Since 2004, I have been interested in microsimulation modelling applied to social and economic policy making, in particular to the redistributive effects of tax and benefit systems.

Microsimulation models are statistical tools developed by governments or research institutes. They can:

(i) help to forecast and simulate the effects of an existing or proposed fiscal or welfare policy change on future costs, poverty and inequality levels, as well as compare them under alternative policy scenarios

(ii) identify – when based on micro data – who gains and who loses by any given policy regime, as well as by how much, thus providing a valuable insight into the political as well as the economic sustainability of a policy change.

Microsimulation modelling was being majorly relied upon by the UK Department for Work and Pensions (DWP), where I worked as team leader of the Policy Simulation Model (PSM), as part of their commitment to evidence-based policy making, in the area of pensions and other welfare policies.

Static Microsimulation Models
The PSM for instance is a static microsimulation tax and benefit model which models the entire UK tax and benefit system rules and thus calculates benefit eligibility and amounts, as well as final disposable incomes, for a representative sample of the population (taken from the Family Resource Survey). This sample is then statistically grossed up to represent the entire UK population in a given year.

The model is so-called static since it assumes no population or other economic changes to affect its input sample in the future. It only limits itself to calculate so called ‘first round’ effects, i.e. those financial effects on household incomes from applying 100% the rules of the system, everything else remaining the same (including the lack of behavioural reactions by individuals or households to the system). For this reason, static tax and benefit models such as PSM are not really suitable for long term projections.

Indeed one can build behavioural additions to static microsimulation models, e.g. models which can predict how the labour supply behaviour of individuals will change in response to a change to work or non-work related financial incentives.

My first job at DWP was to build a labour supply model for the PSM, which the DWP would use to estimate the likely impact and directions of some of the welfare policy changes being considered (e.g. work hours-related tax credits). The LSM simulation model, built on SAS, implemented a theoretical model developed at the Institute for Fiscal Studies. You can download a presentation of the model here.

Dynamic Microsimulation Models
While working at DWP, I also started a collaboration with Dr. Cathal O’ Donoghue, an expert in dynamic microsimulation modelling, who is currently my PhD supervisor.

Dynamic microsimulation tries to surpass the limitations of a static model by using micro-econometric techniques to simulate veritable life histories 50 or 100 years into the future, for every single individual present in the original micro data. This involves modelling as many life events (e.g. births, marriages, educational choices, deaths, labor market choices) as our initial data allow us to predict (given longitudinal information present in the sample).

With Cathal and the rest of his team I am currently developing a dynamic microsimulation for Ireland called LIAM. In particular, I am busy modelling a pension module which allows to estimate both public, occupational and pension coverage and eligibility for Ireland up to 2050.

A main output from developing this pension model within LIAM is going to be the production of a report on old age poverty in Ireland under different pension reforms, commissioned by Combat Poverty Ireland.

Agent-Based Models
Agent Based Models are computer-generated simulation models to study complex systems whose macro properties arise from decentralised interactions between micro 'agents' (be they individuals, households, firms, government or other institutions).

Agent-based models have been increasingly used to model so called micro-macro linkages between individual agents and aggregate level variables, in alternative to e.g. traditional CGE, Overlapping Generation Models, or even game theoretical strategic models, which assume conditions of optimality, rationality and equilibrium, and can become unsolvable as soon as multiple agents are introduced.

Tesfatsion sums up agent-based models as models characterised by the following properties:

(i) the system modelled is composed of interacting agents

(ii) the system exhibits emerging properties from these interactions which cannot be inferred a priori

(iii) agents constantly evolve and learn from their interaction with others and the surrounding environment.