20世纪90年代时间序列预测领域主要研究动态

Main Research Dynamics in Time Series Forecasting in 1990s

2 0世纪 90年代 ,预测领域取得了比较丰硕的研究成果。预测方法除主观判断方法外 ,主要有单变量方法和多变量方法。单变量方法在实际中使用最多 ,主要涉及分数差分模型、结构模型、贝叶斯预测方法。多元回归方法仍是最常用的多变量预测方法 ,但对经济时间序列拟合多元回归模型存在一些问题 ,于是人们对向量回归模型进行了大量的研究。本文着重分析了国外学者关于预测方法的选择以及非线性模型的研究动态。

There were a great of achievements in time series forecasting in 1990s. Forecasting methods mainly include univariate methods and multivariate methods except the judgmental methods. Univariate forecasting methods are still used far more in practice, such as Fractional Differencing models, Structural models and Bayesian Forecasting methods. Multiple regression is still the most commonly used multivariate forecasting methods but there can be problems in fitting such models to economic time series. So there has been much work on vector regression models. This paper will discuss the foreign researchers' achievements in the choice of forecasting methods and their research dynamics in non-linear models.