An agent-based approach to supply side modeling of agricultural and power systems

Doktorsavhandling, 2017

This thesis deals with the modeling of economic systems in the context of agricultural and power systems, and some aspects of the difference between the standard economics equilibrium approach and the agent-based approach. We model the supply side, where agents make decisions on what to produce or in what to invest. These decisions are based on predictions of future prices and other market conditions. In all settings time lags and limited foresight are important. Whereas standard economics is based on the idea of economic equilibrium, agent-based modeling describes dynamic systems based on the interaction of agents who do not necessarily possess perfect information and rationality. This thesis consists of three parts. In papers I-III we present a model of interacting markets with cobweb characteristics, i.e. markets where prices are prone to oscillations due to a time lag between supply and demand decisions. We apply the model to land-use competition between food and bioenergy crops. We show how instability in one agricultural market, e.g. the bioenergy crops market, can be transferred to other agricultural markets, both on the supply side (by the limited availability of land) and on the demand side (by consumers choosing between different goods). Under certain circumstances the agent-based dynamics can be projected to a closed dynamics of aggregate quantities, which allows for the stability characteristics to be analytically approached. In paper IV we present a model of beef cattle dynamics based on decisions taken by boundedly rational farmers. We systematically examine the parameters determining the agents' expectations and decision mechanisms, and their impacts on the dynamics. In paper V we study a power system transition triggered by a carbon tax. We find that the level of carbon tax needed to reach a specific CO$_2$ mitigation target may be significantly higher in an agent-based model than in the corresponding optimization model. In all papers we focus on mechanisms and model characteristics rather than on predictions.