Measuring Economic Uncertainty in China^†

ABSTRACT

This study develops a new economic uncertainty (EU) index based on Chinese newspapers to address the media coverage bias of existing measures. We investigate how the EU affects China’s macroeconomy. Our results suggest that the EU reduces aggregate output. We find that uncertainty predicts fluctuations in economic activity and actual economic activity also predicts EU, but nonlinearly. Furthermore, we show that uncertainty in the United States leads to uncertainty in China, implying that negative EU on the Chinese economy is coming from the U.S. Finally, we conduct some asset-pricing tests, showing that EU can predict stock returns and commands risk premium. Our results are helpful for both researchers and policymakers to stabilize the economy and financial markets in China.

KEYWORDS:

• Economic uncertainty
• china
• newspapers
• Granger causality
• nonlinear causality

JEL:

• C22
• D80
• E32
• E66

Notes

^1. The original statement is “The heightened level of policy uncertainty, especially regarding trade, that has been exacerbated by recent political developments–most notably in the United States and the United Kingdom–may be amplified over time by mounting protectionist tendencies, slower potential growth, and elevated vulnerabilities in some emerging markets and developing countries.”

^2. The term “uncertainty” encompasses both “risk” and “Knightian uncertainty”. The former means the probabilities of potential outcomes are known, but which outcome will occur is not, while the latter means neither the probabilities of outcomes nor the eventual outcome are known (Knight 1921). It is difficult to disentangle these two, so in this study we refer to a single concept of uncertainty that incorporates both elements.

^3. We will discuss the advantages of our index in comparison with other uncertainty measures in Section 2.

4. Policy uncertainty might contribute to a steep economic decline and increase in economic uncertainty (See International Monetary Fund 2012, 2013).

^5. Although our main focus is to measure EU in China, we follow the same method to generate EU in other regions outside mainland China, including Taiwan, Hong Kong, Singapore, Malaysia, and Macau. The EU index for these regions is reported in the Appendix.

^6. Jurado, Ludvigson, and Ng (2015) define uncertainty as the conditional volatility of the unforecastable component of the future values of a series.

^7. The traditional proxy for EU is stock market volatility, but it appears driven by factors associated with time-varying risk-aversion rather than EU (Bekaert, Hoerova, and Duca 2013).

8. In fact, in Davis, Liu, and Sheng (2019) index (available at http://www.policyuncertainty.com/china_monthly.html), China’s EPU index from 2000 to 2018 spikes due to overseas risk events, such as the U.S. government shutdown in 2014 and the Iraq Invasion in 2003, and most of the spikes in their economic policy uncertainty index during 1949–1999 are due to domestic events.

^9. Note that we add one additional criterion when using overseas newspapers because it does not only report uncertainty related to China but also its local economy. Here we use “China (中国)”, and “mainland China” (中国大陆,内地) as location criteria to ensure that this article is related to China’s EU.

^10. Moore (2017) observes that business-focused newspapers have a higher proportion of EU-related articles, given their business focus.

^11. All general-interest newspapers in mainland China are owned and supervised by the Chinese Communist Party Committees. We use economic and financial news instead. Seven out of these 36 (19.44%) newspapers are privately owned.

12. There are 34 mainland China newspapers, and two Hong Kong newspapers in our sample.

^13. We add “risk” as one of the search terms since it is closely related to “uncertainty”, even though these two words are different in English. This situation is closer to Spanish as Ghirelli, Pérez, and Urtasun (2019) also use “risk” as a keyword in Spanish when constructing Spain’s EPU index. In fact, Hassan et al. (2019) study the dictionary and newspaper, and decide to use the synonyms of “risk” and “uncertainty” in generating their political risk/uncertainty indices. “Unstable” and “Unpredictable” are also in their list.

^14. We select 1999 as the beginning of our sample because newspaper data before 1998 is sparse.

^15. Table B in the Appendix lists the newspapers used in constructing the alternative EU index.

^16. Bekaert, Hoerova, and Duca (2013), Bloom (2009), and Caggiano, Castelnuovo, and Groshenny (2014) use stock market volatility as a proxy for uncertainty. We follow Bloom (2009) and Caggiano, Castelnuovo, and Groshenny (2014) to calculate the volatility as the within-month standard deviation of daily percentage changes.

^17. Davis (2016) calculates the monthly global EPU index based on a GDP-weighted average of national EPU indices for 20 countries.

18. Thanks to the referee who provided this suggestion. Note that we use Baidu index instead of using Google because Baidu is the largest search engine in China. Moreover, Baidu provides actual search frequencies of each term, not the normalized index like Google. Many studies have used the search volume from this engine to conduct analysis in China (e.g., Fang, Jiang, and Qian 2014; Zhang et al. 2013). We simply take the average of these three selected terms to construct this search-based measure.

19. Note that we only plot Davis, Liu, and Sheng (2019) beginning from 2000 because their EPU is normalized differently for three different regimes, where the latest regime begins in 2000.

^20. This occurred when Premier Wen Jiabao lowered the 2005 economic growth target, but the fiscal and monetary policies were unchanged (see http://www.chinadaily.com.cn/english/doc/2005-03/06/content_422130.htm).

^21. The data for the Shanghai (Securities) Composite Index is the three-month-averaged daily closing index obtained from the Wind Info database, the data information system created by the Shanghai-based company Wind Co. Ltd., often referred to as the Chinese version of Bloomberg. The benchmark one-year deposit rate, employment number, and the quarterly real GDP are obtained from Chang et al. (2016) and are available at https://www.frbatlanta.org/cqer/research/china-macroeconomy.aspx?panel=3.

22. Please see Table 3 and we will discuss the detail in Section 3.2.

23. The specification of 2) one lag and 3) two lags are the alternative to optimal three lags selected by HQ information criterion in the VAR.

24. Following Bloom (2009), who argues that all macroeconomic variables should be detrended, we use λ = 1,600 because our variables are quarterly observations.

25. The local projections method is proposed by Jordà (2005), which is less vulnerable to misspecification of VAR model.

26. Note that Granger causality is defined in the sense of predictive relationship, even though it is called Granger “causality” test.

27. Note that the nonlinear Granger causality test is based on two variables (EU and GDP (or BCI)) only because this approach only allows for two variables.

28. The data for the benchmark one-year deposit rate, and the quarterly real GDP are from China Stock Market & Accounting Research (CSMAR). We use the same Shanghai (Securities) Composite Index for aggregate stock market as we did in section 3.1. The sample period begins from the fourth quarter of 2004 to the second quarter of 2018.

29. Thanks to the referee who provided this useful suggestion.

30. Thanks to the referee who provided this helpful suggestion.

31. gcovid19.w(access/(accessed on 25 Dec 2020).

32. https://www.ipsos.com/en-us/news-polls/abc-news-coronavirus-poll

33. https://www.wsj.com/articles/chinas-economy-continues-broad-recovery-despite-covid-19-surge-elsewhere-11608013339

Additional information

Funding

This work was supported by the Southern University of Science and Technology [Y01246210, Y01246110].