| Micro Finance Description |
 |
 |
 |
|
Sensitivity Scale of credit cycles does not seem to influence results, apart from the number of events registered. Groups most likely to succeed present similar patterns, even in under different configurations, of losses being covered whilst those most likely to fail register more expelling votes. Groups terminating credit cycles with equal numbers of desirable and undesirable events have yet to be analysed thoroughly to allow better interpretations of their peculiarities.
DOCUMENTATION 1. Purpose This model incorporates the social and financial findings derived from a fieldwork conducted in collaboration with a microfinance institution in Mexico on the group-level mechanisms for dealing with defaults within finance groups. The agent internal state, MeetingTrack, is described in Table 1 and also includes following attributes: SpokenLanguages, BusinessActivity, TotalDebt, Quota, Location and Tolerance.
Table 1: Internal structure of the MeetingTrack in all individual agent memories
The configurable circumstances of a microfinance group are described in Table 2 below.
Table 2: Negative circumstances surrounding a simulated group
The financial parameters of a microfinance group are described in Table 3.
Table 3: Financial parameters
3. Process overview and scheduling The order in which the model is initialised is depicted in the Figure 1 below, starting with basic configurations and instantiation of agents along with their individual properties.
 4. Design concepts 4.1 Emergence Groups classified as mostly likely to succeed (or fail) can have different patterns of registered events that help justifying and explaining why the configuration leads to it. 4.2 Adaptation Individual agents can have different tolerances to each group member and, although these do not change at runtime, the model can be easily changed to allow that. 4.3 Fitness There is no fitness as in the Genetic Algorithms sense, only positive and negative endorsements between agents being registered in their memories, depending on the circumstances in which each one encounter itself in at that moment in the simulation. 4.4 Prediction There is no prediction ability from agents in the model. 4.5 Sensing Agents can sense whether another one has defaulted, their language / business and if is affected by some exogenous problem (illness, unprofitability or poor investments). 4.6 Interaction Interaction takes place in the form of meetings where agents can use their sensing abilities and which endorsements have been registered individually about each group member in past meetings. 4.7 Stochasticity Employed to choose who is affected by the configured exogenous problems (illness, unprofitability or poor investments), the order in which these are processed and when problematic agents will default or miss meetings as there is no evidence to backup such orders. 4.8 Collectives Yes, sub-groups can be done based on business activities, languages and locations. 4.9 Observation All agent decisions are logged and a summary of all these are included at the end. 5. Initialization All is defined as described in section. An extension of the simulation is being tested to read up-to-date data directly from the microfinance institution databases as input. 6. Input All the possible input parameters are described in Sections 2 and 5 of this file. 7. Submodels This is described in Section 3, yet a formalised Markov Chain version of these is being developed in collaboration with the ETH Chair of Modelling and Simulation team.
|