大数据时代下使用互联网搜索量预测CPI——基于LASSO和核偏最小二乘的联合使用

Apply Internet Search Index to Predict CPI in the Big Data Era——Based on Joint Application of LASSO and Kernel Partial Least Squares

本文结合LASSO和核偏最小二乘回归研究百度搜索指数对CPI的预测作用。首先,选取宏、微观经济领域与CPI相关的、有代表性的百度搜索关键词,将其月平均指数作为CPI预测的一类自变量。其次,联合使用LASSO算法和核偏最小二乘回归,用网格搜索法选取最优参数组合,对自变量集进行有效降维得到最优预测。最后,比较单独使用互联网搜索量、单独使用政府统计数据和同时使用两者对CPI的预测效果,结果显示,互联网搜索量和政府统计数据对CPI预测的效果具有互补性,同时使用两者可提升预测的准确性,且能较好地预测转折点。

Based on Chinese residents’daily usage habits of Baidu search engine for gathering information,this paper tries to study the role of Baidu search index in the CPI forecast when LASSO and the kernel partial least squares are jointly applied.We first select the macroeconomic and microeconomic Baidu search keywords associated with the consumer price index,and use their relevant Baidu search average indexes as one class of explanatory variables to predict CPI.Then we jointly use the LASSO algorithm and the kernel partial least squares regression by choosing the optimal parameter combination of the two methods to efficiently reduce the dimension of the variables and achieve the optimal prediction.Finally,we compare the forecast results of the three models only with the government statistics as regressors,only with Baidu search index as regressors,and with both the government statistics and the Internet search index as regressors,and find that government statistics and Internet search index are complementary in predicting CPI and the combined usage of both is a better way to improve the accuracy of CPI prediction and well predict the break point.