The Long-Run Effects of Recessions on Education and Income Bryan - - PowerPoint PPT Presentation

The Long-Run Effects of Recessions

• n Education and Income

Bryan Stuart

George Washington University

June 2018

Any opinions and conclusions expressed herein are those of the author and do not necessarily represent the views of the U.S. Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed.

Introduction

Recessions receive tremendous attention from economists, policymakers, and the public Most attention focuses on short-run effects on

• Workers’ labor market outcomes
• Firms’ investment and employment decisions

Recessions could have long-run effects, especially via human capital

• Past work: Lower opportunity cost increases education
• Other mechanisms could dominate for younger individuals
• Both sign and size of long-run effects are open questions

1

This paper

Setting: double-dip recession in U.S. from 1980-82

• Led to persistent relative decline in earnings per capita in some counties

Data: adult outcomes from 2000-13 and birth county for people born 1950-79 Empirical strategy: generalized difference-in-differences

• 1. People born in counties with more vs. less severe recession
• 2. People who were younger vs. older when recession began

2

Results: Sizable long-run reductions in education and income

People who were born in counties with more severe recession and were children or adolescents during the recession

• Are less likely to obtain a college degree
• Earn less income
• Have higher poverty rates

For people age 0-10 in 1979, a 10% decrease in real earnings per capita from 1978-82 leads to

• 3.0 p.p. (9.4%) decrease in four-year college degree attainment
• $1,300 (3.2%) decrease in earned income
• 1.7 p.p. (13.9%) increase in poverty rate

Little evidence that states with more generous or more progressive transfer systems mitigated these long-run effects

3

The 1980-82 recession could depress aggregate economic output today

Scale diff-in-diff estimates by 105 million people born in U.S. from 1951-79 Aggregate effects from back of envelope calculations:

• 0.9-2.1 million fewer four-year college graduates (1-3% of stock in 2015)
• $64-145 billion less earned income per year (0.4-0.8% of GDP in 2015)
• 0.5-1.3 million more adults in poverty per year (1-3% of stock in 2015)

4

Contribution

New evidence showing that 1980-82 recession led to sizable, long-lasting reductions in education and income Contributes to four distinct literatures

• Effects of labor market conditions on schooling

[Mincer 1958; Becker 1962; Black, McKinnish and Sanders 2005; Cascio and Narayan 2015; Charles, Hurst and Notowidigdo 2015; Atkin 2016]

• Long-run effects of early life conditions

[Almond and Currie 2011; Heckman and Mosso 2014; Chetty and Hendren 2016]

• Welfare costs of recessions [Lucas 2003]
• Recessions and subsequent economic activity - cleansing effects

[Schumpeter 1939, 1942; Caballero and Hammour 1994; Foster, Grim and Haltiwanger 2016]

5

Background: The 1980-82 recession

6

Measuring local economic activity and the severity of the recession

Earnings per capita from BEA (1969-forward)

• By county of residence
• Primarily from administrative unemployment insurance and tax data

Severity of recession: decrease in log real earnings per capita from 1978-82

• Earnings decline captures several possible channels, including job loss
• Some economic indicators began to decline in 1979

7

Which counties experienced a more severe recession?

Darker color indicates bigger decrease in log real earnings per capita from 1978-82

8

Which counties experienced a more severe recession?

Darker color indicates bigger decrease in log real earnings per capita from 1978-82

92 percent of variation is within-region 63 percent of variation is within-state

8

1980-82 recession led to sudden, persistent decrease in earnings per capita in some counties

Mean change in real earnings per capita from 1978-1982 Less severe recession: $44 (0.2%) More severe recession: $-2708 (-10.4%)

20000 25000 30000 35000 40000 Real earnings per capita (2014$) 1970 1974 1978 1982 1986 1990 1994 1998 2002 Year

Less severe recession More severe recession

9

Conceptual framework: Possible long-run effects of a recession on education and income

10

A recession could affect human capital development in childhood

Stock of human capital obtained in childhood depends on

• Sequence of material and time investments from parents
• Sequence of community investments
• Initial human capital endowment

Recession-induced decrease in local wage and earnings could lead to

• Less material investments from parents

[Aizer, Eli, Ferrie, and Lleras-Muney 2016; Bleakley and Ferrie 2016; Hoynes, Schanzenbach, and Almond 2016]

• More time investments from parents

[Aguiar, Hurst, and Karabarbounis 2013; Del Boca, Flinn, and Wiswall 2014; Morrill and Pabilonia 2015]

• Less community investments

[Carrell, Hoekstra, and Kuka 2016; Chetty, Hendren, and Katz 2016]

I focus on individuals born before 1980, so initial human capital is fixed

11

A recession could affect investment in high school or college education

In deciding whether to obtain a college degree, individuals trade off

• Higher lifetime earnings
• Opportunity cost of forgone earnings and cost of tuition

Recession reduces opportunity cost

[Bound and Holzer 2000; Hoynes, Miller, and Schaller 2012]

• Could increase educational attainment

[Black, McKinnish, and Sanders 2005; Cascio and Narayan 2015; Charles, Hurst, and Notowidigdo 2015; Atkin 2016]

Recession could reduce parents’ ability to finance tuition

• Could decrease college degree attainment in presence of credit constraints

[Cameron and Heckman 2001; Belley and Lochner 2007; Lovenheim 2011]

12

Data and empirical strategy

13

Data: Outcomes in adulthood and place of birth

Newly available confidential data on adult outcomes and county of birth

• 2000 Census and 2001-13 ACS linked to SSA NUMIDENT file

Sample: people born from 1950-79 who are 25-64 years old in survey

• Exclude people if relevant variables are imputed

◮ Sex, race, birth year, birth state, or dependent variables

• Exclude people living in group quarters (not in 2001-05 ACS)

14

Regression of interest 1/2

yi,a,c,t = R78−82

c

πa + xi,a,c,tβ + γc + θa,s(c) + δt + εi,a,c,t yi,a,c,t: long-run outcome of individual i, age a in 1979, born in county c,

• bserved in survey year t

R78−82

c

: 1978-82 decrease in log real earnings per capita for residents of c xi,a,c,t: individual and birth county covariates γc: birth county fixed effect θa,s(c): birth cohort by birth state fixed effect δt: survey year fixed effect

15

Regression of interest 2/2

yi,a,c,t = R78−82

c

πa + xi,a,c,tβ + γc + θa,s(c) + δt + εi,a,c,t Sample contains people who were age 0-29 in 1979 29 year olds are omitted reference group: π29 = 0

• Identified parameters are πa − π29
• Education: schooling largely completed before recession (π29 ≈ 0)
• Income: estimates could be biased upwards (π29 < 0)

Use 2SLS to exploit variation in recession severity driven by pre-recession industrial specialization [Bartik 1991]

Details

Estimates of πa are likely attenuated due to measurement error in R78−82

c

• County of birth differs from 1979 county of residence

16

Results: The long-run effects of the 1980-1982 recession

17

Long-run effects on four-year college degree attainment

Comparison Group: Age 23-28 in 1979 H0: π24 = π27 = 0 p = 0.519

• .5
• .4
• .3
• .2
• .1

.1 .2 .3 .4 Four-year college degree attainment 5 10 15 20 25 30 Age in 1979

18

Long-run effects on four-year college degree attainment

Comparison Group: Age 23-28 in 1979 Exposed to Recession: Age 0-22 in 1979 H0: π24 = π27 = 0 p = 0.519

• .5
• .4
• .3
• .2
• .1

.1 .2 .3 .4 Four-year college degree attainment 5 10 15 20 25 30 Age in 1979

18

Long-run effects on four-year college degree attainment

Comparison Group: Age 23-28 in 1979 Exposed to Recession: Age 0-22 in 1979 10% decrease in earnings per capita from 1978-1982 9.9% (3.2 p.p.) decrease in 4-year degree attainment for 6 year olds H0: π24 = π27 = 0 p = 0.519

• .5
• .4
• .3
• .2
• .1

.1 .2 .3 .4 Four-year college degree attainment 5 10 15 20 25 30 Age in 1979

18

Long-run effects on education

Dependent variable: HS/GED attainment (1) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10 0.039 (0.039) 11-19 0.038 (0.031) 20-28 0.017 (0.026) Panel B: Average value of dependent variable in years 2000-13, by age in 1979 0-10 0.936 11-19 0.932 20-28 0.933

19

Long-run effects on education

Dependent variable: Any HS/GED college attainment attendance (1) (2) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10 0.039

• 0.038

(0.039) (0.052) 11-19 0.038

• 0.099

(0.031) (0.066) 20-28 0.017

• 0.054

(0.026) (0.051) Panel B: Average value of dependent variable in years 2000-13, by age in 1979 0-10 0.936 0.588 11-19 0.932 0.537 20-28 0.933 0.540

19

Long-run effects on education

Dependent variable: Any Any college HS/GED college degree attainment attendance attainment (1) (2) (3) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10 0.039

• 0.038
• 0.184***

(0.039) (0.052) (0.068) 11-19 0.038

• 0.099
• 0.122**

(0.031) (0.066) (0.059) 20-28 0.017

• 0.054

0.026 (0.026) (0.051) (0.036) Panel B: Average value of dependent variable in years 2000-13, by age in 1979 0-10 0.936 0.588 0.414 11-19 0.932 0.537 0.380 20-28 0.933 0.540 0.381

19

Long-run effects on education

Dependent variable: Any Four-year Any college college HS/GED college degree degree attainment attendance attainment attainment (1) (2) (3) (4) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10 0.039

• 0.038
• 0.184***
• 0.303***

(0.039) (0.052) (0.068) (0.109) 11-19 0.038

• 0.099
• 0.122**
• 0.159**

(0.031) (0.066) (0.059) (0.080) 20-28 0.017

• 0.054

0.026 0.031 (0.026) (0.051) (0.036) (0.043) Panel B: Average value of dependent variable in years 2000-13, by age in 1979 0-10 0.936 0.588 0.414 0.321 11-19 0.932 0.537 0.380 0.288 20-28 0.933 0.540 0.381 0.292

19

Long-run effects on education

Dependent variable: Any Four-year Two-year Any college college college HS/GED college degree degree degree attainment attendance attainment attainment attainment (1) (2) (3) (4) (5) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10 0.039

• 0.038
• 0.184***
• 0.303***

0.119** (0.039) (0.052) (0.068) (0.109) (0.060) 11-19 0.038

• 0.099
• 0.122**
• 0.159**

0.037 (0.031) (0.066) (0.059) (0.080) (0.048) 20-28 0.017

• 0.054

0.026 0.031

• 0.004

(0.026) (0.051) (0.036) (0.043) (0.033) Panel B: Average value of dependent variable in years 2000-13, by age in 1979 0-10 0.936 0.588 0.414 0.321 0.093 11-19 0.932 0.537 0.380 0.288 0.093 20-28 0.933 0.540 0.381 0.292 0.090

19

Long-run effects on education: context

For people age 0-10 in 1979, the effect of a 10 percent decline in mean earnings per capita

• on any college degree attainment is similar in magnitude to effect of reducing

class sizes from K-3 by about 30 percent [Dynarski, Hyman, and Schanzenbach 2013]

• on four-year degree attainment is similar in magnitude to scholarship programs

that offered free tuition and fees for qualified students [Dynarski 2008]

20

Long-run effects on income, wages, and poverty

Dependent variable: Log earned income (1) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10

• 0.321**

(0.126) 11-19

• 0.272***

(0.098) 20-28

• 0.082

(0.090) Panel B: Average value of dependent variable in years 2000-13, by age in 1979, in levels 0-10 40,942 11-19 48,391 20-28 48,880

21

Long-run effects on income, wages, and poverty

Dependent variable: Log Log earned hourly income wage (1) (2) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10

• 0.321**
• 0.178*

(0.126) (0.099) 11-19

• 0.272***
• 0.318***

(0.098) (0.115) 20-28

• 0.082
• 0.118*

(0.090) (0.070) Panel B: Average value of dependent variable in years 2000-13, by age in 1979, in levels 0-10 40,942 25.52 11-19 48,391 29.81 20-28 48,880 32.04

21

Long-run effects on income, wages, and poverty

Dependent variable: Log Log Log earned hourly family income wage income (1) (2) (3) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10

• 0.321**
• 0.178*
• 0.366**

(0.126) (0.099) (0.166) 11-19

• 0.272***
• 0.318***
• 0.351***

(0.098) (0.115) (0.122) 20-28

• 0.082
• 0.118*
• 0.135*

(0.090) (0.070) (0.076) Panel B: Average value of dependent variable in years 2000-13, by age in 1979, in levels 0-10 40,942 25.52 80,892 11-19 48,391 29.81 93,896 20-28 48,880 32.04 98,157

21

Long-run effects on income, wages, and poverty

Dependent variable: Log Log Log earned hourly family In income wage income poverty (1) (2) (3) (4) Panel A: Interaction between 1978-82 decrease in log earnings per cap. and age in 1979 0-10

• 0.321**
• 0.178*
• 0.366**

0.168*** (0.126) (0.099) (0.166) (0.049) 11-19

• 0.272***
• 0.318***
• 0.351***

0.077** (0.098) (0.115) (0.122) (0.035) 20-28

• 0.082
• 0.118*
• 0.135*

0.018 (0.090) (0.070) (0.076) (0.025) Panel B: Average value of dependent variable in years 2000-13, by age in 1979, in levels 0-10 40,942 25.52 80,892 0.122 11-19 48,391 29.81 93,896 0.103 20-28 48,880 32.04 98,157 0.092

21

Long-run effects on income, wages, and poverty: summary and context

For people age 0-10 in 1979, a 10% decrease in earnings per capita leads to

• 3.2% ($1,314) decrease in earned income
• 3.7% ($2,961) decrease in family income
• 13.9% (1.7 p.p.) increase in poverty

22

Long-run effects on income, wages, and poverty: summary and context

For people age 0-10 in 1979, a 10% decrease in earnings per capita leads to

• 3.2% ($1,314) decrease in earned income
• 3.7% ($2,961) decrease in family income
• 13.9% (1.7 p.p.) increase in poverty

Chetty and Hendren (2016a,b) estimate effect of moving to better county

• Effect of 1 SD increase in recession severity on family income for people age

0-10 in 1979 is similar in magnitude to effect of 0.5 SD decrease in CH “county quality” throughout childhood

22

Long-run effects on income, wages, and poverty: summary and context

For people age 0-10 in 1979, a 10% decrease in earnings per capita leads to

• 3.2% ($1,314) decrease in earned income
• 3.7% ($2,961) decrease in family income
• 13.9% (1.7 p.p.) increase in poverty

Chetty and Hendren (2016a,b) estimate effect of moving to better county

• Effect of 1 SD increase in recession severity on family income for people age

0-10 in 1979 is similar in magnitude to effect of 0.5 SD decrease in CH “county quality” throughout childhood

Simple calculations and additional regressions suggest that much of the long-run effects on income are due to effects on education

22

Evidence supporting empirical strategy

Main threat: Even in absence of recession, long-run outcomes of people born in counties with more severe recession would have evolved differently than for people born in same state in counties with less severe recession

• For example: relative decline in infant health from 1950-79 in severe recession

counties

Evidence suggests that this type of threat is unimportant

• 1. Empirical strategy exploits sudden change in local economic activity
• 2. Birth certificate data show no relationship between recession severity and

pre-1980 evolution of infant mortality, birth weight, or maternal education

• 3. Falsification test for individuals age 23-28 in 1979

23

Mechanisms

Age profile of effects points to decrease in childhood HC or long-term decline in parental resources for college Parental job loss or decline in earnings

• Benchmark: long-run effects of parental job displacement
• Literature generates wide range of estimates, some of which include my results

[Page, Stevens, and Lindo 2007; Bratberg, Nilsen, and Vaage 2008; Oreopoulos, Page, and Stevens 2008; Coelli 2011; Hilger 2016]

Decline in community investment

• Local government expenditures did not fall until early 1990s
• Reductions concentrated in welfare and health, not education

Decline in opportunity cost of schooling

• Could be operative, but dominated by other factors

[Black, McKinnish, and Sanders 2005; Cascio and Narayan 2015; Charles, Hurst, and Notowidigdo 2015]

Decline in match quality or pay level of initial employer

• Could be relevant, especially for individuals age 18-22 in 1979

[Kahn 2010; Oreopoulos, von Wachter, and Heisz 2012]

24

Potentially mitigating policies

State transfer generosity Less More generous generous (1) (2) (3) (4) Interaction between 1978-1982 decrease in log real earnings per capita and age in 1979 0-10

• 0.378**
• 0.257*

(0.173) (0.133) 11-19

• 0.161
• 0.157

(0.131) (0.103) 20-28 0.0130 0.0417 (0.0657) (0.0586) p-value, equal effects 0.498

25

Potentially mitigating policies

State transfer generosity State transfer progressivity Less More Less More generous generous progressive progressive (1) (2) (3) (4) Interaction between 1978-1982 decrease in log real earnings per capita and age in 1979 0-10

• 0.378**
• 0.257*
• 0.339**
• 0.235*

(0.173) (0.133) (0.149) (0.139) 11-19

• 0.161
• 0.157
• 0.193**
• 0.0955

(0.131) (0.103) (0.0965) (0.152) 20-28 0.0130 0.0417 0.0135 0.0625 (0.0657) (0.0586) (0.0509) (0.0899) p-value, equal effects 0.498 0.956

25

The aggregate long-run effects of the 1980-82 recession

Scale diff-in-diff estimates by 105 million people born in U.S. from 1951-79 BOE calculation =

• a
• c

Na,c

• Number of people

born in county c age a in 1979

(R78−82

c Actual log earnings per capita change

− RCF

c

• CF log

earnings per capita change

)( ˆ πa

• Diff-in-diff

estimate

) Counterfactuals:

• 1. No change in real earnings per capita from 1978-1982 (RCF

c

= 0)

• 2. Pre-recession linear trend of 1.9 percent growth per year (RCF

c

= 0.076)

Aggregate effects:

• 0.9-2.1 million fewer four-year college graduates (1-3% of stock in 2015)
• $64-145 billion less earned income per year (0.4-0.8% of GDP in 2015)
• 0.5-1.3 million more adults in poverty per year (1-3% of stock in 2015)

26

Conclusion: The long-run effects of recessions

1980-82 recession led to sizable long-run reductions in education and income For people age 0-10 in 1979, a 10% decrease in real earnings per capita from 1978-82 leads to

• 3.0 p.p. (9.4%) decrease in four-year college degree attainment
• $1,314 (3.2%) decrease in earned income
• 1.7 p.p. (13.9%) increase in poverty rate

Back of the envelope calculations suggest that the recession led to

• 0.9-2.1 million fewer four-year college graduates (1-3% of stock in 2015)
• $64-145 billion less earned income per year (0.4-0.8% of GDP in 2015)
• 0.5-1.3 million more adults in poverty per year (1-3% of stock in 2015)

Every U.S. recession since 1973 has generated persistent relative decline in earnings per capita at county-level

• Other recessions might also have long-run effects on education and income

27

Every U.S. recession since 1973 has led to a sudden, persistent decrease in earnings per capita in some counties

• .12
• .1
• .08
• .06
• .04
• .02

.02 Percent difference in mean real earnings per capita

• 5

5 10 15 20 Years since start of recession

1980-1982

28

Every U.S. recession since 1973 has led to a sudden, persistent decrease in earnings per capita in some counties

• .12
• .1
• .08
• .06
• .04
• .02

.02 Percent difference in mean real earnings per capita

• 5

5 10 15 20 Years since start of recession

1973-1975 1980-1982

28

Every U.S. recession since 1973 has led to a sudden, persistent decrease in earnings per capita in some counties

• .12
• .1
• .08
• .06
• .04
• .02

.02 Percent difference in mean real earnings per capita

• 5

5 10 15 20 Years since start of recession

1973-1975 1980-1982 1990-1991

28

Every U.S. recession since 1973 has led to a sudden, persistent decrease in earnings per capita in some counties

• .12
• .1
• .08
• .06
• .04
• .02

.02 Percent difference in mean real earnings per capita

• 5

5 10 15 20 Years since start of recession

1973-1975 1980-1982 1990-1991 2001

28

Every U.S. recession since 1973 has led to a sudden, persistent decrease in earnings per capita in some counties

• .12
• .1
• .08
• .06
• .04
• .02

.02 Percent difference in mean real earnings per capita

• 5

5 10 15 20 Years since start of recession

1973-1975 1980-1982 1990-1991 2001 2007-2009

28

Additional slides

29

Details on instrumental variable

Isolate local labor demand shifts using 1976 industrial structure and aggregate employment changes: D78−82

c Shift to local labor demand in county c

=

• j

ηc,j,1976

Share of county c’s 1976 employment in 2-digit industry j

(e−s(c),j,1982 − e−s(c),j,1978)

• Log employment change for

industry j in all other states in same region [Freeman 1980; Bartik 1991; Blanchard and Katz 1992; Bound and Holzer 2000; Notowidigdo 2013; Diamond 2016]

• Constructed using Census County Business Patterns data

Back 30