Universiteit van Amsterdam
Graduate School of Communication
When the dragon soars: examining the interactions between
media attention to negative economic news, economic
situation and consumer confidence in China (1996 - 2013)
A thesis submitted by
Peiwen Zhong
Student Number: 10389911
In partial fulfillment of the requirements for the degree of
Master of Science
in
Communication Science (Research MSc)
Supervisor: Prof. dr. Rens Vliegenthart
Date: June 27, 2014
2
Abstract
This study is aimed at investigating the relations between economic situation (which is measured by stock market), media’s attention to negative economic news (which is represented by the counts of negative words appearing in news coverage), and consumer confidence in China during the 1996-2013 period. By running vector autoregressive analysis (VAR) on both weekly and daily-level data to test research questions and hypotheses, only the causal relation between negative economic news and the economic situation under a daily-level has been observed, i.e., media’s attention to negative economic news is Granger-causing stock market. The paper finally provides some explanations for the insignificant causal relations: (1) China is not suffering from the worldwide economic crisis; (2) China’s unique and strict media control as well as economic policy making mechanisms eliminate the impact of media coverage; (3) the data strategy maybe influential to the results.
3
1. Introduction
The world is changing drastically in a way that nobody could have imagined in the past decades. One reflection of this change is the increasing vitality of emerging economies. China, a poor and backward socialist country in the past, is benefiting from its reform and opening policy since the beginning of 1978 and has become one of the most powerful countries in the world. In spite of the global economic crisis since 2009 which struck the entire capitalist world, China's economy still maintains a good momentum of growth. According to the statistics of the World Bank, the GDP growth of China was 9.2% in 2009 and 7.8% in 2012, which outshines other
economies1. Under this circumstance it is conceivable that in this country, an
increasing number of people participate in daily local, and/or global economic activities such as stock market investments. In this process, media play a crucial and not replaceable role beyond all doubt. Stakeholders, such as investors, economic analysts obtain and analyze information from media to make decisions.
In this study I investigate the dynamic relationships between economic situation, consumer confidence and media’s attention to negative economic news in China between 1996 and 2013. This study contributes to existing academic work in the following aspects. First, I look at the case of China, a developing country rarely examined in existing studies which focus on western countries in general (an
exception see: Wu (2011); more discussion about this in upcoming sections); second, I look at a longer period of time (17 years), which largely captures the patterns of economic status fluctuations as well as variation in media attention; also it will better satisfy the minimum sample size for statistical analysis (VAR in this study); third, I
1
4
employ computer-assisted content analysis to retrieve as much information contained in media content as possible; fourth, I compare different levels of data. Due to the reality that the chance of obtaining original data with consistent level of analysis is rare (an exception is the work of Wu et al (2002) in which all data is gathered in a monthly-level), original data transformation is necessary to prepare for a quantitative analysis in most cases when original data source level varies. So, I investigate how different levels of analysis (daily and weekly-level data) reflect the causal relations between variables.
This paper is structured as follows: first, I look at the theories of previous studies and formulate my research questions and hypotheses according to them; second, I present how the data were collected and the methods employed in my analysis; finally, I will show my findings and provide conclusions and explanations.
2. Literature Review, Research Questions, and Hypotheses
2.1 News in economic communication
Economic situation varies from minute to minute everyday but one can hardly experience and be exposed to all of these “key events”, except for a few stakeholders or journalists. In this case, news media seems to be the most essential and effective channel to obtain information for most people (Fogarty, 2005). This is an important reason why researchers look at the role of economic news in economic
communication successively. For one thing, given the channel of news media, people can obtain information that they cannot experience themselves, for another,
5
analyze the government economic policies (such as fiscal policies and interest rate adjustments) and other relevant information revealed in news coverage.
Studies have shown that how much news people exposed to determines the evaluation of the importance of an economic issue (Brosius & Kepplinger, 1990). This finding is highly consistent with agenda-setting theory, which unravels the possibility that the importance of an issue may be enhanced by more exposure to news information about a specific topic (McCombs & Shaw, 1972). After the clarification of media impact researchers start to look at the factors that can determine media’s attention to economic news, an important one of which is the asymmetrical economic news coverage on “good” and “bad” news: news media tends to focus more on negative information compared to positive information (for example: Fogarty, 2005; Headrick & Lanoue, 1991; Hetherington, 1996; Kurtz, 1990; Nadeau et al, 1999; Soroka, 2006). As Fogarty (2005) points out, economic issues are not treated equally by news media agencies. This is also described by Kurtz (1990) as “media malady”. Haller and Norpoth (1997) try to explain this phenomenon: news media treat negative economic reality as a more attractive and valuable information source, hence they are more sensitive and active to make responses to it (by reporting more negative
economic news). In other words, the emphasis on negative economic news is a reflection of the marketization trend of news media.
2.1.1 Economic news and consumer confidence
Straightforwardly, the foundation of economic performance evaluation is the economic situation (economic reality): economic performance evaluation is the reflection of economic situation. The fall of economic indexes (for example, unemployment rate) usually lead to a more pessimistic evaluation to economic
6
situation (an example see: Fuhrer, 1988). Even though research notices that personal perceptions and experiences influence public economic opinion formation (Conover
et al, 1987; Weatherford, 1983), the impact of it on economic opinion is not as large
as people have thought. MacKuen et al (1992) state that the general public acts as “peasants (those who rely on either short-term economic performance or their personal experiences)” other than “bankers (those who look at current economic information in combination with personal future expectations)” when evaluating the economic or political performance of a government. Notably, as Mutz (1992) claims in her paper, the importance of personal considerations in economic performance evaluation is decreased by negative news coverage. Negative economic coverage acts as a mediator in the formation of social-level (other than personal-level) economic consideration. Collective opinions to economy are magnified and enhanced by negative economic coverage in this process, however on the contrary the impact of personal experiences (such as unemployment) and judgments on the social-level economic consideration is constrained by media coverage (Mutz, 1992). Just as mentioned above, one can hardly experience all the necessary information they need before they make an economic evaluation (Fogarty, 2005; MacKuen et al, 1992). Hence, the role of economic news in economic opinion formation is crucial.
The evaluation of economic situation is always shown in a form of “consumer confidence” (Matsusaka & Sbordone, 1995). The relation between economic news and consumer confidence has been investigated in a considerable number of studies. Research reveals that the economic reality perception and economic opinion
formation of the general public are largely influenced by the economic messages conveyed by news media (Alsem et al, 2008; Brosius & Kepplinger, 1990; McCombs & Shaw, 1972). With respect to the direction of influence, a strain of literature
7
suggests that consumer confidence level is influenced by negative news coverage (De Boef & Kellstedt, 2004; Goidel & Langley, 1995; Hollanders & Vliegenthart, 2011; MacKuen et al, 1992; Wu, 2011; Wu et al, 2002):
Goidel and Langley (1995) find out that economic news coverage has an
independent effect on the general public’s evaluation of economy; Blood and Phillips (1995) look at the causal relationship between consumer confidence and negative news coverage during one presidential election cycle (during 1989-1993 period). Wu
et al (2002) examined the impact of media coverage on consumer confidence of the
United States during the 1987-2006 period. By employing vector autoregressive (VAR) analysis and controlling for the economic situation, the authors find out that media’s coverage about economic recession conditionally predicts economic opinion, and the size of the impact varies across time. In other words, only under some
circumstances (economic slump in this case) will media coverage influence economic performance expectation, during the recovery period, the influence is absent.
Alsem et al (2008) try to answer the question whether economic news coverage “spins” public economic perception. By choosing the front page and economic page of two Dutch newspapers (De Telegraaf and NRC Handelsblad) for a content analysis, the authors observe a short-term and non-immediate effect of media on consumer confidence, even though the current economic situation (stock market index) has been controlled; another study of Hollanders and Vliegenthart (2011) empirically test the causal relation between negative news coverage and consumer confidence between 1990 and 2009 period and find out that journalists amplify negative news coverage and decrease consumer confidence.
8
Later Wu (2011) conducts a similar analysis for the cases of United States and China during the global recession (2008-2009) periods. Interestingly, by comparing the causal relation between economic coverage, consumer confidence and economic situation, the causal relation between economic news coverage and economic
expectation has only been observed in China, while this causal relation is absent in the United States.
But another strain of literature criticizes the conclusion introduced above, arguing a reverse relation between negative news and consumer confidence, that is, low consumer confidence is expected to increase the media’s attraction to negative news content (Stevenson et al, 1994). As two interesting studies of Headrick and Lanoue (1991) and Wu et al (2002) conclude, the public reads more news during economic recession periods. According to their findings, the key point is that the general public is more motivated to obtain economic news during economic downturns, other than the market-orientation of news media. This can also be
explained by the fact that some news agencies are market/audience-driven, aiming at attracting more readers. Once readers show more interests in negative economic news compared to positive news, journalists and editors produce more news items which discuss negativities of economy, in order to cater to the demands of readers.
Notably, the paper of Hall and Norpoth (1997) provides strikingly different conclusions. The authors demonstrate that the general public makes economic evaluation according to a relatively stable “path”, regardless of what kind of
economic content is provided. They compare the economic evaluations of non-media readers and media readers and find that non-media readers surprisingly provide dominant economic opinion.
9
From the two strains of literature I have reason to believe that the relation between negative news coverage and consumer confidence can be interpreted by two different ways. On one hand, negative economic news influences consumer
confidence in that news media is an important source of the general public; on the other hand, it is likely that consumer confidence may influence media’s attention to negative economic news, since they have the market-oriented inclination. Hence, based on the theories and findings above I formulate the following hypotheses:
H1: Media’s attention to negative economic news has an impact on consumer confidence. The more negative economic news, the lower the consumer confidence will be.
H2: Consumer confidence influences media’s attention to negative economic news. Lower consumer confidence leads to more media attention to negative economic news.
2.1.2 Economic news and stock market
How economic news is linked with the economic situation, especially as reflected in stock market, has attracted a lot of focus (Engelberg & Parsons, 2011; Hollanders & Vliegenthart, 2011; Tetlock, 2007; Vliegenthart & Mena Montes, 2014). The relation between economic news and stock market has attracted interdisciplinary focus, which is discussed majorly in the realm of economy and finance (for example: Bhattacharya et al, 2009; Engelberg & Parsons, 2011; Fang & Peress, 2009; Peress, 2008; Tetlock et al, 2008). A branch of academic works related to this topic concede that information conveyed in media content significantly affects stock market (an exception is Fama (1970) which claims that media have no effect on stock prices). For
10
example, Fang and Peress (2009) find a cross-sectional impact of media coverage on the stock returns. Stocks with more media coverage obtain less return compared with those stocks that have less media coverage. Engelburg and Parsons (2011) control for geographic factors and find that local media coverage predicts local trade volume.
Another branch of research looks at the role of negative news stories in stock-market behaviors. Tetlock’s (2007) article takes the daily content in a popular Wall
Street Journal column as an example and observes that higher pessimism of news
coverage leads to a downward pressure of stock market, and the trade volume increases in the meantime. This finding is consistent with the “noise trader” theory, which claims that investors naturally have the inability to obtain “inside information”, so they make decisions based on the so-called “noise” irrationally. In this process, trade values increases and media act as a proxy. Also, investors are risk-averse, the more negative information they’ve obtained, the larger the probability they are going to take actions in stock market, thus the trade volume is expected to increase. One year later, Tetlock et al (2008) quantify language (verbal) information contained in financial news stories and find out that the fraction of negative words in news stories lowers the accounting earnings and stock returns of firms.
Nevertheless, an early work of Fama (1970) denies the impact of media on stock prices. The author put forward the Efficient Market Hypothesis (EMH) which states that stock market is high efficient and stock prices have fully reflected the all accessible information. Even though a few stakeholders are able to obtain more information, there is a much larger amount of ordinary investors participating in stock market in the meantime. Hence it is somewhat unrealistic to expect that someone can gain big profits from stock market by relying on obtained information. As a result,
11
changes of stock prices are highly unpredictable. In addition, researchers in the realm of finance and economy define and measure the term “media” and “stock market” differently compared to communication scientists. For example, economists usually look at the number of media coverage in general and how it influences various of outcomes in stock market, such as stock prices (Solomon, 2012), market trading volume (Tetlock, 2007), household trading behavior (Engelberg & Parsons, 2011), and stock returns (Tetlock, 2011; Tetlock et al, 2008). When it comes to media content measurement, some researchers focus on the company-level media content such as earning announcements (Peress, 2008), while some others look at the
linguistic information, such as the tone contained in media coverage (Bhattacharya et
al, 2009; Tetlock et al, 2008), media coverage in general (Bhattacharya et al, 2009;
Engelberg & Parsons, 2011), and how the media “spins” the news content (Dyck & Zingales, 2003).
In the realm of communication sciences, some researchers see economic condition as a control. For example, Wu et al (2002) look at the impact of recession news on public economic opinion during different economic situation and find out that the amount of news coverage has a strong correlation with economic situation; some another similar studies, such as Hollanders and Vliegenthart (2011),
Vliegenthart and Mena Montes (2014) also have similar findings.
Most of the existing communication studies address how media content and economic situation interact with each other. Regarding the media content, it is usually measured by either the frequency of news articles (Vliegenthart & Mena Montes, 2014; Wu et al, 2002) or the frequency of linguistic content (such as negativity) contained in the articles (Hollanders & Vliegenthart, 2011). In a nutshell, economists
12
concentrate on the “impact” of media attention, while communication scientists care about how media content matters in economic communication.
According to these research findings I propose the following hypotheses: H3: Media’s attention to negative economic news has an impact on economic situation, which is reflected by stock market index. Negative economic news lowers stock market performance.
H4: Economic situation (which is reflected by stock market index) influences media’s attention to negative economic news. A worsening economic situation increases media’s attention to negative economic news.
2.2 Regional Studies
Regarding previous regional studies, scholars have focused on the cases of Japan (Wu et al, 2004), the Netherlands (Alsem et al, 2008; Hollanders & Vliegenthart, 2011; Vliegenthart & Mena Montes, 2014), the United Kingdom (Soroka, 2006), the United States (Blood & Phillips, 1995; Wu et al, 2002), Korea (Ju, 2008), Spain (Vliegenthart & Mena Montes, 2014), and China (Wu, 2011). Most of them are single-country studies, except for the ones of Wu (2011) and Vliegenthart & Mena Montes (2014). Depending on the time period that the study has investigated and the different country context, research yields very different findings.
One of the pioneering work in this area maybe Wu et al’s study (2002), in which he investigates the role of recession news in the economic communication of the US in 1987-1996 period (9 years). By using vector autoregression (VAR) analysis, the authors first divide the time series into downturn and recovery period and finally yield various findings regarding the investigation of pairwise relationships between news
13
coverage, the state of economy and public’s perceptions to the current economy. For example, different states of the economy largely influence public’s evaluations to economy; in the long run economic news coverage follows economic situation; economic performance can be predicted by public’s economic evaluation. Later the first author and colleagues (2004) conduct a similar research under the Japanese context in 1988-99 period (11 years). By using a similar method of data collection and analytical strategies, the authors find that the relationships of the three indicators changed: the impacts of media coverage on consumer confidence and economic situation are absent, which is in contrast with the earlier study. Also, economic situation does not influence the tri-variate relationship in this study.
In another more recent study, Hollanders and Vliegenthart (2011) investigate the causalities between media’s attention to negative news, economy and consumer confidence by taking different economic cycles into account in 1990-2009 period under the Dutch context. They find out that negative news coverage is the Granger cause of consumer confidence when stock market (which represents economic situation) is controlled, by taking a computer content analysis of a large Dutch newspaper, NRC Handelsblad. In addition, there is a mutual Granger causailty between stock market and negative news coverage.
The paper of Vliegenthart and Mena Montes (2014) is one of the few
comparative studies. It investigates the relationship between parliamentary questions and negative economic news coverage. In addition, stock market (which represents economic situation) and consumer confidence are controlled for.
Another comparative research of Wu (2011) seems to be the only study that is implemented under the Chinese context. It focuses on how Chinese economic news,
14
public expectation and the Leading Economic Indicator (LEI) interact with each other during 2008-2009 global recession periods. By content-analyzing the economic news of selected three Chinese newspapers (i.e., People’s Daily, Southern Metropolis Daily and China Business News) and two US newspapers (New York Times and Wall Street
Journal), the author finds out that economic news influences economic expectation of
China, while this influence is absent in the US. However the paper still has some drawbacks, as the author has pointed out: first, the author only considers the front-page news stories on traditional print media, which may miss a lot of information, narrowing the sample and consequently misinterpret the causal relationship; second, the study period is only two year (2008-2009), which is not considerably long enough to depict a VAR analysis in this case (See Richman & Moorman (2000) for a
discussion about the small sample effects in time series analysis).
Except for the above-mentioned studies which investigate how three variables of economic communication (economic news, economic situation and the economic expectation of public) interact simultaneously, some other studies narrowed the research topic and only look at the relationship between two of the aforementioned variables. For example, Soroka (2006) and Ju (2008) successively observe an
asymmetrical response of general public to economic news in the UK and Korea, that is, the general public have stronger responses to negative economic news information compared with the positive one.
It is notable that the studies introduced above largely rely on the context of western (especially developed) countries. All these countries share high press and economic freedom; also, general public is guaranteed to speak their minds. However these are not the case in China. As some studies point out, the general public also
15
takes political issues into consideration when evaluating economic situation (De Boef & Kellstedt, 2004). Using China as the research context here naturally reduces the impact of political issues on consumer confidence, since this is a country where no elections are held (even though the authorities officially call some activities “elections”). Consequently, the exogenous impact of election cycles on economic perceptions is eliminated. Even though some might say that some other political issues (such as the changes of political leaders) may also influence consumer confidence, however the general public has much less chance to participate in these processes directly compared with a “true” election. Undoubtedly, this is an excellent chance for researchers to see how economic factors influence consumer confidence, especially when the general public considers less about politics.
To enrich existing research and compare the results with Wu (2011) by using longer and more recent time series, I take China as my context of study and propose the following research question:
RQ1: To what extent do the relationships between economic situation, economic opinion and media’s attention to negative economic news in China differ from the existing studies that focus mainly on Western countries?
2.3 Level of Analysis
Previous studies that investigate relative issues employ different levels of
analysis. Levels of analyses in previous studies include: weekly-level (Wu et al, 2004), monthly-level (Hollanders & Vliegenthart, 2011; Wu et al, 2002), and even quarterly level (MacKuen et al, 1992). To run a VAR analysis we need to make the level of analysis of each variable (time series) consistent. When data level varies between
16
variables, data level transformation is required and researchers should consider whether the data level can provide substantial meanings for the analysis or not.
Different from some analyses in which all data sources share the same time scale level (one exception is the study from Kleinnijenhuis et al (2013) which uses both monthly and daily level data), in this paper I simultaneously look at the results of both weekly and daily-level data. For one thing it is an innovation compared to previous studies that only look at one single data level, for another, I expect that in this way the causalities between variables influenced by different data levels can be unraveled. Since no relevant existing studies have considered both levels of analysis
simultaneously, here I propose the following research question:
RQ2: Does time series level (daily and weekly) influence the results of VAR analysis? If yes, to what extent and how?
3. Methods
To test the aforementioned hypotheses and research questions I collect data for three variables. Details of these series are introduced in the upcoming section.
3.1 Data Collection
Media Attention
To measure media attention I look at the weekly frequencies of some negative keywords in two famous Chinese newspapers, i.e., Southern Metropolis Daily, a relatively liberal and one of the most influential tabloid in China, and People’s Daily, a party newspaper and the mouthpiece of Chinese government. Both newspapers are ranked as the top -10 circulated print media in China. Selecting these newspapers is
17
based on the assumption that Chinese tabloids tends to focus more on “livelihood issues”, while party mouthpiece newspapers pay more attention to the attitude of government, or the indication of the fiscal and/or economic policies. 2
Researchers use multiple ways to extract the content contained in newspapers, such as: focusing on the frequency of news articles (Wu et al, 2002) or the frequency of headlines contained some specific keywords in selected newspapers (Blood & Phillips, 1995; Wu, 2011; Wu et al, 2004), or newspapers’ negative references throughout all pages (Hollanders & Vliegenthart, 2011). Studies which look at only the headlines or the front page find that negative economic information is always a focus for daily news compared to other issues (Blood & Phillips, 1995), so it is reasonable to believe that most newspapers will place the negative economic news on the headline or front page. The study of Hollanders and Vliegenthart (2011) combines these techniques. The authors look at the negative content contained in both the headlines and the texts of selected newspapers. Furthermore, the frequencies of negative references contained in the headline are weighted. In contrast to researchers who concentrate on negative economic news only, Wu (2011) claims that considering negative news only is problematic in that it does not cover the complete spectrum of newspapers’ reports. In this study I focus on the full tests of related newspapers and extract the negative words contained in them.
The development of computer content analysis in the realm of social sciences provides researchers a larger possibility to look at the media content more in-depth (a brief introduction to computer content analysis see: Berg & Lune, 2004). With the assistance of computer content analysis technique, after deciding which newspapers to
2
See Huang (2001) for a detailed introduction to the development and characteristics of Chinese tabloids and mouthpiece newspapers.
18
look at I start to use the research string “economy AND (crisis OR recession OR downturn OR fall or decline)”3
in order to find out the articles which contain negative economic information from Jan 1, 1996 to the Jun 29, 2013. I download and obtain all of the eligible full tests of the two newspapers on the WiseSearch4 database in a weekly level, one of the largest LexisNexis - like search engine for pan-greater China media outlets. After a series of data cleaning procedures, I conduct computer-assisted content analyses to collect the frequencies of the keywords (words mentioned in the search string) appearing in articles to represent media’s attention to negative
economic news.
Economic Situation
To reflect the fluctuation of economic situation of China I use the daily closing value of SSE Composite Index (Shangzheng zonghe zhishu; SHA: 000001) from Jan 1, 1996 to the Jun 29, 2013 to represent the economic situation of China in the short term. The SSE Composite Index is one of the two stock market indexes of mainland China, thus we have reason to believe that it is a suitable reflection of short-term economic situation.
Needless to say, there are many other indexes of economic condition
measurements, such as unemployment rate, ratio of bank deposits, inflation rate, and so forth that can be considered as suitable depictions of current and historical
economic situation. Here I choose the closing value of stock market in order to make the results of this study comparable with many other international studies in the realm of communication sciences (some examples include: Vliegenthart & Mena Montes (2014) and Hollanders & Vliegenthart (2011)).
3
The original search string in Chinese is: 经济 AND (危机 OR 衰退 OR 下落 OR 回落 OR 下降).
4
19 Consumer Confidence
Another variable of this study is Consumer Confidence, a concept widely used in economics that measures the “feelings” of the general public to the current economic situation. To measure it I obtain data from the National Bureau of Statistics of China5.
This is an official dataset and has surveyed a considerably large population in China. The index ranges from 0 to 200. The larger the number, the higher confidence
consumers have.
Before testing my VAR model it is necessary to gain an overall view of the whole dataset. For convenience data is reported in a weekly-level in this section. The overall pattern of media attention of two newspapers is illustrated in Figure 1 below:
= FIGURE 1 ABOUT HERE =
As Figure 1 has shown, the media attention level appears to be very similar across time and both newspapers’ attention to negative news reached peaks at around the first week of 2009, when the global economic crisis and the fiscal crisis of some western countries occurred. The high correlation (r = .73, p = .00) between two newspapers can also corroborate the similar patterns of two curves in Figure 1. More concretely, in this period, the media attention of Southern Metropolis Daily is much higher than the one of People’s Daily. Since the global financial crisis continues after 2008, media attention slightly declines but still maintains a relatively high level.
Then I look at the pattern of stock market, consumer confidence and the
aggregate media attention to negative economic news across time, as shown in Figure
5
20
2. In order to show the trends of all of the three time series clearly, I logged original values.
= FIGURE 2 ABOUT HERE =
From Figure 2 we can observe that Chinese newspapers generally have more coverage when the economic situation is bad (which is reflected by the low closing value of stock market in this study). Similarly, newspapers have less negative economic coverage when economic situation is good. However this trend becomes less manifest after 2012 when stock market indexes slightly fluctuates but stable in the meantime media attention seems to have more drastic fluctuations. Consumer confidence faced a drastic decline by the end of 2008, when the global economic crisis occurred. Also, consumer confidence seems to have a negative correlation with media attention to negative economic news (r = -.59, p = .00). Higher media attention to negative economic news is always accompanied with a lower level of consumer confidence.
3.2 Vector Autoregression (VAR)
In this paper I run a vector autoregression analysis (VAR) to explore the
causalities between the aforementioned three variables. A typical VAR model can be written as:
(1)
Where is a (k * 1) vector and is a (k * k) matrix. From equation 1 we can see that VAR model is a more generalized AR(p) model. Choosing the VAR method to test my research question and hypotheses seems suitable in that it can help to find
21
out the causality between variables across time and the direction (one way/two way) of it can also be identified.
3.3 Level of Analysis
As mentioned before, in this study I compare the results of both the weekly-level data and daily-level one, due to the restriction of data accessibility. The
operationalization of changing data level is as follows: for the weekly-level analysis, I calculate the mean of stock market index each week to represent the economic
situation in weekly-level; for consumer confidence, the index of each month represents the consumer confidence each week of that month; since I collected the media attention data in a weekly level, no changes have to be made in this case. Similarly, for daily level data, the original stock market data is maintained; daily consumer confidence equals to the index of that month and daily media attention equals to the amount of media attention each week. The decision of equating the daily-level media attention to the weekly-level data is a compromise of data
accessibility and collection feasibility. The database where I obtain media content sets a content query limitation which provides a restricted daily times of access; also, media content was downloaded from the database manually. Hence, it is unrealistic and time-consuming to collect daily-level data from the database.
4. Results
Stationarity test
Stationarity is the prerequisite of running a VAR analysis. To test it I run an augmented Dickey-Fuller unit-root tests (ADF; also known as tests for unit roots).
22
The null hypothesis of ADF test is that there is a unit-root and the series is not stationary. Test results are shown in Table 1:
= TABLE 1 ABOUT HERE =
For weekly-level data, the test result of variable media attention rejects the null hypothesis (test value = -5.44 < -3.43), thus this series is stationary. The test values of stock market and consumer confidence show that we cannot reject the null hypothesis, thus both variables are not stationary. Then I differenced all variables and run the test again. All of the test results for the differenced values show that three series are now stationary.
The stationarity test results of daily level dataset are somewhat different from the weekly-level dataset: the series of consumer confidence is not stationary (test value = -1.76 > -3.43) under a 1% critical value, while the other two variables are stationary (test value for media attention = 6.28 < 3.43; test value for stock market = 18.00 < -3.43). All variables are stationary after first order differencing.
Granger Causality Test
To test the hypothesized causal relationships I run Granger Causality test. The null hypothesis of Granger Causality test is that the past observations do not Granger cause the present observations of the dependent variable.
Before running a formal Granger Causality test, the lag length has to be decided upon. There are multiple lag selection criterions that can provide suggestions on lag length decision, such as AIC (Akaike information criterion), BIC (Bayesian
information criterion) and HQIC (Hannan–Quinn information criterion). In this analysis I refer to BIC criterion’s suggestion in combination of substantial meanings
23
of the lag to finally decide which lag to choose. The BIC statistics of weekly-level data suggest a lag length of 1, while for the daily-level data, the BIC statistics suggest a length of 21 lags.
After the decision on the number of lags I start to run formal Granger Causality tests. The null hypothesis of the analysis is “there is no granger causality from variable a to variable b”. Statistical results are shown in Table 2 below:
= TABLE 2 ABOUT HERE =
Statistical results in Table 2 indicate that only one Granger causal relationship is observed in the dataset. For the weekly-level results (the second column) we cannot reject the null hypothesis, hence Granger causalities are absent in all pairwise relationships; for the daily-level data, Granger causalities are absent in all
relationships except for the one between stock market and media attention ( = 43.04, p = .00). According to this result, I conclude that media attention is Granger causing stock market.
I also test the residuals and squared residuals to insure that there is no additional correlation(s) remained in residuals. The null hypothesis for the test is “the residuals are white noise”. For weekly-level data, Ljung-Box Q statistics show that we cannot reject the null hypothesis for the residuals of media attention (Q = 13.96, p > .01), stock market (Q = 2.08, p >.01) and consumer confidence (Q = .40, p > .01), which means there are no autocorrelations left in residuals. For daily-level data, only the residual of consumer confidence is white noise (Q = .40, P > .01), which means that there is still information contained in residuals and the model has to be modified. Hence I extend the lag length to 30. Afterwards, the causal relations between all variables have not changed and all of the three variables are confirmed to be white
24
noise. For consumer confidence, Q = .03, P > .01; for stock market, Q = 14.54,
P > .01; and for media attention, Q = 3.63, P > .01. Thus we can safely say that we are satisfied with the model we’ve run.
To look at how much of the change in one variable can be explained by the other, I deploy the decomposition of forecast error variance (FEVD). Statistics show the result of FEVD from 0 to 8 steps (days), which are all insignificant. At day 1, media attention has no contribution to the changes of stock market, but the contribution gradually increases across periods: at day 2, media attention explains .5% of the change of stock market, and later this number slightly increases to 1.6% at the eighth day.
To observe how a variable reacts to a standardized external shock I run an impulse response analysis. Figure 3 illustrates that there is a time lag for the impulse (media attention) to affect response (stock market). An increase in the orthogonalized shock to stock market causes one period of decrease, but after the second period the response fades away and maintains steady around 0. Later at the sixth period, the response of stock market reaches a peak when stock market index increases by .05% in response to an orthogonalized shock.
= FIGURE 3 ABOUT HERE =
Thus far, I’ve tested the hypotheses I formulated in previous section. As the results have shown, only hypothesis 3 (which expects media’s attention to negative economic news Granger-causes stock market) is supported, while the other three hypotheses are rejected in this study.
25
5. Conclusions and Discussions
This article looks at the trilateral relationship between economic situation
(represented by stock market), consumer confidence and media attention (represented by the references of negative keywords appeared in selected newspapers) in a 17-year period (1996-2013). By running VAR analyses, only the uni-directional causal
relationship between stock market and media coverage is found in daily-level data, i.e., media attention is Granger-causing stock market. This is consistent with Forgaty (2005)’s finding. As one of the most important sources of economic information, media plays a role of conveying economic information and political economic policies. However, the reverse causal relation, i.e., “economic situation affects media’s attention to negative economic news” is absent in both data levels. This means that the influence of media marketization on economic news decision is still very weak in China.
Also, I also looked at whether the causal relationships changed across different levels of original data. Results show that no causal relationships are observed in the weekly-level dataset, but one causal relationship in daily-level data is significant. This may reveal that a daily-level data can capture the dataset more in detail.
In addition, I also compare whether the causal relations between economic situation, media attention and consumer confidence in China are different from other regional studies. First I compare the results of Wu (2011) in which a comparison between the cases of the United States and China is presented. By looking at the global recession period (2008-2009) the author finds out that the consumer confidence of Chinese public is influenced by media coverage, while in this study this causal relation is absent. One possible reason maybe that the causality that Wu (2011)
26
observed is a short-term one. In a longer time period (such as a 17-year period in this study) the effect of media coverage on consumer confidence (especially during the recession period) may be “diluted”. Not unimportantly, the author only looks at the news coverage (in general; not only negative news coverage) on the front pages of two selected newspapers and does content analysis manually, which may have a chance of underestimating or overestimating the results. The finding of the absence of media coverage-consumer confidence causality here is also in line with the finding of Hall and Norpoth (1997) which suggests that economic content conveyed in media does not matter for the economic performance evaluation of general public since they follow a relatively stabilized “path” to make judgments.
Second, I compare the results with studies conducted in some other countries. The absent effect of media coverage on consumer confidence of this study is
consistent with the finding of Wu et al (2004) in their Japanese study during the ‘Lost Decade’. The authors claim that the reason is that the Japanese public has suffered from the recession for a long time, making them feel deeply depressed. Hence, the role of media attention becomes insignificant.
In my study, the two-directional causal relationship between media’s attention to negative economic coverage and consumer confidence is also absent in both levels. Here I propose the following arguments that may help explain these findings:
First, China is not suffering from the economic crisis. Different from a local economic recession/crisis, it is conceivable that China’s general public may become less sensitive towards economic crisis news coverage: as Wu (2011) states in his paper, “… to Chinese, it is an issue of recession threat rather than of reality (p.3)”. The economic growth of China during the global recession may be a good example.
27
Based on the fact that China’s economic growth maintains a very high number, to what extent do the general public in China influenced by the economic news coverage, which is largely related to non-domestic issues, remains to be unknown.
Second, China’s unique media system and mechanism of economic policy making may be an important factor that influences the causal relationships between variables. In this study, the two-way causal relationship between media’s attention to negative economic news and economic situation, between media’s attention to negative economic news and consumer confidence, are absent. In China, a one-party state where no elections are held, media content is strictly controlled and censored. Maybe sounding somewhat unusual, the party worries about “social instability” when developing Chinese economy. The concept of maintaining social stability is usually called “weichi wending; weiwen” which is especially emphasized after President Xi Jinping came to power (Kelly, 2011).
However the concept of Chinese social instability is seldom discussed in academia, except for a recent paper of Knight (2013) which argues that economy is believed to be a “source of social stability” in China. The society needs to be
maintained stable in a long-term. When the political legitimacy of the government is challenged, the growth of economic development acts as “a fire extinguisher” to shift the focus of general public. If economy does not grow, unemployment rate and the gap of wealth increases, the party will have a higher risk of being overthrown. In this case, high economic development can be seen as a tool for current party leadership to maintain and strengthen social stability and political legitimacy. Furthermore, the party manipulates media content for the sake of maintaining what they call “social stability”. News contained negative economic information can be reported only when
28
it is insured to be “under an acceptable level” (not to trigger social instability). For example, news about how economic crisis influences economies overseas is considered to have a higher chance of being published than news regarding the
consequences of economic crisis for domestic issues (such as high unemployment rate, high Gini coefficient, and so forth). Just as Wu (2011) concludes in his work, “news coverage in China is still controlled and manipulated by the authorities to advance their interests and goals (Wu, 2011, p. 15)”.
Last but not least, we cannot neglect the possibility that the data collection
strategy has influenced the results in this study. As Wu et al (2002) state in their paper, “assessing mass media influence is a tough job (p.20)”. Studies related to economic communication largely rely on secondary data, which in some ways restricts
researchers’ possibility and flexibility of choosing a suitable dataset to represent variables as well as to what extent can researchers’ dealt with existed data. Take my study as an example here. Even though I’ve collected the official historical data of consumer confidence from the National Bureau of Statistics of China, details of the methodology employed in the study, such as what kind of questions have been asked in the survey (current economic situation evaluation or future economic situation expectation), the sample size, the means of contacting respondents, are not revealed. However, some other consumer confidence indexes cannot provide a considerably long enough time series for this study. Under this circumstance I have to select the former index and sacrifice knowing the precise details of it. In addition, whether transforming the original data’s level of analysis will damage the intrinsic
characteristics is unknown. Researcher should be cautious about these and choose “the least worst option”. These are yet to be studied in future research.
29
Appendix
Table A
Descriptive Statistics, daily-level data
Variable N Mean Std. Deviation Range (Min, Max) Consumer Confidence 2737 106.18 4.57 97.00, 113.78 ΔConsumer Confidence (logarithm) 2736 -.00 .00 -.05, .08 Media Attention 2737 1221.71 805.99 56, 3767
ΔMedia Attention (logarithm) 2736 .00 .19 -2.28, 2.40
Stock Market 1814 2739.05 924.24 1197, 6092
ΔStock Market (logarithm) 2736 -.00 .17 -.92, .91
References
Alsem, K. J., Brakman, S., Hoogduin, L., & Kuper, G. (2008). The impact of
newspapers on consumer confidence: does spin bias exist? Applied Economics,
40(5), 531-539.
Berg, B. L., & Lune, H. (2004). Qualitative research methods for the social sciences (Vol. 5): Pearson Boston.
Bhattacharya, U., Galpin, N., Ray, R., & Yu, X. (2009). The role of the media in the internet IPO bubble. Journal of Financial and Quantitative Analysis, 44(03), 657-682.
Blood, D. J., & Phillips, P. C. (1995). Recession headline news, consumer sentiment, the state of the economy and presidential popularity: A time series analysis 1989–1993. International Journal of Public Opinion Research, 7(1), 2-22. Brosius, H.-B., & Kepplinger, H. M. (1990). The agenda-setting function of television
30
Conover, P. J., Feldman, S., & Knight, K. (1987). The personal and political
underpinnings of economic forecasts. American Journal of Political Science, 559-583.
De Boef, S., & Kellstedt, P. M. (2004). The political (and economic) origins of consumer confidence. American Journal of Political Science, 48(4), 633-649. Dyck, A., & Zingales, L. (2003). The media and asset prices. Retrieved from
http://www.nber.org/~confer/2003/si2003/papers/cf/dyck.pdf
Engelberg, J. E., & Parsons, C. A. (2011). The causal impact of media in financial markets. The Journal of Finance, 66(1), 67-97.
Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work.
The Journal of Finance, 25(2), 383-417.
Fang, L., & Peress, J. (2009). Media Coverage and the Cross‐ section of Stock Returns. The Journal of Finance, 64(5), 2023-2052.
Fogarty, B. J. (2005). Determining economic news coverage. International Journal of
Public Opinion Research, 17(2), 149-172.
Fuhrer, J. C. (1988). On the information content of consumer survey expectations.
The Review of Economics and Statistics, 70(1), 140-144.
Goidel, R. K., & Langley, R. E. (1995). Media coverage of the economy and aggregate economic evaluations: Uncovering evidence of indirect media effects. Political Research Quarterly, 48(2), 313-328.
Haller, H. B., & Norpoth, H. (1997). Reality bites: News exposure and economic opinion. Public opinion quarterly, 61(4), 555-575.
Headrick, B., & Lanoue, D. J. (1991). Attention, asymmetry, and government popularity in Britain. Political Research Quarterly, 44(1), 67-86.
31
Hetherington, M. J. (1996). The media's role in forming voters' national economic evaluations in 1992. American Journal of Political Science, 372-395. Hollanders, D., & Vliegenthart, R. (2011). The influence of negative newspaper
coverage on consumer confidence: The Dutch case. Journal of Economic
Psychology, 32(3), 367-373.
Huang, C. (2001). China's State-Run Tabloids: 'The Rise of City Newspapers'.
International Communication Gazette, 63(5), 435-450.
Ju, Y. (2008). The asymmetry in economic news coverage and its impact on public perception in South Korea. International Journal of Public Opinion Research,
20(2), 237-249.
Kelly, D. (2011). Stability and social governance in China. Retrieved from
http://www.eastasiaforum.org/2011/09/13/stability-and-social-governance-in-china/
Kleinnijenhuis, J., Schultz, F., Oegema, D., & van Atteveldt, W. (2013). Financial news and market panics in the age of high-frequency sentiment trading algorithms. Journalism, 14(2), 271-291.
Knight, J. (2013). The economic causes and consequences of social instability in China. China Economic Review, 25, 17-26.
Kurtz, H. (1990). Is the economy suffering from media malady. Washington Post, 28, 1.
MacKuen, M. B., Erikson, R. S., & Stimson, J. A. (1992). Peasants or bankers? The American electorate and the US economy. The American Political Science
Review, 597-611.
Matsusaka, J. G., & Sbordone, A. M. (1995). Consumer confidence and economic fluctuations. Economic Inquiry, 33(2), 296-318.
32
McCombs, M. E., & Shaw, D. L. (1972). The agenda-setting function of mass media.
Public opinion quarterly, 36(2), 176-187.
Mutz, D. C. (1992). Mass media and the depoliticization of personal experience.
American Journal of Political Science, 483-508.
Nadeau, R., Niemi, R. G., Fan, D. P., & Amato, T. (1999). Elite economic forecasts, economic news, mass economic judgments, and presidential approval. The
Journal of Politics, 61(01), 109-135.
Peress, J. (2008). Media coverage and investors’ attention to earnings announcements.
INSEAD, Fountainebleau, France.
Richman, J. S., & Moorman, J. R. (2000). Physiological time-series analysis using approximate entropy and sample entropy. American Journal of
Physiology-Heart and Circulatory Physiology, 278(6), H2039-H2049.
Solomon, D. H. (2012). Selective publicity and stock prices. The Journal of Finance,
67(2), 599-638.
Soroka, S. N. (2006). Good news and bad news: Asymmetric responses to economic information. Journal of Politics, 68(2), 372-385.
Stevenson, R. L., Gonzenbach, W. J., & David, P. (1994). Economic recession and the news. Mass Communications Review, 21, 4-19.
Tetlock, P. C. (2007). Giving Content to Investor Sentiment: The Role of Media in the Stock Market. The Journal of Finance, 62(3), 1139-1168. doi:
10.2307/4622297
Tetlock, P. C. (2011). All the news that's fit to reprint: Do investors react to stale information? Review of Financial Studies, 24(5), 1481-1512.
33
Tetlock, P. C., Saar-Tsechansky, M., & Macskassy, S. (2008). More than words: Quantifying language to measure firms' fundamentals. The Journal of Finance,
63(3), 1437-1467.
Vliegenthart, R., & Mena Montes, N. (2014). Media-politics interaction in times of economic crisis: a comparative study of Spain and the Netherlands.
International Journal of Public Policy/Politics, 19(3).
Weatherford, M. S. (1983). Evaluating economic policy: A contextual model of the opinion formation process. The Journal of Politics, 45(04), 866-888. Wu, H. D. (2011). Chinese and U.S. media coverage of the global recession
(2008-2009) - linking economic news and public expectation. Paper presented at the
WAPOR 64th annual conference, Amsterdam, the Netherlands.
Wu, H. D., McCracken, M. W., & Saito, S. (2004). Economic Communication in the ‘Lost Decade’News Coverage and the Japanese Recession. Gazette, 66(2), 133-149.
Wu, H. D., Stevenson, R. L., Chen, H. C., & Güner, Z. N. (2002). The Conditioned Impact of Recession News: A Time‐ Series Analysis of Economic
Communication in the United States, 1987–1996. International Journal of
34
Tables and Figures
Table 1
Stationarity test of variables (weekly and daily level data)
Variable Name (logged value) Weekly-level Daily-level
Media attention -5.44 -6.28 Stock market -2.29 -18.00 Consumer confidence -1.20 -1.76 ΔMedia attention -27.42 -52.28 ΔStock market -14.95 -52.34 ΔConsumer confidence -14.57 -52.29
Note: 1% critical value = -3.43, 5% critical value = -2.86, 10% critical value = -2.57.
Table 2
Granger causality test results, weekly and daily-level data
Relationship Weekly data Daily data
MA – SMKT 15.29 (.81) 43.04 (.00)** CC – SMKT 14.10 (.44) 2.38 (1.00) SMKT – MA 21.69 (.42) 28.01 (.14) CC – MA 5.12 (.65) 9.44 (.22) SMKT – CC 4.69 (.99) 4.58 (.99) MA – CC 6.47 (.49) 5.50 (.60) Notes:
a. SMKT: Stock Market, CC: Consumer Confidence, MA: media attention b. p-values in parentheses, where: *p < .10; **p < .05
35
Figure 1
Media Attention of two Chinese newspapers
0 500 1000 1500 2000 1996w1 1997w1 1998w1 1999w1 2000w1 2001w1 2002w1 2003w1 week
36
Figure 2
Media attention, stock market and consumer confidence (logged value)
4 .5 5 4 .6 4 .6 5 4 .7 4 .7 5 C o n su me r C o n fi d e n ce 4 5 6 7 8 9 1996w1 1997w1 1998w1 1999w1 2000w1 2001w1 2002w1 2003w1 week
37
Figure 3
Impulse response analysis (impulse = media attention, response = stock market) -.005 0 .005 .01 0 5 10 order1, ln_nr_d1, ln_smkt_d1 95% CI orthogonalized irf step
