Agent-Based Models, Macroeconomic Scaling Laws and Sentiment Dynamics
The purpose of this thesis is to show that a certain kind of agent-based model can generate a series of social phenomena, and many of them may be quite close to reality. Instead of emphasizing the microscopic aspect of "utility maximization", we emphasize more that of a "statistical equilibrium", a stage that economy will generally fluctuate around. Through agent-based simulations, some seemingly puzzling stylized facts in macroeconomics can be replicated without too much effort. The first part of the thesis has introduced an agent-based model describing the social production process, which consists of employment, production and consumption. The biggest advantage of this model is that: it generates many macroscopic regularities from microscopic behavior within a single framework, involving intricate interactions among simple micro-economic entities. In particular, the model is capable of explaining many stylized facts, including the frictional unemployment, the firm size distribution, growth distribution, wealth distribution, and etc. The secon part of the thesis dig into the agent-based models of opinion formation in order to find out whether the subjective opinion dynamics may, to some extent, influence objective economic variables. Particularly, we will investigate how consumer sentiment or industrial sentiment can help to explain or predict the movement of industrial production. We first offer a forecast-oriented VAR analysis, in which different kinds of VAR models with various lags will be estimated and evaluated with respect to their overall forecast abilities for industrial production. In the next step, taking nonlinearity and interactions among sentiment and industrial production into account, we set up an agent-based model of opinion formation following the pioneering work of Weidlich and Haag (1983) to model the joint dynamics of sentiment and output. A nonlinear Fokker-Planck equation is added to help in explaining the transient density of this stochastic model. Then the model is further extended into two dimensions. We observe a strong causal relation from industrial sentiment to the output, which confirms our finding in the VAR analysis.