Text analysis of news spots Chinese financial risk, says paper

Machine learning tools can analyse media coverage of the Chinese economy to detect financial risk, a working paper published by the European Central Bank finds.

In Using machine learning to measure financial risk in China, Alexander Al-Haschimi, Apostolos Apostolou, Andres Azqueta-Gavaldon and Martino Ricci note that assessing Chinese stress levels is a difficult task. Sources of risk are rapidly changing, the authors say. It is also is hard to find consistent financial data that would allow standard time series analysis.

They outline a different approach: using machine learning techniques to analyse a large number of newspaper articles for signs of financial stress. They compile a set of relevant newspaper articles, from the South China Morning Post and the Wall Street Journal, by searching for specific risk-related keywords.

The set takes articles from the newspapers’ print editions rather than online versions, to guard against problems caused by the duplication or updates of coverage. The authors then examine these approximately 10,000 articles using topic modelling based on the Latent Dirichlet Allocation algorithm.

This judges whether articles are concerned about a particular topic relating to financial risk. The authors construct a time series based on what proportion of the newspapers’ articles on China concerned financial risk.

They compare their time series to five indexes commonly used by analysts to measure Chinese financial stress. The authors find it closely reflects periods of heightened risk.

Using textual analysis is “a valid alternative to conventional quantitative measures to track financial risk”, they argue, which could be applied to other countries.