如何测度央行行长的口头沟通信息——一种基于监督学习的文本分析方法

How to Measure the Information of Central Bank Governor’s Verbal Communication: A Textual Analysis Method Based on Supervised Learning

本文以2003年1月至2018年8月中国央行行长所有口头沟通内容为文本基础,生成央行行长沟通这一特定领域的专用词典,进而使用短语数量加权的方法分别构造货币政策沟通指数和经济形势沟通指数。其中,货币政策沟通指数与实际基准利率和存款准备金率的变动具有高度相关性,而经济形势沟通指数可以作为经济基本面的信号器。进一步,本文基于监督学习方法,通过训练子样本词典得到具有倾向的短语及其概率分布,利用文本分类器对新的沟通文本进行自动分类,最终对新样本进行指数计算。子样本的监督学习与全样本信息具有一致的结果,表明本文的央行行长口头沟通测度具有可复制性和可延展性。

Based on the verbal communication of the Central Bank Governor of China from January 2003 to August 2018,this paper generates a special dictionary for Central Bank Governor communication,and then constructs the monetary policy index and economic outlook index using the phrase weighting method. The monetary policy index can better fit the changes in the actual benchmark interest rate and deposit reserve ratio,and the economic outlook index can be used as an annulus for economic fundamentals. Furthermore,based on the supervised learning method,tendentious key words and their probability distribution are obtained by training the subsample dictionary,and the new communication text is automatically classified by the texting classifier, finally the new communication event is indexed. The supervised learning of the subsamples has similar results with the full sample information,indicating that the Central Bank’s verbal communication measure is replicable and extensible.