During t his w eek you w ill about many import ant aspect s - - PowerPoint PPT Presentation

GENDER AND WEALTH EQUALI TY

EVA SIERMINSKA

LISER-Luxembourg Institute of Socio- Economic Research IZA Bonn DIW Berlin 2nd Summer School on Gender Economics and Society, Torino, Italy July 6th –July 10th, 2015

2

I NTRODUCTI ON

P R E - I N E Q U A L I T Y

A N D

. . . . G E N D E R

• During t his w eek you w ill about many import ant aspect s

relat ed t o gender affect ing t he everyday lives of w omen and men : family policies, migrat ion, et hnicit y, income, invest ment st rat egies, polit ical part icipat ion, pensions, w ell-being over t he life-course and act ual t est imonials. Very import ant !

• All t hese aspect s feed int o t he t opic of t his lect ure,

w hich is w ealt h and w ealt h building. How ?

Family reconciliation policies Balance work and family Labor market participation Bargaining power within the household Individual savings Accumulated savings = wealth Retirement income Well‐being over the life‐course

4

I NTRODUCTI ON

P R E - I N E Q U A L I T Y

A N D

. . . . G E N D E R

• Here I w ould like t o show you not only t he ext ent t o

w hich gender differences exist in w ealt h, but also

• how to measure gender differences in wealth
• how to measure gender differences in wealth participation.

5

FAMOUS EQUATI ON ON ASSET BUI LDI NG

Wt=(1+r)Wt-1+St = (1+r)Wt-1+Yt-Ct

r-gross rate of return on investments Y t-income in period t C t – consumption in period t

In this model differences in wealth accumulation occur for 3 reasons:

• W/M differ in the amount they save (Y-C)
• Labor market participation (Warren et al 2001);
• Earnings gap (e.g. Blau & Kahn 1997, 2000)
• Occupation (e.g. Goldin 2014)
• Work/ family balance affects your Y
• W/M enter the period with different stocks of assets (W)
• Inheritance laws
• Probability of owning a home;
• discrimination in mortgage lending (Ladd 1998);
• lower income  lower credit scores (Sedo & Kossoudji 2004)

6

FAMOUS EQUATI ON ON ASSET BUI LDI NG

• W/M receive different rates of return (r)
• Differences in portfolio structure (reflecting variation in risk preferences) saving vs. investing
• Women invest more conservatively (Jiankokopolos & Bernasek 1998) More risk averse (Barsky

et al 1997)

• Concept of : wealth escalator and debt anchor (M.Chang 2010)

Additionally:

• Marriage patterns (Zagorsky, 1999)
• Legal environment:
• Joint ownership of assets (and debts) acquired during marriage
• Divorce laws
• Wealth accumulated prior to marriage remains in the hands of the original
• wner
• Marriage contracts can deviate from these standard regulations

7

VARI ABLES OF I NTEREST

• Tot al net w ealt h ~ net w ort h ( NW)
• Componet s of net w ealt h
• Financial assets= stocks+ bonds+ mutual funds+ deposit accounts+

saving accounts+ other financial assets

• Non-financial assets= main residence+investment real estate+

business equity

• Liabilities= mortgages + other debt
• Decision t o ow n asset s

8

Wealt h Escalat or

• allow s for w ealt h creat ion t hrough access t o w ealt h

building product s

• Saving and invest ing

Debt Anchor

• causes w ealt h dest ruct ion

9

WEALTH ESCALATOR ( LABOR MARKET PARTI CI PATI ON) Fringe benefit s

• Direct fringe benefit s
• Employer-sponsored retirement plans (access restricted, i.e. Full-time)
• Defined benefit plans (based on length of employment, age and

earnings history)

• Private pension savings plans (defined contribution) (women fewer

years to contribute and lower income) Women more often change jobs and thus have more opporutnities to cash out of these plans

10

WEALTH ESCALATOR ( LABOR MARKET PARTI CI PATI ON) ( CONT’D) Fringe benefit s

• Direct fringe benefit s
• I ndirect fringe benefit s
• Stock options and profit sharing plans
• Employer sponsored health insurance
• Life insurance
• Paid sick leave
• Flexible spending plans

Less fringe benefits in occupations where women cluster e.g. services

11

WEALTH ESCALATOR ( CONT’D)

• Direct fringe benefits
• Indirect fringe benefits
• Tax-incentives on private pension savings plans (IRAs)
• Long-term capital gains (lower tax)
• Home mortgage deduction

Government benefits are considered as being a trap for women (many means-tested and require liquidation of assets)

12

SAVI NGS VS. I NVESTI NG Saving ( put t ing away cash for a rainy day) I nvest ing ( commit ment of money in t he hope of event ual ret urn)

13

SAVI NGS VS. I NVESTI NG Saving ( put t ing away cash for a rainy day)

• Cash ( in deposit account s, under mat t ress)
• Important foundation of economic security
• Homeow nership
• Access to the wealth escalator (equity, tax advantages, capital gains if

sold)

• Home equity source of reserve
• Income creation (renting, reverse mortgages)

14

SAVI NGS FOR SI NGLE WOMEN AND MEN ( RATI O W/ M I N PARTI CI PATI ON AND LEVELS ( MEDI AN) )

15

SAVI NGS VS. I NVESTI NG I nvest ing ( commit ment of money in t he hope of event ual ret urn)

• St ocks
• I nvest ment real est at e
• Business asset s

16

I NVESTI NG FOR SI NGLE WOMEN AND MEN ( RATI O W/ M)

17

WHY ARE THERE GENDER DI FFERENCES I N I NVESTI NG? Wealt h building requires more risk t aking ( w omen are more risk-averse) Why? difference in access t o financial informat ion confidence in economic mat t ers gender socializat ion discouraging w omen from risk t aking differences in t est ost erone ( ! ! ! ) levels

18

• Labor market access
• Financial know ledge regarding t he w ealt h escalt or and

debt anchor

• Financial educat ion regarding saving and invest ing

19

POLI CY SOLUTI ONS

• Minimal paid parent al leave for bot h parent s
• I ncorporat e caregiving int o t he w ealt h escalat or ( include

years spent caregiving)

• I ncreasing men’s part icipat ion in caregiving
• Same hourly pay for part-t ime and full-t ime w ork
• Allow ing part-t ime w ork in many occupat ions ( same

chances for promot ion?)

• Great er financial educat ion for w omen t o be able t o

access t he w ealt h escalat or ( and avoid debt anchor)

20

DEBT ANCHOR

• “ Good” debt ( debt t hat allow s t o accumulat e more w ealt h

and includes t ax advant ages)

• Mortgages
• Education loans
• “ Bad” debt ( debt used for t he most part t o consume and

not hing else) Women are more likely t o have “ bad” debt

21

FAMOUS EQUATI ON ON ASSET BUI LDI NG

Wt=(1+r)Wt-1+St = (1+r)Wt-1+Yt-Ct

r-gross rate of return on investments Y t-income in period t C t – consumption in period t

22

LI FE-CYCLE PERSPECTI VE ( MODI GLI ANI )

23

LI FE-CYCLE PERSPECTI VE ( MODI GLI ANI )

24

AVERAGE WEALTH LEVELS FOR WOMEN AND MEN

P R E - I N E Q U A L I T Y

A N D

. . . . G E N D E R

Source: Household Finance and Consumption Survey (HFCS) wave 1

100,000 200,000 300,000 400,000 500,000 600,000 700,000 800,000 900,000 AT BE CY DE ES FI FR GR IT LU NL PT SI SK Women Men Single_Womwn Single_Men

25

I NTRODUCTI ON

P R E - I N E Q U A L I T Y

A N D

. . . . G E N D E R

I n a life-cycle set t ing 1 point w here t o observe t hese accumulat ed differences w ould be at ret irement and in fact t his is w hat w e see....

26

GOAL FOR TODAY

• Examine w here are t he differences in w ealt h

accumulat ion bet w een w omen and men coming from

• Met hods t hat allow us t o calculat e t he magnit ude of

t hese effect s

27

OUTLI NE 1. I nt roduct ion 2. Famous equat ion

1. Sources of differences between W/M 2. Examples

3. Concept of w ealt h 4. Research in w ealt h ( first st eps) 5. Met hods ( w it h applicat ions)

A. Levels (status quo—little explanation) B. Distribution and Inequality (summary measure) C. Regressions D. Decomposition methods

6. Summary 7. Ext ensions 8. References

28

RESEARCH STRATEGY

• Differences in w ealt h levels
• Differences in levels of t he component s ( int ensive

levels) —housing , financial asset s

• Decision t o ow n ( ext ensive level)

29

METHODS A. Levels ( st at us quo—lit t le explanat ion) B. Dist ribut ion and I nequalit y ( summary measure)

• A. Gini
• B. Other measures / graphical representation

C. Regressions D. Decom posit ion m et hods

• A. At the mean
• B. Along the distribution

30

• A. LEVELS ( MEAN VS. MEDI AN)

Mean—sum of all component s divided by t he number of

• bservat ions

Median --t he numerical value separat ing t he higher half of t he dat a from t he low er half.

n x X

i i





31

• A. LEVELS ( MEAN VS. MEDI AN)

Mean—

• Pros
• Easy to calcute and interpret
• Means of different wealth components will sum up to mean of net worth
• Cons
• Sensitive to outliers and to the asymmetry of the distribution (both

common to the wealth distribution)

Median—

• Pros
• More stable and robust measure
• Less affected by values at the lower and upper extremes of the

distribution

• Cons
• It’s not additive

32

• A. LEVELS ( MEAN VS. MEDI AN) EXAMPLE

Mean France Italy Median France Italy Difference France Italy All 462 013 280 816 All 94 612 173 000 All 367 401 107 816 Women 164 549 225 477 Women 49 271 152 445 Women 115 278 73 033 Men 246 261 309 125 Men 119 829 190 000 Men 126 432 119 125 Ratio_WM 0,668 0,729 Ratio_WM 0,411 0,802

Mean-median –crude measure of inequality

33

WEALTH GAP FOR WOMEN AND MEN AGED 60 AND OVER ( RATI O W/ M)

P R E - I N E Q U A L I T Y

A N D

. . . . G E N D E R

Source: Household Finance and Consumption Survey (HFCS) wave 1

0.000 0.200 0.400 0.600 0.800 1.000 1.200 1.400 1.600 1.800 2.000 AT BE CY DE ES FI FR GR IT LU NL PT SI SK SK Women_av Never Married Divorced Widowed 0.000 0.500 1.000 1.500 2.000 2.500 3.000 AT BE CY DE ES FI FR GR IT LU NL PT SI SK SK Women_md Never Married Divorced Widowed

34

Result s: Not e: Dat a collect ed at t he household level ( hence couples higher levels) Mean result s different are not so different from median result s, but rankings vary St at a: summarize [ var] , det ail

. su nw, detail nw

• Percentiles Smallest

1% -28960.5 -1370892 5% 0 -1226787 10% 1616 -1212101 Obs 218345 25% 25330 -1143296 Sum of Wgt. 218345 50% 151492 Mean 378287.8 Largest Std. Dev. 1653234 75% 354000 1.49e+08 90% 739261.6 1.49e+08 Variance 2.73e+12 95% 1224736 1.49e+08 Skewness 41.03189 99% 3726498 1.53e+08 Kurtosis 2724.83

36

METHODS A. Levels ( st at us quo—lit t le explanat ion)

• B. Dist ribut ion and I nequalit y ( summary measure)
• A. Gini
• B. Other measures

C. Regressions D. Decom posit ion m et hods

• A. At the mean
• B. Along the distribution

37

• B. I NEQUALI TY AND WEALTH

Most inequalit y measures are defined for non-zero, nonnegat ive values for posit ive component s of w ealt h— same as in t he income lit erat ure/ This w orks for posit ive holdings of many asset s and debt How ever, a common charact erist ic of net w ealt h dat a is t hat at various point s in t he life-cycle, hhlds m ay have negat ive or zero values of w ealt h. As a result , only a subset of inequalit y measures can be used

38

DI SRTI BUTI ON OF NET WORTH I N THE UNI TED STATES, 2007

5 10 15 20 Perce nt median

• 300000

mean 500000 900000 NETWORTH

Source: Survey of Consumer Finances

[R] histogram -- Histograms for continuous and categorical variables Syntax histogram varname [if] [in] [weight] [, [continuous_opts | discrete_opts] options]

39

KERNEL DENSI TY ESTI MATES OF NET WORTH I N THE UNI TED STATES

2 .000e

6 4 .000e

6

D ensity

• 500000

500000 1000000

NET WORTH

Source: SCF 2007 (solid line -20 bandwiths, dashed line -200 bandwiths)

40

KERNEL DENSI TY ESTI MATES I s a non-paramet ric way of est imat ing t he probabilit y densit y f( x) of a random variable. Many funct ions ( Deat on 2000) or Pagan & Ullah ( 1999) Here, , w here K( .) is t he kernel funct ion

St at a: kdensit y produces kernel densit y est im at es and graphs t he result . Synt ax kdensit y varnam e [ if ] [ in] [ w eight ] [ , opt ions]

41

• B. I NEQUALI TY AND WEALTH
• Coefficient of variat ion ( sensit ive at t he t op) [ 0.5CV2]

affected by inclusion/exclusion of just one v. high value

• Gini coefficient ( middle sensit ive)
• G=0 if all people have the same level of

wealth

• n – number of people; μ—mean of household wealth in the population
• ratio to the mean of half the average over all pairs (i,j) of the absolute

deviation of wealth (w) between households.

• Exponent ial measure ( bot t om sensit ive)

X CV

2

 





         

n j i n j j i

w w n G 

2

2 1

42

• B. I NEQUALI TY AND WEALTH

Raw Shave top and bottom 1% Shave top 1% & bottom 0.5% mean 556 846 378 215 559 361 median 120 780 120 780 123 800 Gini 0.82 0.74 0.81 1/2 CV2 18.1 2.42 14.63 p90/p10 30000 3369 3061 p75/p25 26.3 24.5 24.3 p90/p50 7.58 6.97 7.42 n 4418 3698 4359

43

• B. I NEQUALI TY AND WEALTH
• Theil and At kinson indices
• Non-negative values of assets and debts
• Theil useful to decompose the measure of inequality in a population

into the inequality that exists within the subpopulations an dthe inequality that exists between those populations [Cowell, 2011] STATA: ineqdec0 / ineqdeco

44

STATA: I NEQDEC0

. ineqdec0 nw Warning: nw has 10913 values < 0. Used in calculations Warning: nw has 1371 values = 0. Used in calculations Note: p5 (and smaller percentiles) <= 0 Percentile ratios

• All obs | p90/p10 p90/p50 p10/p50 p75/p25
• ---------+-----------------------------------------------

| 457.464 4.880 0.011 13.976

• Generalized Entropy index GE(2), and Gini coefficient
• All obs | GE(2) Gini
• ---------+-----------------------

| 9.54975 0.72608

45

STATA: I NEQDECO

. ineqdeco nw

Warning: nw has 10913 values < 0. Not used in calculations Warning: nw has 1371 values = 0. Not used in calculations

Percentile ratios

• All obs | p90/p10 p90/p50 p10/p50 p75/p25
• ---------+-----------------------------------------------

| 146.339 4.584 0.031 9.135

• Generalized Entropy indices GE(a), where a = income difference

sensitivity parameter, and Gini coefficient

• All obs | GE(-1) GE(0) GE(1) GE(2) Gini
• ---------+-----------------------------------------------------------

| 111.53839 1.36909 1.24274 8.92013 0.70367

• Atkinson indices, A(e), where e > 0 is the inequality aversion parameter
• All obs | A(0.5) A(1) A(2)
• ---------+-----------------------------------

| 0.44962 0.74566 0.99554

• Note: Do not use ineqdeco

if you have 0 and (-) values. Use ineqdec0.

46

METHODS A. Levels ( st at us quo—lit t le explanat ion) B. Dist ribut ion and I nequalit y ( summary measure)

• A. Gini
• B. Other measures

C. Regressions D. Decom posit ion m et hods

• A. At the mean
• B. Along the distribution

47

• C. REGRESSI ONS [ GUI SO, HALI ASSOS, JAPPELLI , 2001]

1. Levels condit ional/ uncondit ional / w / select ion ( ext ensive) 2. Decision t o ow n ( probabilit y of holding) ( int ensive)

• Standard discrete cross-section models (probit, logit)

I dent ify det erminant s of w ealt h holdings W= f( age, age^ 2, educat ion, gender, marit al st at us, parent s background, income, family size, number and age of children, labor market variables)

48

EXAMPLE: EXAMI NI NG GENDER WEALTH GAPS I N GERMANY ( SI ERMI NSKA, FRI CK, GRABKA, 2010)

=50, 000 =30, 000

49

INCOME AND LABOR MARKET SITUATION, GERMANY 2002

Male TOTAL Male married Male single

• never

married Female TOTAL Female married Female single - never married Demographics Age (in years) 47,1 53,3 30,5 49,4 50,2 32,0 Education University degree 21,4 25,5 12,4 14,6 15,7 13,7 Income Relative post-govt. income position 105 111 95 96 108 86 Relative labor income position 143 163 99 62 61 73 Labor market status FT employed 42,6 44,9 37,4 20,6 17,0 29,4 PT employed 2,0 1,5 3,5 13,5 19,3 4,6 self employed 7,3 7,7 4,8 2,7 3,1 2,6 not employed 25,7 33,0 6,6 42,5 46,0 13,3

50

EXAMPLE: EXAMI NI NG GENDER WEALTH GAPS I N GERMANY ( SI ERMI NSKA, FRI CK, GRABKA, 2010)

Are the wealth differences between women and men in Germany due to age differences, educational differences, labor market participation, income? (regs) Can we explain these differences with observable factors? If yes, which factor explains more of the differences? (decomp techniques) How much of the gap is unexplained?

Net worth Coeff Mean Coeff Mean lmarriage Length of marriage 169,95 20,34 206,67 20,77 nrmarriagesNumber of marriages ‐8878,6 ** 1,06 ‐15375,5 ** 1,07 migback Immigrant ‐43758,31 ** 0,14 ‐38022,11 ** 0,14 partner Have a partner ‐26384,98 ** 0,12 ‐8736,97 0,11 loc89east Lived in East Germany before 1989 ‐47856,58 ** 0,21 ‐38151,71 ** 0,21 kids04 Have kids under 5 years old ‐4628,14 0,12 ‐2375,84 0,11 _Iedu_2 Lower vocational education 24301,69 ** 0,51 10449,51 ** 0,49 _Iedu_3 Upper vocational education 38558,3 ** 0,12 31426,18 ** 0,14 _Iedu_4 University degree 54556,42 ** 0,13 45225,72 ** 0,21

• ver65

Being over 65 years old 826,26 1,10 970,63 * 1,73 autonom Have high job autonomy 10678,32 * 0,08 ‐19102,87 ** 0,20 perminc Permanent income 18226,57 * 7,91 196607,4 ** 10,02 expft02 Years working full‐time 1334,38 ** 12,67 2211,98 ** 26,09 exppt02 Years working part‐time 2397,09 ** 4,41 2224,02 ** 0,43 expue02 Years unemployed ‐283,39 0,66 823,48 0,71 notlabor Not in the labor force 1492,83 ** 18,68 2301,76 ** 11,23 expmiss Labor market experience missing 79440,16 ** 0,01 61892,52 ** 0,01 hiedu_f Father with higher education 25907,03 ** 0,06 ‐5286,01 0,06 hiedu_m Mother with higher education ‐5786,83 0,02 6258,57 0,02 hiedu_p Parent with higher education ‐13023,39 0,01 ‐303,07 0,01 inheri1 Recent inheritance (after 1992) 51185,75 ** 0,08 59062,95 ** 0,08 inheri2 "Old" inheritance (1949 till 1992) 52127,71 ** 0,06 63456 ** 0,08 Constant 22553,65 ** ‐21408,13 Observations 7439 7040 Women Men

52

• D. DECOMPOSI TI ON METHODS

A. Decomposit ion met hods

• A. At the mean
• Oaxaca-Blinder decomposition
• B. Along the distribution
• Dinardo, Fortin, Lemieux
• C. Fairlie
• D. Regression distr?

53

• Decompose outcome variable (wealth) into explained and unexplained

variation (i.e. explained variation (level of schooling) and unexplained variation (differences in returns in these endowments, new policies, discrimination, the way transformed into wealth)

D.DECOMPOSI TI ON METHODS

54

• Decompose outcome variable (wealth) into explained and unexplained

variation (i.e. explained variation (level of schooling) and unexplained variation (differences in returns in these endowments, new policies, discrimination, the way transformed into wealth)

• Decomposition of gender wealth gap:

1) The Oaxaca-Blinder decomposition

(endowment) (return) (interaction)

2) Dinardo, Fortin, and Lemieux (DFL) reweighting techniques

• 4 counterfactual distributions (Labor market experience, education,

intergenerational characteristics, demographic characteristics)

            X X X

W W W M ,

D.DECOMPOSI TI ON METHODS

55

• D. DECOMPOSI TI ON METHODS ( OAXACA-BLI NDER, 1973)

and

W = wealth; α = intercept; β = coefficient of years of schooling (S ); ε = error term; M = men; F = women.

• -term of interest

We want to look at wealth differentials between women and men, so we construct a count counter erfactual equation for women factual equation for women (intercept & coefficient replaced with those from men’s equation).

F F F F F

S W      

M M M M M

S W      

M F

W W 

F F M M F

S W      

*

56

• D. DECOMPOSI TI ON METHODS ( OAXACA-BLI NDER, 1973)

and

W = wealth; α = intercept; β = coefficient of years of schooling (S ); ε = error term; M = men; F = women.

• - term of interest

We want to look at wealth differentials between women and men, so we construct a count counter erfactual equation for women factual equation for women (intercept & coefficient replaced with those from men’s equation).

F F F F F

S W      

M M M M M

S W      

M F

W W 

F F M M F

S W      

*

] [ ] [

* * F F F M F M

W W W W W W     

57

• D. DECOMPOSITION METHODS

] [ ] [

* * F F F M F M

W W W W W W     

Characteristics effect Return/coefficients effect

F F M M F

S W      

*

58

What would the gender wealth gap be if women had the same aforementioned characteristics as their male counterparts?

Table 6: Two-fold Wealth decomposition results (Blinder-Oaxaca) Two-fold decomposition Women reference group Wealth gap Average wealth Women Amount of gap due to differences in characteristics Women's av. wealth if male characteristics Amount of gap explained with differences in coefficients + unexplained portion Average wealth Men 79562 98400 107761 28199 * 18838 * 9362 * 67% 33% Note: Women are the reference group.

] [ ] [

* * F F F M F M

W W W W W W     

Return/coefficients effect Characteristics effect

59

DI FFERENT REFERENCE GROUP

Two-fold decomposition Note: Men are the reference group. Men reference group Wealth gap Average wealth Women Amount of gap explained with differences in characteristics + unexplained portion Men's av. wealth if women's wealth function Amount of gap explained with differences in coefficients (1). Average wealth Men 79562 125200 107761 28199 * 45638 *

• 17439 *

162%

• 62%

1) Composition effect accounts for 2/3 of the gap 2) Differences in return offset the differences in characteristics almost entirely, leaving a huge portion of the gap unexplained

60

MAIN FINDINGS – OAXACA DECOMPOSITION

1) Composition effect accounts for 2/3 of the gap 2) Differences in return offset the differences in characteristics almost entirely, leaving a huge portion of the gap

unexpl ai ned

. oaxaca nw ageh educh inc, by(sexh)pooled

Blinder-Oaxaca decomposition Number of obs = 217675 Model = linear Group 1: sexh = 1 N of obs 1 = 122519 Group 2: sexh = 2 N of obs 2 = 95156

• | Robust

nw | Coef. Std. Err. z P>|z| [95% Conf. Interval]

• ------------+----------------------------------------------------------------
• verall |

group_1 | 456421.6 5844.825 78.09 0.000 444966 467877.3 group_2 | 279791.5 3013.46 92.85 0.000 273885.2 285697.7 difference | 176630.2 6575.935 26.86 0.000 163741.6 189518.8 explained | 98664.85 5723.364 17.24 0.000 87447.26 109882.4 unexplained | 77965.32 5804.058 13.43 0.000 66589.57 89341.06

• ------------+----------------------------------------------------------------

explained | ageh | 14493.15 738.5682 19.62 0.000 13045.58 15940.72 educh | -3143.034 576.669 -5.45 0.000 -4273.284 -2012.783 inc | 87314.73 5627.703 15.52 0.000 76284.64 98344.83

• ------------+----------------------------------------------------------------

unexplained | ageh | 131783.4 29020.1 4.54 0.000 74905.03 188661.7 educh | 112991.3 44685.1 2.53 0.011 25410.1 200572.5 inc | 100136.3 34941.77 2.87 0.004 31651.68 168620.9 _cons | -266945.6 41296.48 -6.46 0.000 -347885.2 -186006

62

SUMMARY OF BLI NDER-OAXACA

• Blinder-Oaxaca met hod ( 1973) : decomposit ion of mean

differences int o port ions at t ribut able t o …

• differences in the distribution of endowments
• differences in the return to these endowments / way in which the

endowments are transformed into wealth

• an unexplained part (discrimination)

Problem: requires an assumpt ion of a linear specificat ion ・

• ft en leads t o misspecificat ion

63

DI NARDO, FORTI N, LEMI EUX- SEMI PARAMETRI C DECOMPOSI TI ON



No param et r ic assu m pt ion abou t t h e con dit ion al m ean f u n ct ion



Use r ew eigh t in g t ech n iqu es an d con sider com par ison s of pr obabilit y den sit y f u n ct ion s



Vect or of ch aract er ist ics is par t it ion ed in t o f ou r gr ou ps:

 (1) Labor market experience (lifetime experience working full-time, part-time, being unemployed, not being in the labor force, being over 65 years of age, having high job autonomy, permanent income)  (2) Educational level (no or basic, lower vocational, upper vocational, university)  (3)Intergenerational characteristics (father with higher education, mother with higher education, parent with higher education, received a recent inheritance (since 1992), received an inheritance in the past (1949 to 1992))  (4) Demographic characteristics (have a partner, length of marriage, number of marriages, immigrant or German national coming from abroad, lived in East Germany before 1989, have children under the age of five)

) ( ) (

) , (

w g w g g

W M W M

   (.)

j

g is the marginal distribution of wealth w for group j ; For an observation with characteristics x and it can be expressed via



 dx x h x w f w g ) ( ) | ( ) ( . The conditional distribution .) | (. f can be thought of as being analogous to an estimated regression line and the marginal density of x , (.) h to the vector of characteristics. Next, we can specify each density separately by gender:



    dx i j x h x w f i j w g w g

j j

) | ( ) | ( ) | ( ) ( , where ) , ( , women men F M j  . With this we can specify various counterfactual densities. For example, What would be the wealth distribution of women if they had the characteristics of men?

 

       dx x F j x h x w f dx M j x h x w f F j w g w g

W W W CF

) ( ) | ( ) | ( ) | ( ) | ( ) | ( ) ( The innovation here is the reweighting function (.)  , which is defined in the following way: ) | ( ) | ( ) (

) , (

M j x h F j x h x

W M

    since ) ( ) | ( ) | ( i j P x i j P i j x h     then using Bayes’ Rule:

) ( ) | ( ) ( ) | ( ) (

) , (

F j P x M j P M j P x F j P x

W M

      , where unlike ) | ( i j x h  each of the components can easily be estimated (e.g., survey- weighted logits) ) | ( x i j P  is the probability that a randomly selected individual with characteristics x belongs to groupi if individuals from both groups are pooled in a common population ) ( i j P  is the probability that a randomly selected individual belongs to group j in a pooled population. ) (w g j can be estimated using Kernel density estimators. In this case the decomposition would be the following:

) ( ) ( ) ( ) (

) , ( F F CF F CF M W M W M

g g g g w g w g g       

The second component would express differences due to characteristics and the first would capture the unobservables.

66

MAIN FINDINGS – DFL DECOMPOSITION

• The gender wealth gap is mostly driven by differences in

income and labor market experience

67

MAIN FINDINGS – DFL DECOMPOSITION

• All other factors play little role to the explanation of the gap

68

CONCLUSIONS

• There is a significant gender wealth gap of about 30,000 euros in

Germany for men and women --- 50,000 euros for married couples

• The gap is largely driven by differences in characteristics between

men and women.

• The most important factor is the individual’s own income and labor

market experience, particularly for the bottom and top of the wealth distribution

• Differences for those in the middle of the distribution is mostly

driven by the wealth function (i.e. the way women transform their characteristics into wealth)

69

COMMENTS (1) – BACK TO THE QUESTION …

The gender wealth gap is at the mean as well as at various part of the wealth distribution Why is gender wealth gap a problem (of inequality) if the asset allocation is efficient ? In other words, the equal sharing rule may not be the optimal choice for both husband and wife. Empirical testing of different hypotheses is needed.

70

MECHANISMS ( WHY WOMEN HAVE SMALLER WEALTH ACCUMULATION THAN MEN?) 1) Labor market experience

… but does it further lead to discrimination in loan/ credit market, property ownership (home, vehicle)?

2) Risk aversion? Or spending more? 3) Institutional environment?

• Does the divorce law favor or against women (with respect to

wealth holdings)?

4) Protected by more generous public pensions?

• Maybe women are less likely to be self-employed, and more likely to

be in the civil services?

Another example, Neelakantan & Chang (2010) Risk preferences do not explain the wealth gap at retirement

71

• D. DECOMPOSI TI ON METHODS

A. Decomposit ion met hods

• A. At the mean
• Oaxaca-Blinder decomposition
• B. Along the distribution
• Dinardo, Fortin, Lemieux
• C. Fairlie
• D. Regression distr?

72

ASSET PARTI CI PATI ON FOR SI NGLE WOMEN AND MEN ( RATI O W/ M)

AT BE CY DE ES FI FR GR IT LU NL PT SI SK Single Women Cash 97 97 72 92 91 100 98 71 77 97 91 91 91 92 Stock 2 6 33 7 7 12 8 2 1 6 2 2 HO 28 53 78 20 68 43 34 43 55 56 33 54 68 79 IR 8 8 39 10 24 11 14 21 13 10 4 18 5 9 BA 3 3 11 4 4 6 5 3 8 3 4 3 5 Single Men Cash 90 95 71 92 94 100 96 74 77 96 93 87 48 87 Stock 6 8 36 9 9 15 12 1 7 12 10 4 7 1 HO 25 37 66 21 70 47 32 32 50 41 45 53 54 69 IR 7 11 40 13 28 16 17 18 15 27 6 21 7 12 BA 6 5 18 6 10 7 10 4 11 5 6 6 11 Cash 1,08 1,02 1,02 1,00 0,98 1,00 1,01 0,96 0,99 1,02 0,98 1,04 1,87 1,05 Stock 0,34 0,74 0,93 0,71 0,74 0,84 0,68 0,25 0,33 0,09 0,62 0,46 0,24 0,25 HO 1,11 1,41 1,18 0,93 0,98 0,91 1,06 1,35 1,09 1,36 0,73 1,02 1,25 1,15 IR 1,08 0,78 0,98 0,81 0,86 0,70 0,82 1,20 0,90 0,36 0,57 0,86 0,68 0,70 BA 0,46 0,68 0,64 0,64 0,39 0,79 0,45 0,62 0,75 0,51 0,68 0,53 ‐ 0,50 Saving and Investing.

73

1. What are t he det erminant s of w ealt h port folio part icipat ion? How do t hey differ bet w een w omen and m en? 2. How much of t he port folio diffference can w e explain by demographic and monet ary fact ors and how much by

• t her fact ors?

3. For differences across count ries w e could see w hat inst it ut ions correlat e w it h t he observed differences ( more so for w omen t han men, for example) ?

74

• D. DECOMPOSI TI ON: THE PARTI CI PATI ON DECI SI ON
• To model t he part icipat ion decision you can use an

ext ension of t he Blinder-Oaxaca nonlinear decomposit ion for binary variables ( Fairlie 1999, 2005)

• Estimate a logit for participation in a particular wealth component
• Examine the difference between two groups
• is t he count erfact ual part icipat ion for w omen,

given men’s coefficient s

) ( ) (  X F w pj  w )] ( ) ( [ )] ( ) ( [ ) ( ) (

^ ^ * ^ * ^ ^ ^

w p w p w p w p w p w p

M F F F M F

     men women j ,  ) (

^ * w

pF

75

• D. DECOMPOSI TI ON: THE PARTI CI PATI ON DECI SI ON
• As in t he DFL case, w e can furt her break dow n int o

set s of charact erist ics

• demographics [age, age2, no of children]
• education [low and high]
• labor [employed, self-employed, retired]
• marital status [married, divorced, single]
• income
• STATA: fairlie

X

78

1. Allow s us t o see w hat are t he driving det erminant s of w ealt h port folio part icipat ion and event ually how do t hey differ bet w een w omen and men 2. We can also see how much of t he port folio diffference can be explained and how much cannot be explained by

• bservable fact ors

79

SUMMARY A. Levels ( st at us quo—lit t le explanat ion)

i. Means vs. Medians [STATA: summarize [..], detail]

B. Dist ribut ion and I nequalit y ( summary measure)

i. 0.5 CV^2 [STATA: ineqdec0/ineqdeco] ii. Gini & other measures iii. Graphics [STATA: kernel/ histogram]

C. Regressions [ STATA: 0/ 1 probit , logit ] D. Decom posit ion m et hods

i. Oaxaca- Blinder [STATA: oaxaca] ii. Dinardo, Fortin, Lemieux [STATA: dfl] iii. Fairlie [STATA: fairlie] iv. Firpo [see Nicole Fortin website]

80

EXTENSI ONS

Nicole Fortin, Th. Lemieux & S. Firpo, Decomposition methods in Economics, NBER WP 16045, 2010

• > focus on distributional statistics (quantiles, Gini, variance) along with

examples.

81

REFERENCES

Ch an g , Mar i k o Sh o r t ch an g ed : W h y w o m en h av e l ess w eal t h an d w h at can b e d o n e ab o u t i t , 2 0 1 0 , Ox f o r d Un i v er si t y Pr ess Di Nar d o, J. Fo r t i n , Ni co l e an d T. Le m i e u x , Lab o r m ar k et i n st i t u t i o n s an d t h e d i st r i b u t i o n o f w ag es, 1 9 7 3 - 1 9 9 2 : A Sem i p ar am et r i c Ap p r o ach ,” Eco n o m et r i ca 6 4 : 1 0 0 1 - 1 0 4 4 . Do o r l ey & Si er m i n sk a, “ Cr o ss- n at i o n al d i f f er en ces i n w eal t h p o r t f o l i o s at t h e i n t en si v e m ar g i n : i s t h er e a r o l e f o r p o l i cy ? I ZA Wo r k i n g p ap er, 2 0 1 4 Fai r l i e, R ( 1 9 9 9 ) T h e ab sen se o f t h e Af r i can - Am er i can o w n ed b u si n ess: an an al y si s o f t h e d y n am i cs o f sel f - em p l o y m en t , Jo u r n al o f Lab o r Eco n o m i cs 1 7 : 8 0 - 1 0 8 . Fai r l i e, R. ( 2 0 0 5 ) An Ex t en si o n o f t h e Bl i n d er - Oax aca Deco m p o si t i o n Tech n i q u e t o l o g i t an d p r o b i t m o d el s” Jo u r n al o f Eco n o m i cs an d So ci al Measu r em en t 3 0 : 3 0 5 - 3 1 6 . Gu i so , Lu g i , Hal i asso s, Mi ch ae l an d Jap p e l l i , Tu l l i o Ho u seh o l d Po r t f o l i o s, 2 0 0 1 Neel ak an t an , Ur v i & Ch an g , Yu n h ee, 2 0 1 0 , Gen d er d i f f er en ces i n Weal t h at Ret i r em e n t , Am er i can Eco n o m i c Rev i ew : Pap er s & Pr o ceed i n g s, v o l . 1 0 0 p p 3 6 2 - 3 6 7. “ Si er m i n sk a & Do o r l ey " To Ow n o r No t t o Ow n ? Ho u se h o l d Po r t f o l i o s, De m o g r ap h i cs an d I n st i t u t i o n s i n a Cr o ss- Nat i o n al Per sp ect i v e,” I ZA Di scu ssi o n Pap er No . 7 7 3 4 , No v em b er 2 0 1 3 . Si er m i n sk a, Fr i ck & Gr ab k a " Ex am i n i n g t h e Gen d er Weal t h Gap ,” Ox f o r d Eco n o m i c Pap er s v o l . 6 2 , i ssu e 4 , 2 0 1 0 d o i : 1 0 . 1 0 9 3 / o ep / g p q 0 0 7.

Cont a ct m e : e va . sie r m insk a @lise r . lu