- Optimal policy, which, where, and why: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4276602
- Machine Learning Simulates Agent-Based Model Towards Policy: https://arxiv.org/abs/2203.02576
Policymakers' role in decision making on alternative policies is facing restricted budgets and an uncertain future. The need to decide on priorities and handle effects across policies has made their task even more difficult. For instance, housing policies involve heterogeneous characteristics of the properties themselves and the intricacy of housing markets within the spatial context of cities. Here, we have proposed PolicySpace2 (PS2) as an adapted and extended version of the open source PolicySpace agent-based model. PS2 is a computer simulation that relies on empirically detailed spatial data to model real estate, along with labor, credit, and goods and services markets. Interaction among workers, firms, a bank, households and municipalities follow the literature benchmarks by integrating economic, spatial and transport research. PS2 is applied here as a comparison of three competing public policies aimed at reducing inequality and alleviating poverty: (a) house acquisition by the government and distribution to lower income households, (b) rental vouchers and (c) monetary aid. Within the model context, monetary aid, that is smaller amounts of help for a larger number of households, improves the economy in terms of production, consumption, reduction of inequality and maintenance of financial duties. PS2 is also a framework that can be further adapted to a number of related research questions.
Available here: https://github.com/BAFurtado/PolicySpace and published as a book here: https://www.researchgate.net/publication/324991027_PolicySpace_agent-based_modeling
FURTADO, Bernardo Alves. PolicySpace: agent-based modeling. IPEA: Brasília, 2018.
This was an open agent-based model (ABM) with three markets and a tax scheme that empirically simulates 46 Brazilian metropolitan regions. Now, we have also added a credit market, a housing construction scheme and an incipient land market mechanism.
Bernardo Alves Furtado -- https://sites.google.com/view/bernardo-alves-furtado
Francis Tseng -- http://frnsys.com
Developed by Bernardo Alves Furtado, funded primarily by Institute of Applied Economic Research (IPEA) [www.ipea.gov.br] with one-period grant from [https://www.cepal.org/pt-br/sedes-e-escritorios/cepal-brasilia] (CEPAL-Brasília) and International Policy Centre (https://ipcig.org/). BAF acknowledges receiving a grant of productivity by National Council of Research (CNPq) [www.cnpq.br].
Furtado, B. A. (2019). Modeling tax distribution in metropolitan regions with PolicySpace. Journal on Policy and Complex Systems, 5(1). https://doi.org/10.18278/jpcs.5.1.6
new BOOK describing changes and processes currently on the make
- PolicySpace2 generates data that can be applied at https://github.com/frnsys/transit_demand_model. The transit model in turn generates private and public transport routes, visualization and congestion times.
- Now data is read at the intraurban level of 'áreas de ponderação' (census track/block) from IBGE.
- There is exogenous migration embedded in the model (creating new houses)
- Exogenous growth of firms
- Marriage
- Better support for plot formats
- House prices (supply) are included into families' wealth and updated every month
- Consumption decisions now follow Bielefeld, 2018 and is now a "linear function of current and expected future incomes and of financial wealth." In other words, "all income in excess of permanent income will be saved and added to financial wealth."
- Families composition are now sure to have at least one adult. Children are initially distributed randomly.
- Houses of families whose last member dies are randomly allocated to remaining family members.
- Interest being now paid on families' savings
- When no cash available, families can withdraw from savings for consumption (if any at savings) up to permanent income
- Internal clock updated to use datetime
- Introduced a rental market
- Included construction companies endogenous to the model.
- Financial (mortgage, credit) market structures
- Included transportation costs on workers' decision on choosing jobs
- Introduce inheritance (savings).
- A neighborhood effect (wealth of resident families influence price of houses)
- Decay effect on prices due to excess of houses on offer
- Data updated (from census 2000 to census 2010)
- Offer decay (time depreciation) implemented
- Global availability of houses (vacancy) influences prices
- Neighborhood effects (based on families' wealth) influence prices.
- Possibility of negotiating house prices below the offer
- Vacancy size (offer) now impacts willingness to build of construction firms and prices
- New output analysis
- Endogenous mortgage rate setting by the bank. Default uses exogenous data though.
- Included the SAC -- Constant Amortization System -- mostly used in Brazil.
- POLICY EXPERIMENTAL DESIGN. Testing wether giving houses, paying rent or distribution money is the best policy alternative. Both families in poverty and resources invested are endogenous.
- Included varying oficial monthly interest rate.
We recommend using conda and creating an environment that includes all libraries simultaneously.
First create an environment and install Shapely and GDAL as such:
conda create --name ps2 python=3.6
Activate the environmnet
conda activate ps2
Then add Shapely from conda-forge channel
conda install shapely gdal -c conda-forge
Then the other packages
conda install fiona pandas geopandas numba descartes scipy seaborn pyproj matplotlib six cycler statsmodels joblib scikit-learn flask flask-wtf psutil -c conda-forge
To locally configure the simulation's parameters, create the following files as needed:
conf/run.pyfor run-specific options, e.g.OUTPUT_PATHfor where sim results are savedconf/params.pyfor simulation parameters, e.g.LABOR_MARKET.
The default options are in conf/default/, refer to those for what values can be set.
These optional arguments are available for all the run commands:
-nor--runsto specify how many times to run the simulation.-cor--cpusto specify number of CPUs to use when running multiple times. Use-1for all cores (default).
python main.py run
Example:
python main.py -c 2 -n 10 run
Runs simulation over a range of values for a specific parameter. For continuous parameters, the syntax is
NAME:MIN:MAX:NUMBER_STEPS. For boolean parameters, just provide the parameter name.
It now also accepts selected "PROCESSING_ACPS-BRASILIA-CAMPINAS-FORTALEZA-BELO HORIZONTE"
Example:
python main.py sensitivity ALPHA:0:1:7
Will run the simulation once for each value ALPHA=0, ALPHA=0.17, ALPHA=0.33, ... ALPHA=1.
Example:
python main.py sensitivity WAGE_IGNORE_UNEMPLOYMENT
Will run the simulation once for each value WAGE_IGNORE_UNEMPLOYMENT=True & WAGE_IGNORE_UNEMPLOYMENT=False.
You can also set up multiple sensitivity runs at once.
For example:
python main.py sensitivity MARKUP:.05:.15:7 WAGE_IGNORE_UNEMPLOYMENT
is equivalent to running the previous two examples in sequence.
For multiple combinations of parameters one may try the following rules
Include first the params, separated by '+', then '' and then the list of values also '+' Such as 'param1+param21+2*10+20'. Thus, producing the dict: {'param1': ['10', '20'], 'param2': ['10', '20']}
python main.py sensitivity PRODUCTIVITY_EXPONENT+PRODUCTIVITY_MAGNITUDE_DIVISOR*.3+.4*10+20
Runs simulation over a different distribution combinations: ALTERNATIVE0: True/False, FPM_DISTRIBUTION: True/False.
Example:
python main.py -n 2 -c 2 distributions
Runs simulation over a different ACPs.
Example:
python main.py -n 2 -c 2 acps
You can regenerate plots for a set of runs by using:
python main.py make_plots /path/to/output
In Windows, make sure to use double quotes " " and backward slashes as in:
python main.py make_plots
"..\run__2017-11-01T11_59_59.240250_bh"
There is a preliminary web interface in development.
To run the server:
python main.py web
Then open `localhost:5000` in your browser.