Ex-ante and ex-post evaluation of the 1989 French welfare reform using a natural experiment : the 1908 social laws in Alsace-Moselle

Transcription

1 Ex-ante and ex-post evaluation of the 1989 French welfare reform using a natural experiment : the 1908 social laws in Alsace-Moselle Matthieu Chemin (McGill, CIREQ) Etienne Wasmer (Sciences-Po Paris, LIEPP) June 2013, Public policy and poverty in developed countries, SciencesPo LIEPP Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 1 / 33

2 Introduction France has complex and (intellectually) interesting labor market regulations: compulsory training employment protection minimum wage working time regulation mix of workfare and welfare Unfortunately: little is known on their e ect (countrywide regulations, no regional variance) Identi cation typically rely on discontinuities based on age, rm size or hours. Crépon-Kramarz (2002) on the 1981 reduction in working time, etc. Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 2 / 33

3 Introduction Major reform on Dec. 1988: Minimum income, called Revenu Minimum d Insertion (RMI) Voted after a decade of long-term unemployment Approximately 40pc of the full time minimum wage Aim : reduce poverty and raise employability minimum income should help reinserting excluded individuals But possible disincentive work e ects: In 2007, 1.3 millions recipients for 2.4 millions job seekers Maximum RMI (2009, e/month): 455, +1 child: 682, couple + 1 child: 818. To be compared to the minimum wage (2009, e/month, net): part-time: 600, full-time: 1051 Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 3 / 33

4 Introduction Allocation di erentielle : 1 euro of labor income reduces transfer by 1 eurormi recipients are also eligible to cash and in-kind transfers: Subsidized housing ( aide au logement ), phone bill rebate, TV tax rebate, free public transportation, free museum, christmas bonus (Gurgand, Margolis, 2001) Overall: implicit marginal tax rate above 100%. In 2009, the RSA (Revenu Social d Activité) reform reduced the marginal tax rate to 38% How large are the disincentive e ects? Adressed in Terracol (2009), Gurgand and Margolis (2001), Bargain and Doorley (2009) Empirical question but surprisingly little scienti c debate at the time partly due to the lack of rigorous identi cation strategy, perhaps partly due to political consensus. Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 4 / 33

5 Identi cation Strategy 1 region (Alsace) and 1 sub-region (département de Moselle) were part of Germany between 1871 (Frankfurt treaty) and 1918 (Versailles treaty). Some favorable elements of the German legal system were kept (Droit Local). In particular, guaranteed minimum income (called aide sociale ) at the city level since 1908 In 1989, the rest of France generalized this system into a national law: any citizen above 25 years and below some income level became eligible to an allowance amounting to a large fraction of the minimum legal wage Aide sociale and RMI are substitutes (Woehrling, 2002) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 5 / 33

6 Identi cation Strategy Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 6 / 33

7 Identi cation Strategy In this paper, we use this particular feature of French institutions: Di erence-in-di erences: Alsace-Moselle = control group, rest of France = treatment group, before and after 1989 a similar stategy was used by Chemin, Wasmer (JOLE, 2009) to study the 35 hours workweek in 2000 We then calibrate key parameters of a search and matching model to replicate the di -in-di results We use the calibrated model to run a number of counterfactual policies, in particular a 38% marginal tax rate (RSA reform 2007) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 7 / 33

8 Preview of results: Unemployment Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 8 / 33

9 Preview of results: Employment Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 9 / 33

10 Preview of results Stable nding: the RMI policy was associated with: 3% fall in employment (among unskilled workers years old), leading to an estimated loss of jobs decline in the job-access rate 5-month increase in the average duration of unemployment larger disincentive e ects for single parents Search and matching model calibrated on the di -in-di : if a 38% implicit tax rate had been maintained as in the 2007 reform (RSA), instead of a 100% implicit tax rate due to the RMI the increase in unemployment would have been approximately half of its actual value Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 10 / 33

11 Data Longitudinal nature of France s Labor Force Survey, the Enquête Emploi (1982 to 2002) 44,663 individuals more than 24 years old, AND in a low-income household, AND who have been unemployed for more than 20 months or out of the labor force the year before (hence no longer eligible to unemployment insurance) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 11 / 33

12 Regressions Di -in-di : eligible individuals qualify for welfare payments in Alsace-Moselle and the rest of France after 1988, but only in Alsace-Moselle before 1988 employment ijt = α j + β t + γ 1 (Rest_of _France) (After1989) ijt +θx ijt + u it GLS, robust standard errors, clustered at the département level, Control Variables X Probit results not signi cantly di erent ( GLS/Probit ) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 12 / 33

13 Results Table 1: Impact of the RMI on transitions and stocks (sample restricted to the years old) (1) (2) (3) (4) (5) (6) (U+N) to E U to E N to U Search E U Dependent var. 1 if E, 0 if not 1 if E, 0 if not 1 if U, 0 if not Search e ort 1 if E, 0 if not 1 if U, 0 if not Sample Long term U or N Long term U N Long term U Unskilled (High school dropouts) Low income hh Low income hh Low income hh Low income hh a year before a year before a year before a year before Rest of France *After 1989 (8.41)*** (4.88)*** (1.24) (1.62) (5.50)*** (1.96)* Dep. fe Yes Yes Yes Yes Yes Yes Year fe Yes Yes Yes Yes Yes Yes Control var. Yes Yes Yes Yes Yes Yes Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 13 / 33

14 Common Time E ects Assumption Alsace-Moselle and the rest of France may be on a di erent time path (e.g. common border with Germany, or tougher negotiations in the rest of France) Use the less than 25 year olds as an additional control group: triple di erences employment ijt = α j + β t + γ 1 (Rest_of _France) (After1989) ijt +α j (More_than_25) ijt + β t (More_than_25) ijt +γ 2 (Rest_of _France) (After1989) (More_than_25) ijt +θx ijt + u it Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 14 / 33

15 Results Table 2: Triple di erences with the less than 25 years old (1) (2) (3) (4) (5) (6) (U+N) to E U to E N to U Search E U Dependent variable 1 if E, 0 if not 1 if E, 0 if not 1 if U, 0 if not Search e ort 1 if E, 0 if not 1 if U, 0 if not Sample Long term U or N Long term U N Long term U Unskilled (High school dropouts) Low income hh Low income hh Low income hh Low income hh a year before a year before a year before a year before Rest Of France *After 1989*(25) (2.48)** (4.17)*** (0.60) (4.23)*** (2.46)** (3.77)*** Rest of France *After 1989 (1.07) (1.92)* (0.63) (1.79)* (1.39) (4.21)*** Departement fe Yes Yes Yes Yes Yes Yes Year fe Yes Yes Yes Yes Yes Yes Control variables Yes Yes Yes Yes Yes Yes Year fe interacted Yes Yes Yes Yes Yes Yes with "More than 25" Dep. fe interacted Yes Yes Yes Yes Yes Yes with "More than 25" Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 15 / 33

16 Heterogeneous e ects Table 3: heterogeneous e ects Dependent variable: transition (U+N) to E: "1" if employed, "0" otherwise, restricted to the sample of Long term (20 months) U or N-Low income hh a year before (1) (2) (3) (4) (5) (6) (7) Baseline Part time Full time Single persons Single parents Couples without Couples with children children Rest Of France*After *More than 25 (2.48)** (4.22)*** (0.59) (0.97) (3.18)*** (0.46) (1.00) Rest of France*After (1.07) (3.35)*** (1.97)* (0.88) (1.84)* (0.69) (0.67) Departement fe Yes Yes Yes Yes Yes Yes Yes Year fe Yes Yes Yes Yes Yes Yes Yes Control variables Yes Yes Yes Yes Yes Yes Yes Year fe interacted Yes Yes Yes Yes Yes Yes Yes with "More than 25" Departement fe interacted Yes Yes Yes Yes Yes Yes Yes with "More than 25" Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 16 / 33

17 Robustness checks Appendix Table 2: Robustness checks Dependent variable: transition (U+N) to E: "1" if employed, "0" otherwise, restricted to the sample of Long term (20 months) U or N-Low income hh a year before (1) (2) (3) (4) (5) (6) Baseline No controls Treatment group DDD without dep 35, 25, 90, 38 Rest Of France*After *More than 25 (2.48)** (1.82)* (1.79)* (2.51)** Rest of France*After (1.07) (0.72) (0.55) (1.07) (1.32) Rest Of France*After *More than 25 (1.56) Rest of France*After (4.26)*** Rest Of France*After 1989*Long term Low income a year before (4.22)*** Departement xed e ects Yes Yes Yes Yes Yes Yes Year xed e ects Yes Yes Yes Yes Yes Yes Control variables Yes no Yes Yes Yes Yes Year xed e ects interacted Yes Yes Yes Yes Yes Yes with "More than 25" Departement xed e ects Yes Yes Yes Yes Yes Yes interacted with "More than 25" Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 17 / 33

18 Regression discontinuity design Appendix Table 3: Regression discontinuity design estimates Dependent variable: 1 if working, 0 otherwise (Sample restricted to High School drop-outs) (1) (2) (3) (4) (5) (6) (7) (8) Method Cubic speci cation Local linear reg. Falsi cation exercises Area Rest of France Rest of France Alsace Moselle Time Period After 1989 Before 1989 After 1989 Before 1989 Age interval 5 years 7 years 5 years 5 years 25 dummy * (2.91)*** (3.04)*** (3.04)*** (8.74)*** (0.83) (1.10) (1.90) Age (8.54)*** (8.46)*** (8.46)*** (6.03)*** (6.20)*** (2.80) (1.72) Age square (6.45)*** (6.32)*** (6.32)*** (10.78)*** (5.89)*** (2.34) (1.56) Age Cube (4.87)*** (4.69)*** (4.69)*** (13.51)*** (5.63)*** (1.97) (1.43) Departement fe Yes Yes Yes Yes Yes Yes Year fe Yes Yes Yes Yes Yes Yes Control variables Yes Yes Yes Yes Yes Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 18 / 33

19 Model Based on a simpli ed version of Garibaldi and Wasmer (JEEA 2005) Three-state model of the labor market with matching frictions à la Mortensen-Pissarides and stochastic value of leisure Endogeneous seach e ort, endogenous job creation and free-entry of rms Since UN unchanged and separations typically exogenous (Lise, Seitz, Smith 2005), other transition rates assumed exogenous Wage = minimum wage. rmi(0) = αw α : ratio of RMI to w, ' 0.45 rmi(h) = max(αw wh, 0) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 19 / 33

20 Model Flow incomes E: v E = hw + rmi(h) ψ(h) U: v U = rmi(0) ψ(e) N: v N = rmi(0) Bellman equations (continuous time) re = hw + rmi(h) ψ(h) + s(u E ) + s N (N E ) ru = rmi(0) ψ(e) + (p e)(e U) rn = rmi(0) + λ(u N) where ψ(e) is the increasing, convex function of e ort: ψ 0 e σ with σ 1 Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 20 / 33

21 Model Optimality condition : ψ 0 (e) = p (E U) where p depends on job creations Matching function M L (en U, V) and θ = V/(eN U ) labor market tightness p = M(eN U, N V ) en U q = M(eN U, N V ) N V with p 0 (θ) > 0 and q 0 (θ) < 0. General equilibrium feature: rm s free-entry : c yh wh q(θ = ) r + s u + s n = M(1, θ) = p(θ) = M(1/θ, 1) = q(θ) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 21 / 33

22 Model Proposition Labor market tightness does not depend on α, the level of RMI as compared to the minimum wage. The RMI has only an impact through a reduction in job search e ort e, with e α < 0. Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 22 / 33

23 Corollary The rate of unemployment is u = N U /(1 N N ) = e (α)p s+s N increases with the out ows from employment (s and s N ), decreases with job search e ort e, with the rate of job creation by rms p(θ ) Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 23 / 33

24 First part of the calibration on pre-reform data 1 Select years old, low skills individuals : less than high school degree (70% of the active population between 1982 and 1989) 2 Calculate yearly stocks, yearly and quarterly ows and average wage (1.24 of the minimum wage) 3 Choose parameters to t the average level of unemployment, its duration, labor market participation prior to the reform and various transition rates between E,U and N Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 24 / 33

25 Second part of the calibration strategy 1 Jointly: nd the disutility of additional e ort to t the di erence-in-di erence estimate of the economy post-rmi, which identi es the causal e ect 2 Denote by e0 the value of job search e ort when α = 0 (no RMI), and by e (α) the post-rmi value as calculated above. The causal impact of RMI on hazard rate is given by: (ep) = p(θ ) [e (α) e 0 ] Hence: choose σ to match this decline using the 7% decline of the DD estimates 3 Finally : using the full-parametrized model: i) calculate the value of work and compare it to value of search ; ii) run counterfactual experiments: simulate employment e ects if the 2007 RSA reform if it had been implemented right away in 1989 Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 25 / 33

26 Results Table 4: Benchmark calibration and counter-factual policy exercise: low skill sample (1) (2) (3) Initial calibration Counterfactual Policy variables No RMI RMI RSA Social transfer at 0 hours Social transfer at h hours Monthly wage (fraction of productivity) Hours worked h for employed workers (fraction of full-time) Targets for the calibration Unemployment rate N Reduction in job nding rate due to RMI N Parameters found by the code Scale parameter A M in the matching function Elasticity of e ort in the cost of e ort η ψ 8.9 Other parameters Scale parameter Aψ in the cost function 1 Elasticity of the matching function η M 0.5 Main endogenous variables % RSA/RMI E ort level Unemployment rate Hazard rate from unemployment to employment Derived endogenous variables % RSA/RMI Average duration of unemployment (years) Number of employed workers (fraction of total population) Number of inactive workers (fraction of total population) Gain from employment 1: (wh C (h)) Gain from employment 2: (wh C (h) -transfers at zero hours) Gain from employment 3: (wh C (h) -transfers at zero hours + C(e)) Present discounted value of being unemployed Present discounted value of being out of the labor force Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 26 / 33

27 in Results Table 5: Additional calibrations and counter-factual policy exercises Di erent hours: part-time Single parents Counterfact. Single parents (1) (2) (3) (4) (5) (6) (7) (8) (9) Initial calibration Initial calibration Initial calibration Policy variables No RMI RMI RSA No RMI RMI RSA No RMI RMI RSA Social transfer at 0 hours Social transfer at h hours Monthly wage (fraction of productivity) Hours worked h for employed workers (fraction of full-time) Targets for the calibration Unemployment rate N 14.5 N 14.5 N Reduction in job nding rate due to RMI N N N Parameters found by the code Scale parameter A in the matching function M Elasticity of e ort in the cost of e ort η ψ Other parameters Scale parameter A ψ the cost function Elasticity of the matching function η M Main endogenous variables E ort level Unemployment rate Hazard rate from unemployment to employment Derived endogenous variables Average duration of unemployment (years) Number of employed workers (fraction of total population) Number of inactive workers (fraction of total population) Gain from employment 1: (wh C (h )) Gain from employment 2: (wh C (h ) transfers at zero hours) Gain from employment 3: (wh C (h ) transfers at zero hours + C(e)) Present discounted value of being unemployed Present discounted value of being out of the labor force Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 27 / 33

28 Conclusion Di erence-in-di erences estimates: strong disincentive e ects of the RMI Using the di erence-in-di erences estimates, we then calibrate a model and assess the e ect of a major policy reform in 2007, the RSA, which provides additional income to those who work without a ecting the unemployed The disincentive e ects would have been drastically reduced with the RSA Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 28 / 33

29 Descriptive statistics Appendix Table 1: Descriptive Statistics of Enquete Emploi Alsace-Moselle Rest of France Proportion of unskilled, 25 to 55 years old: Mean SD Obs. Mean SD Obs. Long-term (>20 months) U or N, in a low-income hh Transition from category above last year to employment Search of Long-term (>20 months) U in a low-income hh Employed Unemployed Wage (French Francs) Employed (part-time) Employed (full-time) Single persons Single parents Couples without children Couples with children Control Variables: age: years age: years age: years Male secondary technical degree (CAP, BEP) general secondary degree (BEPC) general primary degree (CEP) no degree Household size (number) Transitions France before, age 25-55, unskilled Transition Full sample Single person Single parent EU EN UE UN NE NU Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 29 / 33

30 Alternative explanations Appendix Table 4: Alternative explanations (1) (2) (3) (4) Dependent variable Search: permanent job Search: demanding 1 if E, 0 otherwise Sample Long term U or N Rest of France Low income hh After 1989, Before 2000 After 2000 a year before Rest Of France*After 1989*More than (0.88) (1.43) Rest of France*After year (1.72)* (0.98) More than 25 dummy (2.87)*** (1.80)* Age (6.66)*** (7.28)*** Age square (4.73)*** (6.18)*** Age Cube (3.26)*** (5.35)*** Departement xed e ects Yes Yes Yes Yes Year xed e ects Yes Yes Yes Yes Control variables Yes Yes Yes Yes Year xed e ects interacted Yes Yes with "More than 25" Departement xed e ects Yes Yes interacted with "More than 25" Observations R-squared Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 30 / 33

31 Controls Back 95 department xed e ects 20 year xed e ects 5 age dummies sex size of household 15 diploma dummies Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 31 / 33

32 GLS/Probit Back We report GLS results: heteroscedastic robust standard errors all predictions between 0 and 1 ease of interpretation no normal or logistic assumptions on error term (as in probit/logit) chapter 15, Wooldridge Probit results not signi cantly di erent Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 32 / 33

33 Migration ows based on Census data % change of pop. due to migration ows Moselle 0.58% 0.55% 0.27% Bas-Rhin 0.16% 0.06% 0.34% Haut-Rhin 0.05% 0.08% 0.15% Hence: small, and increased after 1989, not decreased. Chemin, Wasmer ( ) The unemployment e ects of the RMI 06/13 33 / 33