Information Technology Reference
In-Depth Information
expected utility of each investment option x. The expected subjective overall utility of
investment option x is calculated by adding up the criteria values for a given x
weighted by the respective preference value. The presented utility approach allows for
the representation of agent heterogeneity in various aspects: Firstly, different prefer-
ence sets may be defined representing agent types that differ in their basic orientations
(understood as persistent personality traits). Secondly, agents are heterogeneously
embedded in their local neighbourhood group and include the observed behaviours of
other group members in their subjective utility estimation. Thirdly, subjective utility
reflects the embedding of an agent in its network of social peers and opinion dynam-
ics based on observed behaviours.
In HAPPenInGS-A the final selection of an investment option is represented by a
probabilistic choice model (see e.g. [19]) based on the expected overall utilities of
each of the behavioural options. This decision process is triggered if there is at least
one behavioural option with higher expected utility than the utility achieved by the
agent in the previous time step. Else the previous investment decision is kept. Explo-
ration is triggered with a probability of 1% and modelled by having agents select a
random investment level (uniformly distributed).
Simulations are initialised as follows: The environmental context consists of 20x20
equally sized, square patches. During initialisation 20 patches are randomly selected
and populated by agents. On each selected patch we initialise 20 agents forming a
neighbourhood agent group. Agent types are defined by preference profiles according
to HAPPenInGS which are represented in the agents' persistent state variables. For
heterogeneous populations the relative ratio between agent types is preserved on the
level of neighbourhood groups. For all agents the investment state variable is initial-
ised to 0.0. After setting up the agents in the environmental context, the social net-
work is initialised. We use a stylised version of the network initialisation process
proposed in 20: To account for baseline homophily we link each agent to 5 randomly
selected agents from its neighbourhood group. To account for inbreeding homophily
each agent is in addition linked to 5 agents of the same agent type selected at random
from the full population. Agent locations and the social network remain fixed
throughout the simulation.
4
Dynamical Analysis
The main goal of this section is to investigate the macro-level patterns generated by
HAPPenInGS-A. Such patterns constitute and describe the collective as well as the
temporal-dynamic implications of the HAPPenInGS theory.
4.1
Basic Agent Types
We first summarise previously reported results [1, 21] which investigate the influence
of agents' social orientation (i.e. the weighting between
selfPreference
and
other-
sPreference
) on agent behaviours and on the success of the collective action in terms
of a sensitivity analysis of the respective preference parameters of HAPPenInGS-A.
For this sensitivity analysis the preference for social conformity was set to 0 (
social-
ConformityPreference
=0.0
)
, i.e. agents disregard social influences. A particular
