Trust and Social Control: Sources of cooperation, performance, and stability in informal value transfer systems

Claudius Gräbner, Wolfram Elsner and Alex Lascaus

This repository contains the source code for the model used in the abovementioned paper. The paper is published in the journal Computational Economics and is available here (open access, main paper and supplementary material).

Folder structure of the repository

bash: contains bash scripts to run the simulation with particularly relevant parameter combinations (not necessary any more)

python: contains all Python scripts and, in python/parameters/ the json files containing relevant parameter specifications.

notebooks: contains Python notebooks, including the one used to create the figures for the final paper

tex: contains TeX code used to create some of the figures in the final paper

output: contains all the output produced by the source code

Main elements of the model

The general functioning of the model is explained in the main paper. Moreover, the code is well documented and should be largely self-explanatory. The general structure and the main elements of the model are as follows:

• The parameter files saved in python/parameters/. These are json files specifying all relevant model parameters. If a parameter value is given as a list and the model is called via Main, then the model is run for all possible parameter constellations for those specified in the lists.
  • For example, if the parameter C_trust is specified as [0, 1] and C_control_1 as [0, 1], then calling this parameter file means to runs one model instance with C_trust=1 and C_control_1=1, one with C_trust=0 and C_control_1=1, one with C_trust=0 and C_control_1=1, and one with C_trust=0 and C_control_1=0.
• The main model is implemented via the class Model, which is defined in model.py. Some parts were sourced out in separate files:
  • During the initialization of a Model, an initial agent population must be set up. This is done by the class PopulationGenerator, which is defined in population_generator.py.
  • The class Agent is defined in agent.py.
  • The strategies of the agents are defined via their own class Strategy. This class is defined in strategies.py.
• The class Main, which is defined in main.py reads a parameter file, creates many instances of Model, runs these models and summarizes their results in a single data frame.
• The function conduct_meta_mcs() as defined in MCS.py facilitates Monte Carlo simulations. When it is called one has to supply the skeleton of the names of parameter files. Then main.py is executed for all parameter files matching this name and the results are aggregated into a single data frame that can then be further processed.

Thus, the recommended way to use the model is via MCS.py. To run the model for all parameter constellations that are specified in the parameter files starting with 'hawala_shocks' we can use the following call:

python python/MCS.py python/parameters/hawala_shocks 10

This runs the model 10 times for each of the following parameter files: hawala_shocks_c.json, hawala_shocks_t.json and hawala_shocks_tc.json, and aggregates the results into a single data frame.

Using the model

There are three ways to run the model:

1. Conduct a single model run using model.py directly. This only works for a single parameter constellation and is not recommended.

2. Run the same model for many times using main.py. This allows to sweep through different parameter constellations if the different parameter values are specified as a list in the json file.

3. Run the model for several parameter constellation using MCS.py. This is the recommended way to use the model and effectively calls main.py several times and can process more than one parameter file.

Replicating the figures of the main paper

To replicate the figures in the main paper you must first create the data by runing the following simulations.

For figure 5:

python python/MCS.py python/parameters/hawala_baseline 30

For figures 6 and 7:

python python/MCS.py python/parameters/hawala_shocks 30

For figure 8:

python python/MCS.py python/parameters/hawala_framework_nbhawaladars 30

and

python python/MCS.py python/parameters/hawala_framework_intdensity 30

For figure 9:

python python/MCS.py python/parameters/hawala_framework_forgiveness 30

For figure 10:

python python/MCS.py python/parameters/hawala_framework_pop_growth 30

For figure 11:

python python/MCS.py python/parameters/hawala_framework_mistakes 30

Then you can use the Python Notebook in notebooks/Create-Figures.ipynb to create all the figures. They are automatically saved to output/figures.

Comments

In case you have questions, feedback or other comments feel free to contact Claudius here.