|
|
|
| News Data and Real-Time Consumption Growth Projections: A View from Narrative Economics |
| ZHANG Yifan, LIN Jianhao, FAN Jiacheng
|
Business School, Sun Yat-sen University;
Lingnan College, Sun Yat-sen University |
|
|
|
Abstract At the beginning of 2020, the COVID-19 pandemic had a massive impact on the Chinese economy, leading to debates about whether consumption would slow down after recovery from the pandemic. It is crucial to achieve real-time forecasting consumption growth, which is important for government policy implementation. Moreover, accurately predicting the changes in consumption growth can provide timely warnings about downward pressures on consumption, allowing sufficient time for stimulus policy plans. Consumers' expectations of future economic conditions, personal incomes, price levels, and macro policies, among other factors, affect their investment and consumption decisions. Narrative economics suggests that consumer expectations are largely guided by media narratives. Therefore, consumption-related media coverage can be used to reflect changes in consumer expectations and predict future consumption activities.
This article constructs a consumption-related news sentiment index (CNSI) based on close to 600,000 news articles from five mainstream media outlets in China from 2007 to 2020. We evaluate the performance of the news-based CNSI and the survey-based consumer confidence index (CCI) in real-time consumption projections. We study the time-series characteristics of consumption growth and find an obvious trend disconnection between consumption growth and consumer sentiment in China, especially after the advent of the “new normal” period, with a decline in consumption growth and an increase in consumer sentiment. This trend disconnection is an important feature that is ignored in the literature. By using wavelet decomposition, or detrending processing, we find a positive and significant correlation between the CNSI and the short-term cyclical component of consumption growth, whereas the CCI does not identify this correlation. Correspondingly, our CNSI can considerably improve the out-of-sample forecasting of short-term consumption, and it can be applied to nowcasting and mixed-frequency forecasting. We further explore the content structures of news texts and find that the “current status” content of news performs better for real-time nowcasting, whereas the “foresight analysis” content is more effective in predicting future consumption growth. Furthermore, more objective and emotionally neutral (non-seditious) media texts perform better in nowcasting and forecasting. Compared with the Internet-based CCI, the construction of CNSI is simpler and more transparent, which results in outstanding advantages in consumption projections.
This article makes the following contributions to the literature. First, to the best of our knowledge, this study is the first to focus on the trend disconnection between consumption growth and consumer sentiment in China. We address this issue by identifying the long-term trend and short-term cyclical components of consumption growth through cycle decomposition methods, and find that their performance differs. This indicates that economic forecasting studies cannot ignore the cyclical changes in China's economy, and future research must consider cycle decomposition.
Second, this article combines consumption projections with high-frequency and real-time textual data. Compared with the official CCI, the weekly news-based CNSI is timelier and more responsive to economic changes. The mix-frequency prediction results indicate that high-frequency data can further improve the accuracy of forecasting consumption growth, which can be an appropriate direction for further research.
Third, this article examines multi-dimensional text features, such as “current status vs. foresight analysis”, and “seditious vs. non-seditious” features. This not only supports the theoretical conclusion that consumer confidence and expectations are driven by information on economic fundamentals but also provides a new perspective on the application of textual data in economic research.
|
|
Received: 26 January 2022
Published: 08 August 2023
|
|
|
|
|
|
|
|
