The impact of expectations in an agent-based model

Within the context of an agent-based model, model selection by the economic agents is introduced and investigated. To achieve this, a specific agent, the “economic research institute†, is set up and produces regular forecasts of the economy which are published to the economic agents. The forecasts are based upon linear regressions updated each period. Expectation errors may arise and agents may chose to change their behaviour based upon these forecasts. Moreover, agents might change their interpretation of the forecast data and decide upon using this data or not. Alternatively, a simple VAR model of the economy is implemented and provided by a different research institute. Model selection is implemented by a genetic algorithm, and each agent uses a linear function to produce its individual forecasts. Thus, each agent has its own “private†model of the economy, based upon the two alternative forecasts of the research institutes. Following this "competition" of the research institutes and the build-up of a large number of different meta-models, several aspects of different expectation building processes are investigated: e.g. information diffusion processes, model selection of smart agents based upon a GA, the impact of different forecast mechanisms on the economic outcome