Federaal Planbureau Economische analyses en vooruitzichten Integrating a random utility random opportunity labour supply model in MIDAS Belgium: presentation of on-going work Gijs Dekkers, Federal Planning Bureau CESO, KU Leuven CEPS/INSTEAD André Decoster CES, KU Leuven Bart Capéau CES, KU Leuven European Meeting Of The INTERNATIONAL MICROSIMULATION ASSOCIATION, October th, 2014, MAASTRICHT
Economische analyses en vooruitzichten Federaal Planbureau Integrating a random utility random opportunity labour supply model in MIDAS Belgium –Current versions of MIDAS include simple, reduced-form behavioural equations –Not ideal for reform analysis –complicating factor: MIDAS is dynamic –Another complicating factor: alignment –This presentation reports on on-going work to introduce the “random utility– random opportunity model” (a.k.a. RURO) in the dynamic-ageing microsimulation model MIDAS of Belgium. –Brief overview of this presentation A birds-eye view on RURO Simulation in LIAM2: a simple example of code Oh, static is static, and dynamic is dynamic, and never the twain shall meet. Wage thrift Stability Alignment Some preliminary results
Economische analyses en vooruitzichten Federaal Planbureau standard model –choice of discrete h –h: uniform distr. –gross wage given –tax-benefit system –functional form U(.) –assumptions about stochastic part –=> prob (h) RuRo-model Oslo model – choice of j: (h,w,k) – h: non uniform – gross wage distrib. – tax-benefit system – functional form U(.) – assumptions about stochastic part – => prob (h,w)
Economische analyses en vooruitzichten Federaal Planbureau probability: standard multinomial logit-model (relative attractiviness of the choice) RuRo weighted by measure of ‘availability’RuRo-model
Economische analyses en vooruitzichten Federaal Planbureau Structural => empirical specifications –preferences –opportunities (job availability)RuRo-model
Economische analyses en vooruitzichten Federaal Planbureau preferences: Box-CoxRuRo-model
Economische analyses en vooruitzichten Federaal Planbureau preferences couples unitary modelRuRo-model
Economische analyses en vooruitzichten Federaal Planbureau job availability –market versus non-market –market subsetRuRo-model
Economische analyses en vooruitzichten Federaal Planbureau job availability –market subset wages: lognormal (covariates: age, education) hours:RuRo-model
Economische analyses en vooruitzichten Federaal Planbureau what is identified? hinges on the separability of g(h,w) non parametrically identified: –v(C,h).g 2 (h) –q0–q0 –g 1 (w)RuRo-model
Economische analyses en vooruitzichten Federaal Planbureau ML-estimation –200 draws to approximate Choice Set on EU-SILC 2007 –571 single females –449 single males –1457 couples tax benefit simulator of EUROMODRuRo-model
Economische analyses en vooruitzichten Federaal Planbureau coefficients for utility function coefficients for opportunities –market versus non market (q 0 ) –hours (peaks): g 2 (h) –wage distribution: g 1 (w)RuRo-model
Economische analyses en vooruitzichten Federaal PlanbureauRuRo-model malesfemales CoeffSEt-valueCoeffSEt-value Leisure coefficients M/F in couples exponent constant ln(age) ln(age)^ # children between 0 and # children between 4 and # children between 7 and region WAL region BXL Educ LOW Educ HIGH Leisure coefficients single M/F exponent constant ln(age) ln(age)^ # children between 0 and # children between 4 and # children between 7 and region WAL region BXL Educ LOW Educ HIGH Wage equation M/F Sigma (RMSE) constant potential experience potential experience^ Educ LOW Educ HIGH Some quite very extremely preliminary estimation results
Economische analyses en vooruitzichten Federaal Planbureau RURO in MIDAS BE: A simple example of LIAM2 code ad_earnings: args: gender, age code: [...] return: [...] ad_welfare: args: income code: [...] return: [...] ad_unemployment: args: entitlement conditions code: [...] return: [...] utility_optimisation: - i: 1 - max_u: 0 - utility_rndm: normal(0.0, 1.0) * while: cond: (i < 200) code: - joboffer: [make a MC simulation] - hours: if(joboffer, [make a random draw of discrete hours], 0) - hourly_wage: if(joboffer, ad_earnings(gender, age), 0) - incomeW: if(joboffer, hourly_wage * hours, ad_unemployment(...)) - welfare: ad_welfare(incomeW) - leisure: 1 - hours / (168 * 52) - utility: function of (incomeW + welfare, leisure, utility_rndm) - max_u: max(max_u, utility) - opt_hours: if(i == 0, hours, if(max_u == utility, hours, opt_hours)) - i: i + 1 Function: generate earnings Function: generate unemployment benefit Function: generate welfare benefit 200 iterationsTake max(utility) utility Draw a number of hours (or not) Does the individual gets a job offer? Optimal choice after i iterations
Economische analyses en vooruitzichten Federaal Planbureau RURO in MIDAS BE: MIDAS is dynamic Wages increase with productivity Social and fiscal parameters increase, but at a lower rate in the short and middle run This will cause the RURO model to keel over as simulated time goes by
Economische analyses en vooruitzichten Federaal Planbureau Complicating factor: MIDAS is dynamic Starting dataset ± 2.2K 2 individuals in 2002 DEMOGRAPHIC MODULE t LABOUR MARKET MODULE t PENSION & BENEFITS MODULE t CONTRIBUTIONS AND TAXATION MODULE t REDISTRIBUTION, POVERTY, INEQUALITY OTHER OUTPUT Simulate earnings i, t= A* Simulate alternative incomes i, t= A Derive net income i, t= A Select hours where U(i)=Max, t Simulate job-offers i, t Draw hours i t=2002 to 2060 i= 1 to 200 A = year of estimation – currently 2007 * Stochastic components are constant over t (exception is ‘joboffer’ and only the random component of earnings changes with labour market transitions). Derive utility i, t= A* RURO MIDAS
Economische analyses en vooruitzichten Federaal Planbureau Complicating factor: alignment It is of course sad, but MIDAS is being used in a policy-assessment environment. Therefore, we use alignment by sorting to be able to assess policy measures in conjunction with a semi-aggregate model (see Dekkers, Inagaki and Desmet, 2012) Alignment includes: –Who works and who does not –Unemployment –Early retirement/CELS –Private and public sector employment –… –And all this to age, gender and period Hence, heterogeneity in choice sets needs to be included in an alignment procedure in simulation. Who receives a job-offer at period t? –‘risk’ based on individual characteristics, using estimation results of RURO –Aligned to gender, age and period
Economische analyses en vooruitzichten Federaal Planbureau Complicating factor: alignment Foreach t = 2002 to 2060 Foreach i = 1 to 200 MC simulation of inversion at i Logit simulation of ‘risk’ joboffer J(i) at i, given working(t – 1) Joboffer(i)=inverse(joboffer(i-1)) Joboffer(i) = ALIGNMENT(age, gender, t) If inversion at i simulation of hours h simulation of earnings at A If Joboffer(i) Unemployment benefit if eligible at t Apply means-test for welfare at A Add family benefits Derive net total income at A Derive utitlity(i) MAX=max(MAX, utility(i)
Economische analyses en vooruitzichten Federaal Planbureau Some extremely preliminary simulation results
Economische analyses en vooruitzichten Federaal Planbureau Some extremely preliminary simulation results
Economische analyses en vooruitzichten Federaal Planbureau Some extremely preliminary simulation results
Economische analyses en vooruitzichten Federaal Planbureau Some extremely preliminary simulation results
Economische analyses en vooruitzichten Federaal Planbureau Integrating a random utility random opportunity labour supply model in MIDAS Belgium Thank you
Economische analyses en vooruitzichten Federaal Planbureau Assumptions and hypotheses of the Study Committee on Ageing Key demographic hypotheses Fertility Life expectancy at birth Men women Key macro hypotheses Up to ≥ 2015 Yearly productivity0.01%1.28%1.50% Unemployment rate14.75 in 2014Decreasing towards 8% Social policy hypotheses ≥ 2015 Wage ceilingCurrent legislation1.25% Minimum right per working year1.25% Welfare adjustment non-lump-sum benefits Employed and self-employed 0.50% Welfare adjustment of lump-sum benefits1.00%