Measuring the Economic Impact of Pension Reform with Microsimulation: an Introduction

1. Measuring the Economic Impact of Pension Reform with Microsimulation: an Introduction Elisa Baroni National University Ireland, Galway Institute for Futures Studies, Stockholm Sept. 20, 2006

2. About me…. • MSc Economics (2003) from LSE, PhD Economics (ongoing). • Since 2004, Micro-simulation Modelling of tax/benefit policy + pension systems. Team leader of DWP policy micro-simulation model (PSM). • Currently at IFS, Stockholm. Developing new simulation model for Sweden. • Research Focus: developing micro-simulation model for analysing the impact of pension reforms on poverty and inequality (Ireland + Sweden)

3. Premise • Pension Systems redistribute incomebetween • workers and retirees • birth cohorts • individuals’ life periods • Redistributive Systems change: • households’ incomes and povertylevels • households’ positions in the income distribution and inequality levels • Pension Systems and Pension Reforms affect Poverty and Inequality levels (e.g. pre- and post-retirement incomes)

4. Our Questions… • What is the Redistributive outcome of a Pension System or Pension Reform ? • What will their Redistributive outcome be in the future ? • Who pays what and whogains what under a given Pension System ? • Whatincentives or disincentives are created? • What are the costs of Pension Reform ?

5. Our Answers… • We can measure (redistributive) outcomes of Pension Reform, e.g. : • Winners and Losers • Change in Poverty + Inequality indicators • Poverty Reduction Efficiency measures • Related Distortionary Impact measures (e.g. effective marginal tax rates) • Redistributive outcome does not depend only on the Pension System / Reform alone  need to study interactionswith the whole Tax and Benefit System + Demographic + Labour Market trends

6. Our Tool: Micro-Simulation Modelling • MSM are micro-based statistical tools used by many governments to inform social policy decisions, including Pension reforms. • MSM simulate how heterogeneous individuals are affected by policy changes (incomes+ behaviours) • MSM allow to compare costs + redistributive outcomes of alternative reforms  choose the policy which fits better government’s aims

7. Our Aims…. • To Introduce MSM • To understand MSM key features + advantages + limitations • To understand MSM use for measuring effects of Pension Systems and Pension Reforms • To show examples of models used for pension reform analysis in OECD (EU, UK, Ireland, Sweden) • To discuss applicability of MSM to Developing Economies • To summarise: how to build a MSM

8. An (incomplete) map of MSM DYNACAM LIFEPATHS , MOZART SESIM SVERIGE MICROHUS , POLIMOD, PSM, PENSIM II, IGOTM, TAXBEN LIAM / SMILE EUROMOD NEDYMAS ESPASIM , DYNAMITE , DESTINIE , DYNASIM CORSIM BRAHMS , NATSEM STINMOD ,

9. “ The Department for Work and Pensions (DWP) has for the past few years been building and validating a dynamic micro-simulation model called Pensim2. This is a highly sophisticated model which attempts to mimic the evolution of both private and state pension accumulation and decumulation between now and 2050. We have used Pensim2 to help inform our recommendations for the UK private and state pension systems. In particular we use Pensim2 to estimate the cost of state pension reforms, the number of individuals on Pension Credit and the impact of the proposed National Pension Savings Scheme on private pension incomes.” 2nd UK Pension Commission Report, November 2005

10. Uses of MSM • Projections • Evaluation of Public Policy e.g. effects on Inequality and Redistribution • Designing new Policy Reforms • International comparisons of Policy Reforms • Studies of inter-temporal processes / behaviours

11. MSM Structure

12. Types of MSM • Static MSM: simulates effects of Policy Change on net incomes without demographic, economic or behavioural changes  not suitable if major population changes expected • Dynamic MSM: simulates effects of Policy Change in conjunction with demographic, labour market and behavioural changes  suitable if e.g. population aging expected

13. Input Data • Without data, no MSM ! • Cross-Sectional or Panel Data • Variables include e.g. demographic, socio-economic, health, housing information • Often used data sources: • Household Surveys • Administrative Data (!) • Census Data • Synthetic Data

14. Cross-Section + Panel Data Set X = Age

15. IT Infrastructure • Large memory needed (1 Gigabyte) • Programming software e.g. : • Excel • SAS • Stata • Object-Oriented Java languages e.g. C ++ • Existing modelling platforms e.g. Genesis, LIAM, JAS, SIMULA • Average running time: • Static: 2 minutes of CPU time for 300,000 individuals • Dynamic: 2 hours for 300,000 individuals

16. Static MSM • Uses cross-sectional data as inputs • Evaluates the short term effects of policy change: first round impact • Accounting tool • Static Aging: • future population is aged to look like current population. • Monetary variables are “up-rated” using exogenous assumptions e.g. inflation  no dynamics

17. EXTRA DATA Example I: PSM (UK) P FRS FRS 2001/02 2001/02 S (selected variables) M PSM INITIAL PREPROGRAMS PSM INITIAL PREPROGRAMS Read in Data Merge Outside Data EXTRA EXTRA P DATA DATA Derive new Variables e.g. Rent Officer statistics, Create Flags and Store FRS values R Calibration E PSM PSM P INTERIM INTERIM PREPROGRAM PREPROGRAM DATASET DATASET R PSM OUTPUT 2005/06 PSM UPRATING PREPROGRAMS PSM UPRATING PREPROGRAMS O Static Aging Tabulations Outputs G Uprate Benefit Values EXTRA DATA PSM YEAR SPECIFIC R Adjust Benefits Caseloads PSM OUTPUT 2006 / 07 Population Grossing A M S PSM INPUT PSM INPUT PSM Rules 2005/06 2004/05 2004/05 PSM INPUT PSM INPUT PSM OUTPUT 2007 / 08 2005/06 2005/06 SIMULATION = USER CHANGES RULES

18. Static MSM Assumptions • Key assumptions made: • No Tax Evasion  individuals report all their incomes • Full Benefit Take – up  individuals cash in all the benefits they are entitled to • Policy changes do not change individual propensity e.g. to evade tax or claim benefits

19. Example I: PSM Output

20. Example II: BRAHMS (Brazil)

21. Static MSM: Limitations • MSM outputs are as good as microdata  compare them with other analysis or admin data • Model benefit entitlement only, given rules  not actual take-up • Non-Behavioural: only first round policy effects are estimated, no responses • Non-Dynamic: no evolution in sampled population • Non-Longitudinal: no past information  time-series data required for pension modelling

22. Dynamic MSM • Use panel data = past information to simulate future life “histories”  Time + Behavioural Dynamics are introduced. • Dynamic Aging: • Individual transitions predicted over the life cycle  synthetic future panel data set from input data. • Transitions estimated under different policy scenarios  compare inter-temporal effects on life income path + behaviours • Better suited for Pension analysis

23. Example SESIM (Swe)Simulated Transitions LINDA data (Lennart Flood, Ministry of Finance and Gothenburg University)

24. Transitions • Individual Probability of a status change • Can be estimated by: Behavioural Methods = Behavioural Responses are included to generate new aggregate patterns Statistical Methods = Future individual events Made to reproduce existing aggregate patterns and behavioural preferences, through Stochastic methods Deterministic Methods (certain events)

25. From Transitions Probability … • Discrete Choice Estimation: Probability that a certain event, e.g. death, happens, at time t + 1 is a function of input variables at time t: Pr (si, t + 1 = 1) = f (xi, t , β| P ) Pr (si, t + 1 = 0) = 1 - f (xi, t, β| P )) • If no micro data available, conditional probability based on observed flows (Transition Matrices or Markov Chains): Pr (si, t + 1 = 1 | si, t ) = 0.02 if male, 0.01 if female

26. …To Event Simulation • From individual probabilities  must simulate for whom event actually happens  stock: • Random Simulation (Cramer, 1991) Pr (si, t + 1 = 1) = Pr (ui < pi ) • Alignment (O’ Donoghue) Pr (si, t + 1 = 1) = Pr (ui - pi < Z)

27. …To Behavioural Simulation • Behavioural responses to redistributive policy can be estimated • Stuctural Models: panel data used to infer ”preference functions” i.e. the behavioural rules (unobservable): F(y, β, P) • Estimated parameters (β) used to simulate (future) responses to policy reform: e.g. ∆(Labor Supply) = F(y, β, P) - F(y, β, Palt)

28. Dynamic MSM: Limitations • inputs • Limited Panel Data availability • Greater resources (e.g. IT infrastructure) needed to build + maintain model  high costs (ii) types of processes simulated • Sufficient knowledge of micro-behaviours? • Generally productive or financial sector not modelled (growth) • No Macro feedbacks (attempts ongoing) (iii) types of policies modelled • Not possible to include policies which depend on non-financial criteria But …..“MSM provides an organising framework” (Burtless,1996)

29. MSM & Pension Analysis 3 AIMS: • to simulate income distribution under pension system P (static MSM) • to simulate future public + private pension accumulation and de-cumulation over the life cycle, underpension system P, given demographic + behavioural changes (dynamic MSM) • to simulate effects of reforms to P on (life-cycle) incomes distribution + costs (both Static and Dynamic MSM).

30. 1. Static MSM and Pensions • Can be used to simulate short-term effects of policy (change) on current pensioners + current income distribution • Can be used to make comparisons between pension systems  pure redistributive impact of each system can be isolated • Cannot track pension entitlement accumulation through life cycle or simulate who will benefit from what pension in the future

31. 1. Example: Pensions Outcomes by Gender for UK, Fr, D

32. 2. Dynamic MSM and Pensions • Given P or Palt, DMSM needed to estimate future: • Pensions’ participation (for workers) • Pensions’ coverage (for retirees) • (Difference between) work and retirement incomes (RR) • (Difference in) n. of poor pensioners • (Difference between) incomes from each pension pillar • (Difference in) Pension Wealth (= lifetime return to system) • (Difference) in future income inequality between groups • Levels of Analysis: • Population (cross-sector differences between groups) • Intra-Cohort (life-time differences) • Multi-Cohort (birth-cohort differences  intergenerational equity)

33. 2. Pension Modelling in DMSM • For each individual in sample, Pension Module simulates over time: • Public Pension Coverage + Benefit Income • Occupational Pension Coverage + Benefit Income • Private Pension Coverage + Benefit Income • Retirement behaviour

34. 2. Statistical Modelling of Pensions • For working age individuals  must simulate pension coverage: • Is person accumulating State Pension rights ? • Is person member covered by an Occupational Pension plan? If so which are plan characteristics? • Is person saving in a Personal Pension ? How much? • For retirees  must simulate pension participation and retirement income: • Is retiree receiving what type pensions ? How much from each type? Replacement rate ? • For deceased  must simulate pension inheritance by spouse / dependents

35. 2. Behavioural Modelling of Retirement • For working age individual, what age to retire given financial incentives built in pension system? Ex. Option Value Model (Stock and Wise, 1990): in any given year, the individual will retire if the expected gain from retiring is greater than that of waiting, given pension system P • Useful to measure pension incentives for early retirement

36. 3. Why Pension Reform: Aging • Aging = lower fertility + mortality rates + higher life expectancy • Doubled Dependency Ratios expected by 2050 • More pressure for redistribution from shrinking active population to growing inactive population • Growing pressure on future performance of existing (public) pension systems: • Higher pension / health care costs ? • Higher taxes ? Lower Savings ? Lower Private Transfers ? • Lower Output Growth ? • Higher pensioner poverty ? • Higher inequality ?

37. 3. Aging in Latin AmericaSources: Calculations using the United Nations data base (1999).

38. 3. Life Expectancy at 60 Latin America

39. Future pension outcomes will depend on complex interactions between Demography + Labour + Macro + Pension System. Aging can affect Pension System outcomes but Pension Reform can curb the effects of aging e.g. by introducing incentives for later retirement What Pension System is better to cope with aging problem, for a given context? MSM can help to simulate these interactions under different assumptions  sensitivity analysis 3. MSM and Pension Reform

40. 3. MSM & Pension Reforms • Parametric Reforms: • Changes to retirement age, replacement ratio, contribution rate, indexing, or taxation of pensions • Systemic Reforms: • Changes to system structure or financing of the system • Moving from PAYG to Funding • Making benefit more actuarial (DC)

41. 3. Ex: From DB to DC Systems: Pension Outcome by Gender

42. 3. STATIC ex: EUROMOD (EU) • Common Pension Reform Package simulated for Dk, D, IT , UK (Working Paper EM5/05): • Lowering replacement rates of contributory earnings-related pensions between 5 – 10 %  lower cost • Introducing Minimum Pension = 40% average earnings  reduce poverty + inequality • Increasing Contribution rates between 1 -3%  revenue neutrality • Effects on current pensioners’ incomes, poverty, inequality ? How large are the differences ?

43. 3. STATIC ex: EUROMOD (EU) • 4 Systems’ main differences: • It / D: more earnings related pensions • Dk/UK: more flat pensions • Common Reform Package lowers poverty and inequality in all 4 countries, but... • Size of distributional impact and beneficiaries vary depending on initial conditions • Conclusion: different systems across EU must follow different pathways to reach common redistributive goal.

44. EUROMOD: Distributional Effects of Reform Package

45. EUROMOD: Distributional Effects of Reform Package

46. 3. Dynamic ex: SESIM (SW) • By 2030, number of +85 to double in Sweden • Public Pension Reform in 1999 • Notional DC PAYG system (16% of pension base)  income pension • Advanced-Funded DC System (2.5% of pension base)  premium pension • Fully covers only those born > 1953  different birth cohorts subject to different pension systems • Flood, L. (2003): Uses SESIM to simulate incomes for multiple cohorts for 1999 – 2041 : • Compare HH incomes before and after retirement (Replacement Rate) by birth Cohorts • Decompose income by pension pillars, by Cohorts • Evaluate importance of private (real + financial) wealth

47. SESIM Simulation • LINDA panel data used • Assumptions: retirement age is 65, 2% yearly inflation, 3% real growth, variable return on financial assets 3-7% • Results: The 1999 Reform is less generous. Younger cohorts (born after 1953) enjoy lower replacement rate, unless retirement age is delayed to 67, and returns on savings are high

48. SESIM Output

49. MSM for Developing Economies • Redistribution increasingly recognised as essential for economic growth and development • Static MSM well spread among developing countries • World Bank • UN Wider: Dart (Russia), 5 MSM for Africa online • NUIG (BRAHMS, XLSIM) • Dynamic MSM not spread among developing countries

50. MSM for Developing Economies • Challenges due to: • Weak Administrative Structures  gap between policy and reality ? • Private, in-kind redistribution might be larger than public transfers ? • Behavioral responses to policy changes might be different ? • MSM must be modified accordingly !