A Recursive Kalman Filter Forecasting Approach
研究时变系数时间序列模型的预测精度和成本效益，通过模拟实验提出一种卡尔曼滤波自适应估计方法，发现时变系数模型在适当时优于常系数模型，并给出一个简单决策规则以提高计算效率。
This paper examines the forecasting accuracy and the cost effectiveness of time series models with time-varying coefficients. A simulation study investigates the potential forecasting benefits of a proposed Kalman filter type adaptive estimation and forecasting approach. It is found that: When appropriate, the time-varying coefficient approach leads to better forecasts than the constant coefficient procedures. A simple decision rule, which indicates whether time-varying coefficient models are in fact needed, increases the computational efficiency.