Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 1 A gentle introduction to microsimulation Gijs Dekkers
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 2 What is microsimulation ? The purpose of any microsimulation model is to impute missing data Alternative realities Prospective scenarios Based on actual or synthetic micro-level datasets n i
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 3 What is microsimulation (2) Microsimulation models are used to evaluate the effects of changes (policy, socio-demographics, economic, …); at the level of the decision making units rather than focusing on the aggregate information. Focus is on the distribution of the target values, rather than on the means or the aggregates. Estimates of aggregate outcomes can however still be derived by summing up individual predictions. Units can be individuals, households, firms, bacteria, viruses, etc.
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 4 A classification of microsimulation models in social sciences Behavioural Dynamic ageing ArithmeticStatic ageing Include a notion of time Behaviour
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 5 Arithmetic (i.e. no time element, no behavioural equations) Take individual characteristics and behaviours as exogenous Evaluate the immediate “overnight” distributional impact of (possible) policy changes on individuals or households E.g. gross income_2=max(5000, gross income) Results are presented as counterfactuals or “alternative realities”. Applications: Traditional: redistributive and/or budgetary impacts of tax-benefit policy instruments, including income tax, family ad social assistance benefits, austerity measures. Indirect taxes and Consumption, Housing benefits Regional disparities from country-wide data
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 6 Arithmetic (i.e. no time element, no behavioural equations) The most well-known model of this type is EUROMOD Openly accessible, uses the EU-SILC Exists for all 27 EU member states + Serbia, Russia, RSA, China and …? Based at the Institute for Social and Economic Research, University of Essex. Other models: STINMOD (AUS), TAXBEN (UK), IZAψMOD (Ger), TRIM (US), MIMOSIS (B), MISIM (B)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 7 A classification of microsimulation models in social sciences ArithmeticStatic ageing Behaviour Include a notion of time
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 8 Static ageing Transform the weights of a dataset to meet J exogenous marginal distributions describing a (current or projected) population Deville and Särndal (1992), Creedy and Kalb (2006) Packages include GREGWT, CALMAR, Clan97 (SAS), Calibrate, Reweight, Sreweight, Gomulka (Stata); see Dekkers, IJM, forthcoming. “uprate” monetary variables and parameters following a hypothetical evolution Note that the individual characteristics remain unchanged!
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 9 A classification of microsimulation models in social sciences Dynamic ageing Static ageing Behaviour Include a notion of time
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 10 Dynamic ageing M t u < p x u~U(0,1) M t+1 N t+1 Basic principle: Monte-Carlo Simulation
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 11 Other techniques Matching Deterministic simulation Cloning Alignment through sorting M t : rank rank < p x % M t+1
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 12 Dynamic simulation: MIDAS as an example DEMOGRAPHIC MODULE t LABOUR MARKET MODULE t PENSION & BENEFITS MODULE t CONTRIBUTIONS AND TAXATION MODULE t REDISTRIBUTION, POVERTY, INEQUALITY OTHER OUTPUT t=2002 to 2060 MIDAS Starting dataset ± 2.2K2 individuals in 2002
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 13 MIDAS is working in conjunction with the other models of the FPB Macro model S3BE Semi-aggregate model MALTESE Microsimulation model MIDAS Demographic projections MALTDEMO Parameter changes Numbers/proportions* of individuals in various states Joint social hypothesis Alignment *
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 14 Applications of dynamic microsimulation Assess the consequences of the projections and hypotheses of the Study Committee on Ageing on poverty, income inequality Assess the consequences of the AWG projections and hypotheses on poverty, income inequality. Assess the consequences of policy, and demographic and economic trends on LWI Assess the (dynamic) consequences of indexation policy, fiscal policy, social policy (e.g. pension reform, austerity policy)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 15 An example: the WGA projections for Belgium (MIDAS) and Sweden (SESIM)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 16 What models are there Static (behavioural) models mainly in academics Static arithmetic models (EUROMOD) highly popular, both in academics and the government Static ageing models not very common Dynamic (reduced-form) behavioural ageing models increasingly popular, but outside academics SESIM, PENSIM II, MOSES MIDAS_BE, _LU, _HU, _?? DYNAMOD, APPSIM, INAHSIM DESTINIE, TRAJECTOIRE Dynamic arithmetic models (“Markov Chain type”) scarce and structural behavioural models almost non-existent
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 17 More information ? O’Donoghue (2015) Handbook of Microsimulation. Emerald Dekkers, Keegan, O’Donoghue (2014) New Pathways in Microsimulation. Ashgate.
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 18 Conclusions Microsimulation is now a well-established part of socio-economic (applied) research and policy evaluation On the EU-level, especially static ageing is now the standard, but attention is gradually turning to dynamics Many public or semi-public institutions in EU member states have dynamic models Still, the MSM research community remains fairly small
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 19 Now… on to LIAM2 !
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 20 LIAM2 tool for the development of microsimulation models mostly dynamic models (but static too) cross-sectional ageing tries to be easy yet flexible and powerful NOT a model in itself Developed in Python Very high level and readable language Free & Open source Large community (especially in the scientific world) Performance scales linearly with the sample size
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 21 Motivations behind the development of LIAM2 Most existing microsimulation models have been developed by separate (teams of) researchers. No economies of scale, No distribution of tacit or codified knowledge Inefficient Furthermore, as modellers often are not professional programmers, the result is not necessarily the most efficient in terms of simulation speed. This is the reason why several partners joined their efforts to develop a dynamic Microsimulation modeling toolbox (“LIAM2”)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 22 LIAM2 – a history Initiative by Gijs Dekkers and Philippe Liégeois; inspiration provided by Cathal O’Donoghue’s LIAM. Developed through a collaboration between the Federal Planning Bureau in Brussels (development by Gaëtan de Menten, testing by Raphaël Desmet and Gijs) LISER and the General Inspectorate of Social Security in Luxembourg (additional testing and complementary funding) under European funding (MiDaL Project , PROGRESS programme, Grant VS/2009/0569)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 23 LIAM2 – a history LIAM MIDAS (Belgium, Italy, Germany) LIAM2 MIDAS (Belgium) T-dymm (Italy) MIDAS (Hungary) MIDAS (Luxembourg) DYNAPOR (Portugal) Individual researchers Or independent teams (T-dymm, IPP, …)
Autumn School Dynamic MSM16-18 November 2015 | L-Esch-sur-Alzette Slide 24 And, finally… LIAM2 is free and open source liam2.plan.be Documentation New versions for download Source code Google groups