This work focuses on the investigation of economic preferences, particularly within economic systems, using agent-based models (ABMs). Economic systems are a challenging area of research, exhibiting complex adaptive, non-linear, recursive properties. ABMs are now relatively widely used in economics as experimental tools across many economic schools. While these schools are presented as distinct theoretical frameworks, typical experimental approaches are strikingly similar: quantitatively rigorous, highly abstracted models of core preferences and economic behaviours are typically bundled into centralised structures and other behaviours ignored. Vast amounts of data are potentially available from computational models making validation & verification of experimental results problematic. Researchers have relied on replication of time series artefacts yielding little explanatory value, rather than examining agent behaviours themselves.

To date relatively little effort has been made to codify, or critically discuss experimental protocols around these models; to verify and replicate findings; or to develop realistic validation metrics. This is a failing in the literature limiting the value of ABM approaches and leaving them open to criticism. This thesis directly addresses these issues from a practical standpoint, presenting discussion of what approaches are appropriate and what features robust methodologies should exhibit. Situation is presented as a basic validation metric, using agent-level performance measures and exploratory tools for verification and validation. Risk-adjusted, agent-level population relative performance measures are developed and tested on a typical agent-based artificial financial market (AFM) demonstrating their effectiveness, while highlighting potential methodological issues and confirming the importance of agent-level exploration alongside other validation and verification methods.

The case for an overall modelling framework based around a subsumptive meta-heuristic, population-based architecture is presented and discussed. In this framework, the SHaaP architecture, core preferences, preference modifying behaviours, and structural modifiers are unbundled so that they can be investigated systematically, while potentially being subsumed into larger, sophisticated non-linear preference structures. A functional SHaaP architecture is developed & implemented as part of the research. A larger case study examining the role of heuristic preference modifying behaviours in economic agent behaviours is presented, also demonstrating the architecture in operation and exploring its dynamics.

Heuristic preference modifiers are widely observed in real economic entities and systems: they have a potentially critical role in regulation & risk management under uncertainty but remain relatively unexplored. The case study demonstrates the potential of the SHaaP architecture in progressively developing and testing both particular behaviours and performance measures. Analysis of the dynamics of preference modifying heuristics in a population-based structure illustrates the complexity of such systems, serving to identify areas for future research, particularly for regulatory policy design. Potential deficiencies in the relative performance measures were highlighted leading to proposals for further development and testing. Finally the role of the subsumptive architecture as a key, complementary component to traditional agent-based experimental economics models is discussed, particularly as an exploratory tool in developing & testing dynamic systems for regulatory frameworks subject to uncertainty.