Agent-based model of income selection/migration
Initial set up of project June 21 2016, Andrew Bell
This setup is built in Matlab 2014a.
As built, the model generates a random landscape of hierarchical administrative units (states within countries, districts within states, etc.) and a random network of agents connected with each other, living in these administrative units.
Agents earn income from layers at fixed intervals which they share with those in their personal networks. In between these intervals, agents may participate in social interactions with other agents, sharing information about income opportunities, and they may make choose to change their set of income sources - either by choosing new layers in the same place, or moving to a new place.
Agents compare different potential portfolios of income layers by constructing future income streams, converting to expected utility, and discounting those streams back to a net present value. wherever possible, income streams preserve patterns along time cycles (i.e., seasons) and preserve correlations in time across income layers, thus allowing patterns such as seasonal migration and income diversification to (in theory, with the right parameters) emerge in the simulation.
This model fits the 'push-pull-mooring' model of migration as follows:
i. Push: Layers accessed by agent in current location shift to lower utility (e.g., economic downturn, climate shift, etc.).
ii. Pull: Layers known by agent in different location offer better utility; benefits from sharing over social network in different location offer better utility than current location.
iii. Mooring: Layers accessed by agent in current location maintain utility (e.g., assets, secure employment, etc.); benefits from sharing over social network in current location are preferable to expected benefits in other locations
Future versions of the code will include a batch-run script. At present, a single simulation run can be made from runMigrationModel.m. Current configurations include a few layers whose income potential increases down and to the right, leading to a mass migration toward the bottom right corner along the course of the simulation.
Input data to develop different scenarios can be added to the following routines:
-
setupLandscape.m: all agent preference parameters, random network parameters, landscape size and timestep parameters, as well as shapefiles for specific landscapes
-
createUtilityLayers.m: all specifications for utility layers, including the income generation function, time constraint, and access restrictions
-
assignInitialLayers.m: assigns the initial distribution of income layer use/access ... could be informed by census data
-
createMovingCosts.m: specifies the cost of moving from one place to another in the map
-
createRemittanceCosts.m: specifies the cost of sharing income with others in the social network, across space in the map