Assessing the Impact of the Maternity Capital Policy in Russia Using - PowerPoint PPT Presentation

Assessing the Impact of the Maternity Capital Policy in Russia Using a Dynamic Stochastic Model of Fertility and Employment Fabián Slonimczyk Anna Yurko ICEF – Higher School of Economics Moscow IZA–HSE Workshop: Labor Market Adjustment in the Wake of the Great Recession. October 2012

Fertility and Economic Incentives • For decades now, fertility has been related to women’s labor supply decisions 1 Static models: Becker (1968), Willis (1974) 2 Life cycle models: Hotz & Miller (1988), Eckstein & Wolpin (1989) • A more recent phenomenon is the explicit use of economic incentives by governments concerned with demographic trends • Australia, France, Germany, the province of Quebec in Canada, and Spain have all offered “baby bonuses" to couples

Russia’s Demographic Crisis • Russia’s TFR over the period 2001–2005 was only 1.3 • To encourage women to have more children, the State Duma passed a law in December of 2006 establishing new measures of government support for families with children • Maternity Capital (MC)

Maternity Capital Policy • Starting in January 2007, women that give birth to or adopt a second or consecutive child are entitled to special financial assistance • Program scheduled to expire by the end of 2016 • Assistance consists of a certificate that entitles its holder to receive funds ($11,000) at any time after the child reaches the age of three • Women can apply for MC funds only once in their lifetimes and the money can be used for a limited number of purposes: 1 Acquiring housing 2 Paying for children’s education 3 Investing in the mother’s retirement fund

Overview • We estimate a dynamic stochastic discrete choice model of fertility and employment • Women are forward looking and make decisions in order to maximize their expected discounted lifetime utility • The MC policy enters the model through the budget constraint • Estimation based on maximum simulated likelihood • We simulate alternative policy scenarios • Preliminary findings 1 The MC policy does not seem to have had a strong impact on fertility 2 Women in Russia are sensitive to economic incentives, so a well-designed pro-natalist policy should be effective 3 The design of the MC policy, in particular the fact that it can only be used for specific purposes, deems it ineffective

Outline The Model 1 Model Solution and Estimation 2 Data Description Estimation Results 3 Model Fit 4 Simulations and Preliminary Conclusion

A model of fertility and labor supply • Women choose among discrete alternatives at each point in time  1 if no birth and no work     if no birth and work 2  j = 3 if birth and no work     4 if birth and work  • Only full-time work is considered • Fertility decisions are deterministic. Fertile period ends at age 40 • The decision process start at age 22 and ends at the official retirement age of 55

A model of fertility and labor supply • The woman’s objective function can be written � 54 � � ρ t − 22 U t ( c t , l t , n t , X t − 1 , N t , B t , S, m t ) E t =22 • Marital status evolves following a first-order markovian process Table Transitions • The specific functional form for the utility function is U t = c t + α 1 l t + ( α 2 + ǫ n t ) n t + α 3 I N t =1 + α 4 I N t =2 + α 5 I N t > 2 + β 1 c t l t + β 2 c t n t + β 3 l t n t + ( δ 1 n t + δ 2 l t + δ 3 I N t =1 + δ 4 I N t =2 + δ 5 I N t > 2 + δ 6 l t n t ) m t � + γ 1 X t − 1 + γ 2 S 1 + γ 3 S 2 + γ 4 S 3 + γ 5 S 4 � + γ 6 I N t =1 + γ 7 I N t =2 + γ 8 I N t > 2 + γ 9 B t l t

A model of fertility and labor supply • The budget constraint is written: c t = y f t l t + y o t + ( φ 1 + φ 2 H ) MCn t K − b 1 l t − b 2 n t − b 3 I N t =1 − b 4 I N t =2 − b 5 I N t > 2 • Women receive labor income y f t when employed and income from other household members y o t , including the spouse’s income when married t = c 0 + c 1 m t + c 2 t + c 3 t 2 + c 4 S 1 + c 5 S 2 + c 6 S 3 + c 7 S 4 log y o Other Income Regression

A model of fertility and labor supply • The earnings offer function depends on the woman’s accumulated human capital: log y f t − 1 + a 3 S 1 + a 4 S 2 + a 5 S 3 + a 6 S 4 + ǫ y t = a 0 + a 1 X t − 1 + a 2 X 2 t t , ǫ y • The two shocks ( ǫ n t ) are jointly normally distributed with zero mean, finite variance, and non-zero contemporaneous covariance • The shocks are assumed to be serially independent, so past realizations do not provide information on the future • Unobserved individual heterogeneity • utility of giving birth ( α 2 , δ 1 ) • utility associated with having children ( α 3 , α 4 , α 5 , δ 3 , δ 4 , δ 5 ) • baseline earnings ( a 0 )

Solution and Estimation • For given parameter values, the solution to the finite-horizon dynamic programming problem is found using backward recursion • Letting d i,t denote the combination of the choice and earnings, we have � � Pr( d i,t | Ω d t ) =Pr j = arg max V k (Ω t ) for j = 1 , 3 k � � Pr( d i,t | Ω d t ) =Pr j = arg max V k (Ω t ) k � � y f × Pr t | j = arg max V k (Ω t ) for j = 2 , 4 k • We generate the probabilities in the right hand side by solving the dynamic program for 20 simulations of the random shocks

Solution and Estimation • Given the serial independence of the shocks, the joint probability of a sequence of choices is 54 � Pr( d i, 22 , . . . , d i, 54 | Ω d Pr( d i,t | Ω d 22 ) = t ) t =22 • The introduction of unobservable types into the model modifies the objective likelihood function as follows H 54 � � Pr( d i,t | Ω d L i ( θ ) = µ h t , type = h ) t =22 h =1

The Data • The Russian Longitudinal Monitoring Survey • Rounds XIII–XIX (2004–2010) • In typical round, 10,000 individuals in 4,000 household • We use the family roster to create a fertility history for each woman in the panel • The adult questionnaire contains information on employment, earnings, and other characteristics • Sample is composed of women ages 22–54 and observed at least 3 times during the period • Unbalanced panel of 2,031 individuals and 12,117 person-year observations

Variable Definitions • Employment: A woman is considered employed if she usually works 10 or more hours per week at all jobs • Experience: Data used to determine experience in the first interview. We let experience evolve in a way that is consistent with the observed employment history • Births: Determined on the basis of the household roster • Number of Children: Data used to determine the number of children in the first interview. Evolution consistent with birth history • Marital Status: We consider a woman as married when there is a cohabiting spouse in the household roster

Comparing Rosstat and RLMS data 0.055 0.050 0.045 0.040 0.035 0.030 2004 2005 2006 2007 2008 2009 2010 Rosstat RLMS Figure : Birth Rates for Women Ages 15-49

Maximum Likelihood Estimates U t = c t + α 1 l t + ( α 2 + ǫ n t ) n t + α 3 I N t =1 + α 4 I N t =2 + α 5 I N t > 2 + β 1 c t l t + β 2 c t n t + β 3 l t n t + ( δ 1 n t + δ 2 l t + δ 3 I N t =1 + δ 4 I N t =2 + δ 5 I N t > 2 + δ 6 l t n t ) m t � + γ 1 X t − 1 + γ 2 S 1 + γ 3 S 2 + γ 4 S 3 + γ 5 S 4 � + γ 6 I N t =1 + γ 7 I N t =2 + γ 8 I N t > 2 + γ 9 B t l t log y f t − 1 + a 3 S 1 + a 4 S 2 + a 5 S 3 + a 6 S 4 + ǫ y t = a 0 + a 1 X t − 1 + a 2 X 2 t c t = y f t l t + y o t + ( φ 1 + φ 2 H ) MCn t K • α 1 , the disutility of work, is negative as expected. In

Predicted vs. Actual Behavior 0.45 0.8 data model, simulated 0.75 0.4 0.7 0.35 0.65 0.3 0.6 0.25 0.55 0.5 0.2 data 0.45 model, simulated 0.15 25 30 35 40 45 50 25 30 35 40 45 50 Age Age (a) No Work – No Child (b) Work – No Child data data 0.07 model, simulated model, simulated 0.06 0.06 0.05 0.05 0.04 0.04 0.03 0.03 0.02 0.02 0.01 0.01 0 0 25 30 35 40 45 50 25 30 35 40 45 50 Age Age (c) No Work – Child (d) Work – Child

Predicted vs. Actual Behavior data 0.8 0.09 model, simulated 0.08 0.75 0.07 0.7 0.06 0.05 0.65 0.04 0.6 0.03 0.02 0.55 data 0.01 model, simulated 0.5 0 25 30 35 40 45 50 25 30 35 40 45 50 Age Age (a) LF Participation (b) Total Births

Data versus Model: Analysis by Type Births (per 1,000) Participation Rate 13 . 7597 0 . 9802 Type 1 9 . 1318 0 . 9892 34 . 0408 0 . 1005 Type 2 34 . 5695 0 . 1072 19 . 0583 0 . 7722 Type 3 16 . 5762 0 . 7759 19 . 8894 0 . 7224 All 17 . 1140 0 . 7289 Note: Gray cells contain model predictions based on 200 simulations.

Simulations Births Participation N X (per 1,000) Rate avg. avg. 22 . 584 0 . 645 1 . 186 22 . 428 Baseline model MC policy efficacy ( φ 1 ) 0 . 1 +16 . 367 − 0 . 012 +0 . 594 − 0 . 413 0 . 5 +21 . 055 − 0 . 021 +1 . 007 − 0 . 721 1 +15 . 565 − 0 . 027 +1 . 025 − 0 . 941 Net utility of birth ( α 2 ) +5000 +14 . 434 − 0 . 014 +0 . 524 − 0 . 448 +10000 +23 . 836 − 0 . 024 +0 . 896 − 0 . 780 Net utility from children ( α 3 – α 5 ) +500 (per child) +19 . 670 − 0 . 025 +0 . 758 − 0 . 833 +1000 (per child) +28 . 461 − 0 . 041 +1 . 193 − 1 . 334