Challenges and Opportunities of Dynamic Microsimulation Modelling

1
Challenges and Opportunities of Dynamic
Microsimulation Modelling

• Ann Harding
• Plenary presentation to the 1st General
  Conference of the International Microsimulation
  Association,
• Vienna, 21 August 2007

National Centre for Social and Economic
Modelling (NATSEM), University of Canberra
2
What are dynamic microsimulation models?

• Try to move individuals forward through time, by
  updating each attribute for each micro-unit for
  each time interval
• Dynamic cohort models move single cohort or
  number of cohorts forward through time
• Lifepaths, Harding, LIFEMOD, ODonoghue
• Dynamic population models move representative
  sample of population forward through time
• Start with large microdata sets (often admin data
  or Census 1 samples) use behavioural equations
  to simulate state transitions (e.g. single to
  married)

3
A Dynamic MSMs Simulation Cycle
V0.2 8 Feb 2006
4
Why has dynamic MSM flourished in past decade?

• Orcutt conceived vision of dynamic MSM in late
  50s had DYNASIM model finished in 1970s
• Dynamic MSM languished in 80s and early 90s
• Policy challenges of population ageing require
  dynamic rather than static MSM to analyse them
• Future distributional impact rather than current
  impact is crucial
• Retirement incomes take decades to accumulate -gt
  long term policy effects (and inter-generational
  effects)
• Are current pension and health care systems
  affordable in 20 years time?

5
Major progress in past 10 years

• In many countries dynamic MSM now being used
  actively in government policy formation
• SESIM (Sweden) MOSART (Norway) SAGE PENSIM
  (UK) DYNACAN (Canada) DYNASIM3, MINT CBOLT
  (US)
• But many challenges remain
• Following based on authors recent experience in
  constructing APPSIM in Australia (and,
  previously, DYNAMOD)

6
Budgets and project management

• Construction is a multi-million dollar exercise
• US6m for CBOLT, MINT and POLISIM
• Similar for Nordic models and DYNACAN
• Academic models often constructed on shoestring
  budgets
• Need strong project management skills
• More important to produce a working model on time
  and within budget
• keep it simple but credible
• Keep stakeholders engaged by producing results

7
Collaboration and Co-operation

• Now an accumulated body of knowledge about how to
  build dynamic MSM
• Marriage market, alignment to macro totals,
  stochastic error reduction
• Some examples of collaboration -gt quicker
  construction
• DYNACAN and POLISIM leveraged off CORSIM
• Australian Treasury had access to MITTS model
• Collaboration between academics and government
• Produces benefits for both parties
• Construction outside policy hot house of govt can
  help documentation maintenance (eg STINMOD)
• Govt funding probably required if models are to
  survive outside govt
• Can IMA do more to help foster collaboration?
• Website divided into more specific sections eg
  Dynamic MSM earnings
• Can we help technology transfer? (eg, visiting
  fellowships)

8
Quality of microdata used to estimate transitions

• In many countries longitudinal sample survey
  microdata used
• Often short period (4 yrs) and small samples
• Endless fine tuning of predictive equations can
  be a waste of time
• For example, theoretical model might suggest
  likelihood of having a child depends upon
• Mothers age, education, marital state, duration
  since last birth, current labour force status,
  duration of relationship

9
Limits to accuracy of predictive equations
Source From Bacon and Pennec
10
Alignment

• Five year snapshot captured in longitudinal
  sample survey data may not capture long term
  dynamics
• Increasing age at first marriage
• Or behaviour may change quickly in future
• Consensus has emerged about need to align summed
  micro projections to macro targets
• Official population or actuarial projections
• Expected labour force participation rates (women
  increasing)
• MIC-MAC bridge between aggregate demographic
  projections of cohort behaviour and projections
  of lifecourses of individual cohort members
• How many components of model should be aligned
  and at what degree of disaggregation?

11
User friendliness

• Modularisation of model structure
• Holding key parameters separate from the model
  source code
• Make output easily available (eg DYNACAN)
• User consultation

12
Behavioural response

• Most dynamic MSMs are dynamic arithmetical
• Individual behaviour does not change in response
  to govt tax-transfer policy change
• Individual behaviour does not change in response
  to macroeconomic change due to govt policy change
• In DYNASIM3, all demographic and labour force
  histories first simulated and written to file and
  then rules of tax-transfer programs simulated -gt
  changes in SS cant affect retirement behaviour
• Some progress being made on these fronts
• MOSART linked to macroeconomic effects, SESIM
  apparently incorporates labour supply
• But extremely difficult task

13
Conclusions

• Construction of a dynamic model is a daunting
  task requiring large investments and good project
  management
• Despite improved computer hardware and software,
  data demands remain problematic
• But population ageing should prompt on-going
  investment in this technology
• And success of these models in last decade in
  many countries is very encouraging!