摘要
研究目标:解决响应变量与解释变量存在混频观测时的多分类分析问题,扩展多项选择Logit(Multinomial Logit, Mlogit)模型到混频数据环境下,使之能够处理混频数据并实现多分类预测。研究方法:将无约束混频数据抽样(U-MIDAS)技术引入Mlogit模型中,构建U-MIDAS-Mlogit模型,给出其极大似然估计。研究发现:新建的U-MIDAS-Mlogit模型,不但能够高频预测响应变量的多分类结果,而且具有比传统Mlogit模型更高的分类精度。研究创新:提出了一个新的U-MIDAS-Mlogit模型,能够直接对原始混频数据进行建模,克服了传统Mlogit模型需要进行数据同频化处理的局限,提升了Mlogit模型的功能,提高了多分类预测效果。研究价值:U-MIDAS-Mlogit模型具有广阔的应用前景,能够解决一类混频数据环境下的多分类分析问题,本文从数值模拟与模型应用两个层面进行了证实。
Research Objectives: To solve the problem of multi-classification analysis on the case where the response variable and covariates are not observed at the same frequency, we extend the Multinomial Logit(Mlogit) model to the mixed frequency data. It enables to deal with mixed frequency data and predict multi-classification. Research Methods: We introduce the unrestricted mixed data sampling(U-MIDAS) technology into the Mlogit model to construct a new U-MIDAS-Mlogit model, and derive its maximum likelihood estimation. Research Findings: The proposed U-MIDAS-Mlogit model can not only timely predict the multi-classification of the response variable, but also obtain higher classification accuracy than the traditional Mlogit model. Research Innovations: The novel U-MIDAS-Mlogit model is developed to handle the raw mixed frequency data directly, which overcomes the limitation of the traditional Mlogit model with the common frequency data. It also improves the performance of Mlogit model in terms of multi-classification prediction. Research Value: The U-MIDAS-Mlogit model is promising in solving the multi-classification problem on a kind of mixed frequency data, which has a broad application prospect. This has been demonstrated from both numerical simulations and a real-world application.
作者
蒋翠侠
赵婷婷
许启发
Jiang Cuixia;Zhao Tingting;Xu Qifa(School of Management,Hefei University of Technology;Key Laboratory of Process Optimization and Intelligent Decision-making,Ministry of Education)
出处
《数量经济技术经济研究》
CSSCI
CSCD
北大核心
2022年第4期168-188,共21页
Journal of Quantitative & Technological Economics
基金
教育部人文社会科学研究规划基金项目(19YJA790035)
全国统计科学研究重大项目(2019LD05)
国家自然科学基金面上项目(71671056,72171070)的资助。
