China EPU Index Based on the South China Morning Post

Download Data

To measure economic policy uncertainty for China, we construct a scaled frequency count of articles about policy-related economic uncertainty in the South China Morning Post (SCMP), Hong Kong's leading English-language newspaper. The method follows our news-based indexes of economic policy uncertainty for the United States and other countries.

First, we identify SCMP articles about economic uncertainty pertaining to China by flagging all articles that contain at least one term from each of the China EU term sets: {China, Chinese} and {economy, economic} and {uncertain, uncertainty}. Second, we identify the subset of the China EU articles that also discuss policy matters. For this purpose, we require an article to satisfy the following text filter: {{policy OR spending OR budget OR political OR "interest rates" OR reform} AND {government OR Beijing OR authorities}} OR tax OR regulation OR regulatory OR "central bank" OR "People's Bank of China" OR PBOC OR deficit OR WTO. We use this compound filter because it outperforms simpler alternatives in our audit study. Third, we apply these requirements in an automated search over every SCMP article published since 1995. This automated search yields a monthly frequency count of SCMP articles about policy-related economic uncertainty. Fourth, we divide the monthly frequency count by the number of all SCMP articles in the same month. We then normalize the resulting series to a mean value of 100 from January 1995 to December 2011 by applying a multiplicative factor.

A few examples clarify how the compound text filter works. If an article includes both "policy" and "government", we regard it as at least partly about government policy. Therefore, if it also contains a word in each of the China EU term sets described above, the article enters our frequency count for economic policy uncertainty. However, the word "policy" is not sufficient for an article to count as policy uncertainty; the article must also contain one of "government", "Beijing" or "authorities". Certain other terms in our text filter - e.g., "tax" or "regulation" - do not involve a compound requirement. We determined when to apply a compound requirement based on our audit study.

Our audit study considers 500 randomly sampled articles drawn from the universe of SCMP articles that satisfy the China EU term sets. The sampling period is January 1995 to February 2012. We subject all 500 sampled articles to human readings to evaluate the accuracy of various text filters. In assessing accuracy, we regard the classifications produced by the human readings as correct.

According to the human readings, 492 of the 500 sampled articles pertain to economic uncertainty for China. The remaining 8 articles were incorrectly flagged by the automated search method as pertaining to economic uncertainty for China. In other words, the China EU term sets produce a very small false positive error rate for economic uncertainty pertaining to China.

Using automated methods to further classify the articles as about policy-related economic uncertainty, or not, is more challenging. Here as well, however, our preferred text filter (described above) produces good results:

  • The policy-related economic uncertainty count produced by automated search methods exhibits a correlation of 0.82 with the true count (human reading) in quarterly time-series data.
  • The net error rate produced by automated search methods is nearly uncorrelated (-0.15) with the true count in quarterly time-series data.
  • The overall false positive rate produced by the automated method is 0.11. The overall false negative rate is 0.21.

Baker, Scott, Nicholas Bloom and Steven J. Davis, "Measuring Economic Policy Uncertainty," Quarterly Journal of Economics, November 2016, Vol 131, Issue 4

Baker, Scott, Nicholas Bloom, Steven J. Davis, and Xiaoxi Wang, 2013. unpublished paper, University of Chicago