toLabor DemandjG + δ Controlsj,t0 + ∆t−toǫjt for genders G = {M,W} Outcomes of interest

Population (migration) Housing rents Employment and wages by gender

Need exogenous variation for changes in labor demand for males and females

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 8 / 17

Gender-specific Bartik shocks BG

j,t−to =

• ind

sind,j,t0

• Local industry

share at t0

gG

−j,ind,t−t0

• National growth in gender G’s

employment in industry

for genders G = {M, W } Predicts gender-specific employment growth in region j over a given period, based on:

The industries located in the region in the base year, and, How gender G employment grew in the rest of the country in those industries.

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 9 / 17

Identification

Recent research on Bartik-style (shift-share) shocks (Goldsmith Pinkham et al. 2019, Adão et al. 2018, Borusyak et al. 2018.) In this light, I assess the validity of identifying assumptions and find similar issues as Goldsmith-Pinkham et al. (2019) do in the U.S. context:

The shares are correlated with baseline characteristics potentially endogenous (i.e. initial income and education levels) Pre-trends are present

I introduce start-year level controls and lagged-growth controls

After controls, shocks are uncorrelated with a large set of observable confounders

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 10 / 17

Addressing identification concerns

Figure: Gender-specific Bartik shocks and Female informality rate

Simple Residualized Male shocks Female shocks

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 11 / 17

Distributions of residualized shocks 1991-2010

Female Shocks Male Shocks

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 12 / 17

Effects on Population

Own-gender shocks Other-gender shocks Hypothesis tests Females Males Females Males (1)-(2) (3)-(4) (1) (2) (3) (4) (χ2 and p-val.) Panel A: Effects on aggregate working -age population Two decades (1991 - 2010) 0.15 0.60* 1.75 (0.21) (0.36) 0.19 The nineties (1991 - 2000) 0.23 0.76** 12.00 (0.21) (0.31) 0.00 The two thousands (2000 - 2010) 0.24 0.88*** 2.34 (0.34) (0.23) 0.13 Panel B: Effects on working-age population by gender Two decades (1991 - 2010) 0.10 0.63* 0.54 0.19 2.34 1.12 (0.22) (0.37) (0.36) (0.21) 0.13 0.29 The nineties (1991 - 2000) 0.22 0.78** 0.74** 0.23 8.35 11.60 (0.21) (0.32) (0.31) (0.20) 0.00 0.00 The two thousands (2000 - 2010) 0.21 0.90*** 0.87*** 0.26 2.80 1.90 (0.33) (0.24) (0.23) (0.36) 0.09 0.17

Note: All coefficients are estimated with SUR regression models with a common set of controls. Outcomes are measured restricting the sample to individuals aged 15 through 64, excluding individuals in school, employers, civil servants, and public

• security. Robust standard errors clustered at the mesoregion level in parentheses, except for the hypothesis tests. The

hypothesis tests are Wald chi-square tests of the hypothesis H0 : βmales − βfemales = 0 on SUR models including the respective female and male regressions. JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 13 / 17

Effects on Population

Own-gender shocks Other-gender shocks Hypothesis tests Females Males Females Males (1)-(2) (3)-(4) (1) (2) (3) (4) (χ2 and p-val.) Panel A: Effects on aggregate working -age population Two decades (1991 - 2010) 0.15 0.60* 1.75 (0.21) (0.36) 0.19 The nineties (1991 - 2000) 0.23 0.76** 12.00 (0.21) (0.31) 0.00 The two thousands (2000 - 2010) 0.24 0.88*** 2.34 (0.34) (0.23) 0.13 Panel B: Effects on working-age population by gender Two decades (1991 - 2010) 0.10 0.63* 0.54 0.19 2.34 1.12 (0.22) (0.37) (0.36) (0.21) 0.13 0.29 The nineties (1991 - 2000) 0.22 0.78** 0.74** 0.23 8.35 11.60 (0.21) (0.32) (0.31) (0.20) 0.00 0.00 The two thousands (2000 - 2010) 0.21 0.90*** 0.87*** 0.26 2.80 1.90 (0.33) (0.24) (0.23) (0.36) 0.09 0.17

Note: All coefficients are estimated with SUR regression models with a common set of controls. Outcomes are measured restricting the sample to individuals aged 15 through 64, excluding individuals in school, employers, civil servants, and public

• security. Robust standard errors clustered at the mesoregion level in parentheses, except for the hypothesis tests. The

hypothesis tests are Wald chi-square tests of the hypothesis H0 : βmales − βfemales = 0 on SUR models including the respective female and male regressions. JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 13 / 17

Effects on housing rents 1991-2010

Female Male Hypothesis tests Shock Shock (χ2 and p-val.) (1) (2) (3) Effects on nominal housing rents 0.23 1.09*** 6.08 (0.27) (0.32) 0.01 Effects on residualized housing rents 0.15 0.71** 2.82 (0.25) (0.32) 0.09

Note: All coefficients are estimated with SUR regression models with a common set of controls. Outcomes are measured restricting the sample to individuals aged 15 through 64, excluding individuals in school, employers, civil servants, and public security. Robust standard errors clustered at the mesoregion level in parentheses, except for the hypothesis tests. The hypothesis tests are Wald chi-square tests of the hypothesis H0 : βmales − βfemales = 0 on SUR models including the respective female and male regressions. JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 14 / 17

Effects on employment and wages 1991-2010

Own-gender shocks Other-gender shocks Hypothesis tests Females Males Females Males (1)-(2) (3)-(4) (1) (2) (3) (4) Panel A: Employment effects

(χ2 and p-val.)

Effects on employment 0.41* 0.99*** 1.09** 0.30 2.78 3.68 (0.23) (0.37) (0.43) (0.25) 0.10 0.06 Panel B: Residualized wage effects Montly wage

• 0.02

0.57*** 0.43* 0.04 6.88 2.95 (0.12) (0.22) (0.23) (0.10) 0.01 0.09 Hourly wage

• 0.12

0.45** 0.32

• 0.10

7.67 4.10 (0.10) (0.21) (0.21) (0.10) 0.01 0.04

Note: All coefficients are estimated with SUR regression models with a common set of controls. Outcomes are measured restricting the sample to individuals aged 15 through 64, excluding individuals in school, employers, civil servants, and public

• security. Robust standard errors clustered at the mesoregion level in parentheses, except for the hypothesis tests. The

hypothesis tests are Wald chi-square tests of the hypothesis H0 : βmales − βfemales = 0 on SUR models including the respective female and male regressions. JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 15 / 17

Effects on employment and wages 1991-2010

Own-gender shocks Other-gender shocks Hypothesis tests Females Males Females Males (1)-(2) (3)-(4) (1) (2) (3) (4) Panel A: Employment effects

(χ2 and p-val.)

Effects on employment 0.41* 0.99*** 1.09** 0.30 2.78 3.68 (0.23) (0.37) (0.43) (0.25) 0.10 0.06 Panel B: Residualized wage effects Montly wage

• 0.02

0.57*** 0.43* 0.04 6.88 2.95 (0.12) (0.22) (0.23) (0.10) 0.01 0.09 Hourly wage

• 0.12

0.45** 0.32

• 0.10

7.67 4.10 (0.10) (0.21) (0.21) (0.10) 0.01 0.04

Note: All coefficients are estimated with SUR regression models with a common set of controls. Outcomes are measured restricting the sample to individuals aged 15 through 64, excluding individuals in school, employers, civil servants, and public

• security. Robust standard errors clustered at the mesoregion level in parentheses, except for the hypothesis tests. The

hypothesis tests are Wald chi-square tests of the hypothesis H0 : βmales − βfemales = 0 on SUR models including the respective female and male regressions. JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 15 / 17

Conclusions

The effects of local labor demand shocks can be very different if they are male-leaning than if they are female-leaning In Brazil: Male shocks lead to larger immigration ...and made localities more expensive Male shocks increased local gender employment and wage gaps Female shocks decreased the employment gap but not necessarily the wage gap

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 16 / 17

Implications for policy

The gender composition of local demand shocks can affect their impact on welfare:

Male shocks are more likely to benefit immigrants and landlords Female shocks are more likely to favor current residents

Highlights the importance of incorporating the gender structure of labor demand in the design and evaluation of local development policies

JP Chauvin (IADB-RES) Gender and Local Demand Shocks Washington DC, Sept. 2019 17 / 17