Abstract for:Analysis of dynamic models by optimization
A virtue of simulation models is that one can postulate and analyze current and future policies in light of cause and effect relationships. The analysis should provide correct explanations of the behavior of the system. However, when models are dynamic and nonlinear, explaining without tools gets complicated. Several formal methods have been developed, e.g. eigenvalue analysis and pathway participation metrics. Such loop analysis methods are efficient and powerful in understanding how model structure at each point in time explains basic dynamic modes of behavior: exponential growth, exponential decay, and oscillations. However, understanding the dynamic dominance may not answers all questions of importance for decision-makers. There is need for a tool that is capable of pinpointing elements of model structure that influence phenomena of interest. We present an optimization method, which helps identify the model structures of greatest importance for phenomena. We use a simple second order model to compare the results of optimization to the results of loop analysis methods. The example shows that optimization can give different explanations of observed phenomena and point to different places for policy intervention. More examples will appear in the final version of the paper.