Abstract: Macroeconomic indicators reflect status of economic entities in different domains, and are important for economic trend prediction, policies making and consumption trends nowcasting. As the advanced search engine service provider in China, Baidu has massive time-series data of different searches that uncover user search behaviors, which have some relationship with economic activities. As economic indicators infer to many different fields, how can searches be utilized to nowcast leading economic indicators is still an unsolved problem with significant meanings. To figure out this problem, this paper proposes a method called PS（Predictable Searches） to mine the relationship between Baidu’s massive search query data and economic indicators automatically. Moreover, PS can filter some representative queries for further nowcasting tasks, as a result, not only is the field knowledge requirement eliminated, the nowcasting abilities of different kinds of searches are also figured out, which is conducive to economic development. Experiments results of the nowcasting of CPI and CCI in China verify the effectiveness of PS.

Key words: search time-series data, economic indicators, automatically nowcasting

摘要： 宏观经济指标能够反映经济实体在多个领域中的活动状态，对经济走势的预测，相关政策的制定以及消费趋势的预判都有重要意义。作为中国领先的搜索服务提供商，百度拥有海量的搜索时序数据，暗含着亿万用户的搜索行为，切实反应了用户的关注焦点，某种程度上构成了与经济活动的间接联系。由此，利用搜索时序数据预测经济指标变得意义重大，然而，如何根据搜索行为预测经济指标这样涉及多个领域的宏观指标，仍然是一个悬而未决的难题。针对这种情况，提出了PS（Predictable Searches）方法，自动地挖掘百度搜索查询数据与经济指标间的关系，筛选出具有代表性查询数据，预测经济指标，不仅消除了同类方法中领域专家知识的成本代价，同时提升了对经济指标的预测效果，并且揭示了不同种类的搜索查询数据预测经济指标的能力，有利于指导经济活动的健康进行。对中国的CPI（Consumer Price Index，居民消费价格指数）和CCI（Consumer Confidence Index，消费者信心指数）等先行经济指标的预测，充分证明了PS方法的有效性。

关键词: 查询时序数据, 经济指标, 自动化预测