基于Lasso-MIDAS模型的混频时间序列预测研究

针对传统的时间序列预测方法在处理复杂丰富的大数据时常面临变量间抽样频率不同、数据相关性复杂等问题，基于Lasso算法和混频数据抽样模型(MIDAS)提出了不改变数据结构的混频时序预测模型Lasso-MIDAS。该模型通过融合MIDAS处理混频信息的机制和Lasso算法的压缩特性来实现估计预测，实时修正对预测最有效的混频变量集；根据常见的正则化方法岭回归设计了Ridge-MIDAS模型用做对比。实验结果表明，Lasso-MIDAS在预测性能上优于标准MIDAS模型及对比模型，验证了该方法在混频时间序列预测方面的有效性。

Research on mixed-frequency time series prediction based on Lasso-MIDAS model

Traditional time series prediction methods often confront problems such as different sampling frequency between variables and complex data correlation when dealing with complex and rich big data. Based on Lasso algorithm and Mixed Data Sampling(MIDAS) method, this paper proposed a mixed-frequency time series prediction model Lasso-MIDAS without changing the data structure. It combined the mixed-frequency information processing mechanism of MIDAS and the compression characteristics of Lasso algorithm to achieve the estimation and prediction, and to modify the most effective set of mixed-frequency variables in real time. According to the common regularization method Ridge regression, this paper also designed a mixed-frequency data model Ridge-MIDAS for comparison. Experimental results show that Lasso-MIDAS is superior to the standard MIDAS model and comparison models in terms of prediction performance, which verifies the effectiveness of the this method in mixed-frequency time series prediction.