基于多源异步混频CPI数据的预测方法研究

Research on CPI Prediction based on Multi-source Asynchronous Mixed Sampling Data

研究目标:基于线上消费者价格指数和网络搜索价格指数预测CPI。研究方法:在卷积神经网络(CNN)框架中融合MADL_MIDAS模型,建立异步混频卷积神经网络(AMCNN)模型,并通过选用2016年1月至2019年12月的数据验证该方法的有效性。研究发现:日度线上CPI及日度网络搜索指数属于CPI的领先指标,同时引入并保留原有数据特征有助于改进CPI预测精度,提高CPI“拐点”捕捉能力。研究创新:揭示了高频日度线上CPI和网络搜索数据对低频月度CPI的预测能力,提出了一种融合神经网络与传统计量模型的异步混频数据处理方法。研究价值:预测CPI波动水平和“拐点”时,可辅助利用线上CPI、网络搜索高频数据,结合AMCNN模型提高预测精度。AMCNN模型可用于处理异步混频数据、探究变量间复杂不确定性(线性、非线性)关系,具有很强的适应性和扩展性,可应用于其他经济、金融领域,应用价值较高。

Research Objectives:CPI is predicted based on internet consumer price index(ICPI)and internet search index(ISI).Research Methods:In the framework of convolution neural networks(CNN),MADL_MIDAS model is integrated,and the asynchronous mixed frequency convolution neural network(AMCNN)model is established.The data from January 2016 to December 2019 are used to verify the effectiveness of the method.Research Findings:High frequency daily data ICPI and ISI are the leading indicators of CPI.Adding all variables and retaining the original data characteristics will help to improve the CPI prediction accuracy and the ability of capture CPI s“change point”.Research Innovations:This paper reveals the prediction ability of low frequency monthly CPI by ICPI and ISI of high frequency daily data,and proposes an asynchronous mixed sampling data processing method combining neural network and traditional econometric model.Research Value:When forecasting CPI fluctuation level and“change point”,it can help to use ICPI and ISI,combined with AMCNN model to improve prediction performance.AMCNN model can be used to deal with asynchronous mixed sampling data and explore the complex uncertainty(linear or nonlinear)relationship between multi-variables.It has strong adaptability,scalability and high applicability.It can be applied to other economic and financial fields.