Using Differential Equation-Based Models to Calibrate Agent-Based Diffusion Models
提出一个两阶段校准框架，先用微分方程模型为基于主体的扩散模型提供初始搜索点，再通过局部搜索找到最优参数，案例显示只需评估少量点即可得到最优解，且部分主体模型解释和预测能力优于微分方程模型。
Agent-based diffusion models (ABMs) have been increasingly used for making decisions in today's complex and dynamic managerial environment. To be a reliable managerial decision tool, they should be rigorously calibrated against empirical data sets to capture key features of real-world scenarios. However, since that the implementation of ABMs is time-consuming, their efficient calibration is still an open challenge. In this article, we present a calibration framework for a parsimonious ABM by using a simple assistant model (AM) to provide an initial searching point. With a differential equation-based diffusion model (DEM) as the AM, we construct a two-stage calibration procedure for the ABM, in which the first stage is to build an explicit connection between parameter spaces of the ABM and the DEM and obtain an initial searching point, and the second stage is to search for the optimal estimates via an iterative local search method. The case study demonstrates that the proposed framework can identify an optimal solution by evaluating only a few points. It also reveals that some of the ABMs have better explanatory and forecasting performance than the DEM.