Media Coverage and Sustainable Stock Returns: Evidence from China

Abstract

This paper explores the relationship between media coverage and stock returns using monthly data of news reports from major Chinese newspapers. We find that firms with higher media coverage in the current month have higher sustainable stock returns in the following months over a one-year period compared with those with lower media coverage, which means that media coverage has a more significant and positive influence on sustainable stock returns in the markets, dominated by individual/immature investors. These results are largely robust to various robustness checks. Further empirical results demonstrate that in the Chinese stock market, a higher level of media coverage might cause higher sustained investor attention, which may drive up the buying pressure and thus lead to higher sustainable stock returns in the following year. Our results show that the effect of media coverage on stock returns depends on the characteristics of investors.

1. Introduction

The visibility of information plays an imperative role in investment decisions [1,2]. As an essential information communication channel, the media is one of the most convenient approaches for investors, especially for small and medium investors to obtain information for investment decisions. Hence, a considerable number of studies in the literature have devoted to examining the relationship between media coverage and stock returns.

There are two conflicting views regarding how media coverage affects stock returns. On the one hand, the “risk compensation” theory proposed by Fang and Peress [3], which is consistent with “investors recognition hypothesis” [4], suggests that media coverage negatively affects stock returns. Fang and Peress [3] argue that stocks with no media coverage or less media coverage must provide a risk premium because they have a lower transparency level and higher recognition risk caused by asymmetric information. On the other hand, Barber and Odean [5] argue that media coverage positively affects stock returns in the short term. They argue that it is difficult for investors in the real world to make fully rational choices in the stock market. Investors’ portfolio decisions, particularly those of individual investors, are largely driven by attention, suggesting that investors tend to buy stocks that draw their attention. Practically, stocks with intense media coverage induce individual investors’ attention easily, leading to a higher buying pressure and thus an upward movement of stock prices. If this “attention-driven effect” does not evaporate instantaneously, media coverage may have a positive effect on stock returns. Relying on ~2.2 million articles from forty-five national and local U.S. newspapers during 1989 through 2010, Hillert et al. [6] found that firms with higher media coverage exhibit, ceteris paribus, significantly stronger stock return momentum, suggesting that media coverage can exacerbate investor biases, leading return predictability to be stronger for firms in the spotlight of public attention. Their findings show that the effect of media coverage on returns still does not reach on a consensus in mature markets. Therefore, an issue arises whether media coverage has a positive or negative effect on stock returns.

In this paper, we reexamine the relationship between media coverage and stock returns using monthly data of news reports from major Chinese newspapers. Our focus on the Chinese A-stock market is of interest for several reasons: First, retail investors dominate the Chinese A-stock market. According to the Shanghai Stock Exchange Statistics Annual (2016), retail investors as of 2015 initiate 86.91% of trading volume. Since the trading behaviour of retail investors is more likely to be subjected to the attention-driven effect, the Chinese A-stock market provides us with a special sample for understanding the relationship between media coverage and stock returns in a market dominated by individual investors. Secondly, there is no tax on capital gains from stock investments in the Chinese A-stock market. As such, tax issues will not make our analysis complex.

Our empirical results show that stocks with higher media coverage in current month persistently have higher stock returns in the following months of the year. These empirical results are robust to various robustness checks. The findings in this paper contradict to the “risk compensation” theory suggested by Fang and Peress [3], but are consistent with the “attention-driven buying” theory proposed by Barber and Odean [5] and the “momentum theory” proposed by Hillert, Jacobs, and Müller [6]. Also, the results show that media coverage has a significant and positive influence on stock returns in the markets dominated by individual investors or immature investors. This paper indicates that the effect of media coverage on stock returns depends on the characteristics of investors (immature vs. mature; individual vs. institutional) in the market. Our paper hence sharpens the understanding of the relationship between media coverage and stock returns, especially in the emerging markets dominated by individual investors instead of institutional investors.

The remainder of the paper proceeds as follows. Section 2 provides the background and a brief literature review on the relationship between media coverage and sustainable stock returns. Section 3 describes data issues. Section 4 conducts empirical analysis and presents the main regression results. In Section 5, we conduct robustness checks. Section 6 makes a further explanation and exploration. Section 7 concludes.

2. Background and Literature Review

The China security market started in the early 1990s. The Shanghai Stock Exchange was established in December 1990 and the Shenzhen Stock Exchange was established in the following year. In the A-shares market, shares are traded and exchanged with Renminbi currency in both Shanghai and Shenzhen stock exchanges. To help domestic firms not only to raise funds from abroad, but also to reduce the negative impacts of the foreign capital on the immature China security market, in 1992, China’s Securities Regulatory Commission set-up the officially domestically listed foreign investment Shares (also called B shares). In the B shares market, shares are traded in foreign currencies on both stock exchanges. Until now, two types of shares are traded on the two mainland Chinese stock exchanges in mainland China.

Both A-shares and B-shares have distinct features. Firstly, the face value of A-shares is set in Renminbi; A-shares are traded in Renminbi as well. Secondly, only domestic citizens are allowed to trade A-shares, which means the price movement of A-shares can adequately reflect the behaviours of relatively immature investors in an emerging market. Also, it is one of the main distinctions between the China A-shares market and the stock market of any other developed country. Statistically, retail investors dominate in the A-shares market. According to the Shanghai Stock Exchange Statistics Annual (2016), the trading volume of retail investors constitutes 86.91% of the whole trading volume in 2015. Finally, there is no tax on capital gains from stock investments in the China A-shares market. As such, tax issues will not complicate our analysis.

On the other hand, B shares are utterly different from A-shares. First of all, the face value of B shares is set in Renminbi. In the Shanghai Stock Exchange, B shares are traded in US dollar, whereas in Shenzhen stock exchange they are traded in the Hong Kong dollar. Moreover, B shares are limited to foreign investment to some extent, even though the China Securities Regulatory Commission has begun permitting the exchange of B shares via the secondary market to domestic citizens since 19 February 2001, which means that its exchange is a mixture of transaction behaviours of domestic individual investors and that of foreign investors. Thus, it is hard to distinguish the effect of immature investors like domestic individual investors from that of mature investors like foreign investors. Eventually, all investors investing in B share companies need to pay taxes on capital gains from stock investments, except for the overseas individual investors investing in B share companies with foreign investment. Therefore, in comparison with the B shares market, we can find that the A-shares market has two significant characteristics, which are the reason for this paper to incorporate A-shares. That is, firstly, retail investors dominate the A-shares market. Secondly, there is no tax on capital gains from stock investments in the A-shares market.

Literature Review

The role of media in capital markets has been a topical issue in finance literature in recent years. Besides investigating the role of media in corporate governance [7,8,9], another key research direction focuses on the effect of media on asset pricing. One of the earliest research in this direction is Niederhoffer [10], who tests the stock market reaction to the world events reported in New York Times and finds that world events exert a discernible influence on the movements of the stock market averages. Klibanoff et al. [11] tested whether exciting country-specific news in the front page of the New York Times affects the response of closed-end country fund prices to asset value. They found that in weeks of the news appearing on the front page of New York Times, prices reacted much more to the asset value, and the elasticity of price concerning asset value is closer to one, which is consistent with the hypothesis that news events lead some investors to react more quickly. Huberman and Regev [12] conducted a case study on EntreMed and found that a New York Times article on the potential development of new cancer-curing drugs attracted enthusiastic public attention which caused a permanent rise in EntreMed’s stock price, even though no genuinely new information had been presented. Chan [13] analyzed differences in stock price reactions after large absolute returns in the previous month depending on whether the underlying company was mentioned in a headline or lead paragraph. He found evidence of a steady drift without a reversal after bad news, which he interpreted as evidence of slow information diffusion. Veldkamp [14] found that asset market movements generate news and news raises prices as well as price dispersion. Tetlock [15] quantitatively measured the interactions between the media and the stock market using daily content from a famous Wall Street Journal column. He finds that high media pessimism predicts downward pressure on market prices followed by a reversion to fundamentals. Furthermore, Tetlock et al. [16] discovered that the fraction of negative words in firm-specific news stories forecasts low firm earnings and stock returns. Earnings and return predictability from negative words are most significant for the stories that focus on fundamentals. The findings suggest that linguistic media content captures otherwise hard-to-quantify aspects of firms’ fundamentals, which investors quickly incorporated into stock prices.

As one of the most representative research among the numerous studies in the literature on the effect of media on asset pricing, Fang and Peress [3] first investigated the cross-sectional relation between media coverage and expected stock returns. They explored that stocks with no media coverage earn higher returns than stocks with high media coverage, even after adjusting for well-known risk factors. They further argue that this cross sectional media effect is consistent with the “investor recognition hypothesis” advanced by Merton [4], positing that stocks with lower investor recognition need to offer higher returns to compensate their holders for being imperfectly diversified and riskier. By disseminating information to a wide audience, media coverage broadens investor recognition. Thus, stocks with intense media coverage earn a lower return than stocks in oblivion. Following Fang and Peress [3], Peress [17] further investigated news media’s causal impact on trading and price formation by examining national newspaper strikes in several countries, and concludes that the media contributes to the efficiency of the stock market by improving the dissemination of information among investors and its incorporation into stock prices. Ahmad et al. [18] found a negative relationship between media expressed negative tone and stock returns, which is very similar as that between media coverage and stock returns found by Fang and Peress [3]. They further confirm that the general media is significantly biased towards negative news stories, and this biasness cannot be traced to newswires or news releases by the firm in the form of 8-Ks. It follows that the news media is more likely to pick up and run with bad news stories. Thereby, they argue that the count of media articles might be a proxy for negative tone, which provides an alternative explanation for the finding of Fang and Peress [3], that high media coverage stocks underperform low or no coverage stocks.

The implicit assumption in Fang and Peress [3] is that investors can rationally analyze the investment targets’ risk and return based on the information they obtain, and then make optimal investment decision according to the trade-off between risk and return. Media reports on the stocks can provide more information, reduce the investors’ risk of recognition and thus lead to a decline in their required rate of return.

However, the “attention-driven buying” theory represented by Barber and Odean [5] argues that it is often difficult for investors to make a fully rational choice from a large number of stocks based on the trade-off between risk and return. Instead, investors have limited attention, and they are more inclined to select the stocks that have caught their attention. Media coverage on a stock will increase the investors’ attention and drive up its price. A bunch of previous studies has provided empirical evidence for the “attention-driven” theory. For example, Cook et al. [19] found that investment banks have incentives to make use of news media to influence investors’ sentiment and drive new investors to buy new shares, increasing IPO pricing. Solomon et al. [20] showed that media coverage of mutual fund holdings affects how investors assign money across funds. Fund holdings with high past returns attract extra flows; but only if these stocks are just featured in the media. The contemporary evidence that media coverage tends to contribute to investors’ chasing of past returns rather than facilitate the processing of useful information in fund portfolios, suggesting that media coverage can exacerbate investor biases instead of reducing them. Hillert, Jacobs, and Müller [6] also advocated that media coverage can exacerbate investor biases. They found that firms with higher media coverage exhibit significantly stronger stock return momentum, suggesting the “attention-driven effect” of media coverage could persist. Using inclusive data on media coverage and merger negotiations, Ahern and Sosyura [21] found that acquirers in stock mergers instigate substantially more news stories after the start of merger negotiations, but before the public announcement. This strategy generates a temporary run-up in bidders’ stock prices during the period when the stock exchange ratio is determined, which substantially affects the takeover price. Their findings implicate that firms have an incentive to manage the attention effect of media coverage to influence their stock prices during important corporate events.

From the above brief review of the literature, we can conclude that previous studies have reached a consensus that media coverage plays a vital role in the movements of stock prices, but they diverge in the direction and mechanism about how media coverage affects stock returns. Notably, there are significant differences between literature supporting the “risk compensation theory” represented by Fang and Peress [3] and literature supporting the “attention-driven buying” theory represented by Barber and Odean [5]. Existing literature has provided evidence for both theories, but it is hard to answer which theory prevails because the answer may vary with market characteristics. As a result, the negative relationship between media coverage and stock returns found in developed stock markets like the US might not necessarily hold in an emerging stock market like China. The main objective of this paper is thus to reexamine the relationship between media coverage and stock returns in China’s emerging stock market and testify whether the “risk compensation theory” or the “attention-driven buying” theory prevails in this market dominated by individual investors and by immature investors.

3. Data

The data of media coverage is collected from the “full-text database of China’s major newspapers” provided by CNKI, one of the most important databases of journals and newspapers in China. Specifically, for each A-share listed company in China in each month from 2001 to 2011, we measured its media coverage by counting the number of news reports whose title or subject includes the abbreviated name of the company. The reason why we only search the listed company’s abbreviation in news title or subject instead of full-text is to ensure that the matched news reports are closely related to the listed company. Meanwhile, the abbreviation of some listed companies may include several other meanings that are irrelevant to the listed firms. For example, if we directly search keywords, such as “China Software”, there will be many pieces of the news related to China’s software industry instead of the listed companies “China Software” (600536). Therefore, we eliminate the sample of stocks whose abbreviation has different meanings besides a stock name. Our measure of media coverage is similar to Fang and Peress [3] who use the number of major financial newspapers’ reports on listed companies so that we can have a direct comparison with their empirical findings. The impact of media on stock return still cannot be confirmed due to the contrary results between Fang and Peress [3] and Barber and Odean [5].

In addition to media coverage, our paper also constructs an investor attention index (ATT) to investigate whether investor attention plays a critical role in the transmission mechanism between media coverage and stock returns. We manually sort out the investor attention index from the search volume index data on the website of “Baidu index” (http://index.baidu.com). Especially, using each company’s abbreviation or stock code as a keyword to search on the web page of the Baidu Index, we can obtain the search volume index for each stock during each period. This method just follows Da et al. [22]’s approach to collect data of the Google search volume index. Because investors may use either a company’s abbreviation or stock code to search the companies’ information, we use the average value of monthly search volume index on either keyword (abbreviation or stock code) to construct the monthly investor attention index. The data collection on investor attention index spans from June 2006 to March 2011. Baidu’s search volume index data date back to June 2006 and are continuously updated till 2011. Recording this data is an enormous task. The reason why we only collect search volume data before March 2011 is that the data after April 2011 were not available when we began data collection, and the Baidu Index website had stopped providing search volume Index publicly since 2012 when we wanted to update it. We guess that stocks with higher investor attention probably have higher returns because of the buying pressure, as does the analyst coverage.

Other control variables are common factors that probably influence stock return, including total asset, book-to-market (B/M) ratio, analyst coverage, monthly closing price, average returns in past 12 months, ⊿turnover, Amihud’s liquidity, and volatility of returns. As can be seen from the book-to-market effect proposed by Fama and French [23], stocks with a higher book-to-market ratio have higher returns. It can be observed from the momentum effect proposed by Jegadeesh and Titman [24] that stocks with higher returns in the past also have a higher performance in the future. According to the liquidity premium theory, ⊿turnover has a negative impact on stock return, while Amihud’s liquidity should affect positively stock return. Eventually, as for volatility of returns, although risk and return trade-off theory illustrates stocks with higher volatility have higher returns, Ang et al. [25] found that stocks with recent past high volatility have low future average returns around the world. So the effect of volatility needs to be reexamined in the paper.

We use two indicators to measure monthly stock return, the main dependent variable of this research. The first indicator is the monthly stock holding period raw return, while the second indicator is the monthly DGTW excess return. We employ the approach of Daniel, Grinblatt, Titman, and Wermers [26] to construct monthly DGTW abnormal returns. Specifically, we firstly sort stocks into 125 portfolios based on the quintiles of firms’ market value, book-to-market radio, and average monthly returns in the preceding year, and then we use the average return of each portfolio as a benchmark. We calculate DGTW excess return by the difference between the raw monthly return and its benchmark.

We collect the data of stock turnover and stock returns from the RESSET Financial Research Database (http://www.resset.cn/databases), while the data of all other control variables are from China Stock Market and Accounting Research (CSMAR) database. The screening procedure of the sample is as follows. First, we drop the observations of financial companies and also companies with special treatments (ST/*ST) from all observations of A-share listed companies in China from 2001 to 2011. Second, we drop the observations of companies whose stock name has multiple meanings. Third, we drop the observations whose asset value is less than its debt value. Then, we eliminate the observations with missing values in the key variables used in regressions. Finally, 143,003 monthly observations remain after the above screening procedure. Due to the restraint of data availability, the Baidu search volume index only ranges from June 2006 to March 2011, yielding 68,306 observations of investor attention index (ATT).

Table 1. Variables’ definitions, descriptions, and predicted effects on stock return.

Variable	Definition	Description	Predicted Effect on Stock Return
Stkret	Monthly raw stock return	The stock holding period raw return in each month, expressed in percentage.
DGTW	Monthly DGTW excess return	We employ the approach of Daniel, Grinblatt, Titman, and Wermers [26] to construct monthly DGTW abnormal returns. Specifically, we firstly sort stocks into 125 portfolios based on the quintiles of firms’ market value, book-to-market radio, and average monthly returns in the preceding year, and then we used the average return of each portfolio as a benchmark. We calculated DGTW excess return by the difference between the raw monthly return and its benchmark.
Media	Monthly media coverage	The natural logarithm of one plus the number of title news reports from China’s major newspapers. We acquired the number of title news reports by searching the titles of news reports in China’s major newspapers database.	?
ATT	Investor attention	The natural logarithm of one plus the monthly searching volume index obtained from the website of Baidu index (index.baidu.com).	+
Asset	Total assets	The natural logarithm of the total assets at the previous year-end	−
BM	Book-to-market ratio	The logarithm of the book to the market ratio of equity at the previous year-end.	+
⊿Turnover	Movement of turnover	The logarithmic difference of turnover between the current month and the preceding month.	−
Analyst	Analyst coverage	The natural logarithm of one plus the number of analysts’ reports on each listed firm in the previous month.	+
Cprice	Closing price in each month	The natural logarithm of the monthly closing price of each listed company.	+
Pret12	Average Return in the preceding 12 months	The average value of monthly holding period return of each listed company in the preceding year expressed in percentage.	+
Amihud’s Liquidity	Liquidity Index	We employ the approach of Amihud (2002) to construct the liquidity index. Specifically, in each month, the liquidity index is the average value of the ratio of daily return to daily trading volume. Lower average value represents higher liquidity.	+
Volatility	Volatility of return	The standard deviation of stock daily returns in each month expressed in percentage.	?

describes variables’ definitions, descriptions and their predicted effects on stock return.

Table 2. Description of the main variables.

Variables	Observations	Mean	Std. Deviation	Min.	Median	Max.
Stkret [t+1]	143,003	1.29	13.65	−31.11	0.28	44.93
Stkret [t+2]	140,914	1.30	13.71	−31.16	0.28	45.22
Stkret [t+3, t+12]	119,233	17.13	53.22	−95.81	7.39	172.76
DGTW [t+1]	143,003	−0.05	8.46	−19.98	−0.77	29.47
DGTW [t+2]	140,914	−0.04	8.48	−20.04	−0.77	29.46
DGTW [t+3, t+12]	119,233	−0.01	28.02	−62.54	−2.60	86.12
Media	145,386	0.68	1.59	0.00	0.00	10.00
Asset	145,386	21.46	1.15	19.35	21.28	25.78
BM	145,386	−1.08	0.66	−2.76	−1.06	0.28
⊿Turnover	143,446	−0.01	0.66	−3.91	−0.04	8.69
Analyst	145,386	1.72	4.74	0.00	0.00	30.00
Cprice	145,386	2.26	0.66	−0.17	2.23	5.61
Pret12	132,577	1.62	4.83	−8.01	0.66	15.74
Amihud’s Liquidity	145,386	0.05	0.09	0.00	0.02	0.54
Volatility	145,282	2.83	1.19	0.83	2.62	6.17
ATT [t+1]	68,306	6.08	2.27	0.00	6.60	9.64
ATT [t+2]	66,641	6.10	2.25	0.00	6.62	9.65
ATT [t+3]	64,974	6.13	2.23	0.00	6.63	9.65
ATT [t+4]	63,552	6.17	2.21	0.00	6.64	9.66
ATT [t+5]	62,154	6.20	2.18	0.00	6.66	9.68
ATT [t+6]	60,755	6.24	2.15	0.00	6.67	9.68
ATT [t+12]	52,382	6.39	2.01	0.00	6.73	9.72
ATT [t+3, t+12]	52,375	8.47	2.48	0.00	8.96	11.93

provides a statistical description of each variable. We winsorized the top/bottom 1% value of each continuous variable to eliminate the interference of the outliers in the regression results. The line graph (Figure 1) gives information on the number of media coverage and analyst reports every year during the period from 2001 to 2011. As can be seen from Figure 1, the amount of media coverage witnessed a steady increase over the period, reaching the highest point in 2007. From this point onwards, media coverage marginally decreases while the figures hover after 2008, which means that the financial crisis did not bring out adverse effects on media coverage. Also, it can be observed clearly from graph one that the number of analyst reports is exploding. Especially, analyst reports soared after 2007, which means that the financial crisis seemed not to hinder the growth of the China security industry.

4. Results

4.1. Univariate Analysis Results

Table 3. Media coverage and stock returns: univariate analysis.

Average Monthly Return (Equal-Weighted)					Total Number of Observations
	Media Coverage				Media Coverage
	No	Low	High	High–No	No	Low	High
All stocks	1.01	1.91	2.10	1.09 ***	103,966	25,139	16,281
Panel A: By asset size
1	1.34	2.21	1.92	0.58 ***	31,217	7438	4723
2	1.08	2.04	2.07	0.99 ***	42,311	9743	5858
3	0.55	1.49	2.25	1.70 ***	30,096	7869	5529
Panel B: By book-to-market ratio
1	0.78	1.73	2.37	1.59 ***	30,944	7461	5160
2	1.07	1.79	2.18	1.11 ***	42,029	9909	6205
3	1.17	2.24	1.73	0.56 ***	30,993	7769	4916
Panel C: By stock returns at current month
1	1.34	2.21	1.92	0.58 ***	31,217	7438	4723
2	1.08	2.04	2.07	0.99 ***	42,311	9743	5858
3	0.55	1.49	2.25	1.70 ***	30,096	7869	5529

*** indicates t-test’s statistical significance at 1% level.

reports the next month average returns of stocks double-sorted by firm characteristics and media coverage. For each month, we first sort stocks into three groups from lowest 30%, middle 40%, to highest 30% by various firm characteristics, such as firm size, book-to-market ratio and stock returns at the current month. We employ the previous year-end data for assets size and B/M ratio, and monthly data for all the other variables. Next, we sort each characteristic-based group into three media portfolios: no media coverage, low media coverage, and high media coverage. Specifically, we divide the stocks with media coverage into the low and high media coverage groups according to the median number of title news reports on each stock in each month. Then, we calculate the next month average return of each portfolio with different media coverage (expressed in percentage); there are 129 months during the sample period ranging from January 2001 to September 2011.

Table 3 shows the next month average returns of stocks with no media coverage, low coverage, and high coverage: 1.01%, 1.91%, and 2.10%, respectively. Stocks with intense media coverage earn higher returns than those with little or no media coverage. Also, the difference of returns between stocks with high media coverage and those with no media coverage is significantly positive at 1% level. Meanwhile, we find that the difference of returns between high media coverage and low media coverage portfolios is more significant among the groups with smaller assets size, lower B/M ratio and the higher stock returns. These preliminary statistical results are just opposite to the “rick compensation theory” proposed by Fang and Peress [3], who argue that no or low media coverage stocks ask for higher risk premiums and thus yield higher returns. Instead, the above univariate analysis results are consistent with the “attention-driven buying” theory proposed by Barber and Odean [5], who argue that investors tend to buy stocks that draw their attention. Stocks with higher media coverage induce investors’ attention more easily, leading to a higher buying pressure and thus higher stock returns.

4.2. Regression Results

To further investigate whether the relationship between media coverage and stock returns stay robust at individual stock level, we ran cross-sectional regressions. The dependent variables of our baseline regressions are the stock returns in the next two months respectively, as well as the cumulative stock return from the third month to the twelfth month in the following year. To address the robustness concern, we used both monthly DGTW excess returns, constructed using the approach of Daniel et al. [26], and raw monthly stock returns. Besides the main explanatory variable—the media coverage of each stock at current month—we include firms asset, B/M ratio, and previous stock returns (including the current month stock returns and the average monthly returns in the preceding year) as control variables according to the four-factor model suggested by Carhart [27]. Meanwhile, we also control the change of turnover (⊿turnover), stock liquidity measure constructed based on the approach of Amihud [28], closing stock price in the previous month (Cprice), stock return volatility (volatility), analyst coverage, and industry fixed effects.

Since we mainly focus on the cross-sectional differences among stocks with different degrees of media coverage, we employ the method proposed by Fama and MacBeth [29] to run cross-sectional regressions with multiple time periods. The basic idea of this methodology is to run cross-sectional regressions for each time period and then take the average of the regression coefficients of all time periods. This approach is very popular in finance literature, and employed by a most recent research by Huang [30] on Chinese stock market.

Table 4. The baseline regression results: the effect of media coverage on stock returns.

	DGTW [t+1]	DGTW [t+2]	DGTW [t+3, t+12]	Stkret [t+1]	Stkret [t+2]	Stkret [t+3, t+12]
Media	0.110 *	0.176 ***	1.373 ***	0.140 **	0.251 ***	1.253 ***
	(1.97)	(3.26)	(6.40)	(2.16)	(4.11)	(5.11)
Asset	0.026	−0.037	−0.683 ***	0.014	−0.104	−0.881 **
	(0.43)	(−0.61)	(−4.32)	(0.13)	(−0.99)	(−2.30)
BM	−0.169	−0.096	−0.352	0.111	0.228 **	3.146 ***
	(−1.25)	(−0.74)	(−0.90)	(0.96)	(2.04)	(11.33)
⊿Turnover	−0.482 ***	−0.219 ***	−0.039	−0.746 ***	−0.260 ***	0.061
	(−5.21)	(−2.62)	(−0.15)	(−6.23)	(−2.69)	(0.24)
Analyst	0.408 ***	0.379 ***	2.286 ***	0.480 ***	0.427 ***	1.945 ***
	(3.24)	(3.96)	(6.66)	(3.90)	(3.52)	(4.88)
Cprice	−0.187	−0.147	−0.677	−0.521	−0.305	−1.996 **
	(−0.83)	(−0.65)	(−0.94)	(−1.64)	(−0.96)	(−2.18)
Pret12	0.033	0.050 *	−0.134	0.060	0.042	−0.162
	(1.19)	(1.74)	(−1.52)	(1.58)	(1.05)	(−1.52)
Amihud’s Liquidity	17.195 ***	8.860 ***	16.868 **	18.737 ***	13.363 ***	39.869 ***
	(3.69)	(2.98)	(2.51)	(3.18)	(3.40)	(4.87)
Volatility	−0.288 ***	−0.136	−1.078 ***	−0.275 **	−0.106	−0.851 ***
	(−2.96)	(−1.39)	(−4.58)	(−2.02)	(−0.83)	(−3.00)
Constant	0.327	1.530	22.984 ***	2.531	4.472	48.944 ***
	(0.23)	(1.04)	(4.66)	(0.87)	(1.45)	(5.12)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	128,846	127,030	108,569	128,846	127,030	108,569
R2	0.122	0.119	0.127	0.198	0.185	0.192

We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 4. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

shows that the media coverage in the current month has a significantly positive effect on the stock returns in the following months during a year, either using DGTW excess returns or raw stock returns as dependent variables.

On average, stocks with 1% higher media coverage will earn 0.11% higher next month DGTW excess returns and 0.14% higher next month raw returns. At the same time, 1% higher current month media coverage will result in 0.176% higher next two-month DGTW excess returns and 0.251% higher next two-month raw returns. Furthermore, stocks with 1% higher current month media coverage earn 1.373% higher cumulative DGTW excess returns and 1.525% cumulative raw returns from the third month to the twelfth month in the following year. The regression results indicate that stocks with higher media coverage earn significantly higher returns at least in the following one year than those with lower media coverage. Our findings contradict the conclusion drawn from the “risk compensation theory” proposed by Fang and Peress [3], but are consistent with the “attention driven theory” suggested by Barber and Odean [5] and the “momentum theory” proposed by Hillert, Jacobs, and Müller [6].

5. Robustness tests

5.1. Eliminating Special Events’ Impact

It is well documented that postearnings announcements, dividend announcements, and merger & acquisition announcements may cause return anomalies, and hence we need to check that they do not drive the media effect. The media effect we found so far might be spurious if media coverage is more intense for firms announcing earnings for stocks with dividends announcements and stocks with merger and acquisition announcements. For instance, a company’s announcement of positive earnings will ignite investors’ investment passion, resulting in a short-term rise in stock price, while at the same time this announcement may attract more media coverage. In this circumstance, we will find a spurious relationship between media coverage and stock returns. Similarly, both dividends announcements and merger & acquisition announcements may also cause a spurious effect of media coverage on stock returns.

To ensure that our results are not driven by the return anomalies during the periods of postearnings announcement, dividends announcements, and merger and acquisition announcements, we exclude the observations in the month of these events as well as those in the following one month after these events. We collect the data of earnings announcements from the RESSET database, while both the dividends data and merger & acquisition data are from the CSMAR database.

Table 5. Robustness tests: eliminating special events’ impact.

	DGTW [t+1]	DGTW [t+2]	DGTW [t+3, t+12]	Stkret [t+1]	Stkret [t+2]	Stkret [t+3, t+12]
Panel A: Excluding stocks with Earnings Announcements
Media	0.124 **	0.213 ***	1.296 ***	0.153 **	0.307 ***	1.140 ***
	(2.07)	(3.75)	(5.61)	(2.28)	(4.72)	(4.25)
Asset	0.042	−0.028	−0.650 ***	0.021	−0.101	−0.929 **
	(0.67)	(−0.44)	(−3.77)	(0.21)	(−0.96)	(−2.44)
BM	−0.174	−0.100	−0.225	0.117	0.235 **	3.358 ***
	(−1.22)	(−0.72)	(−0.52)	(0.99)	(2.00)	(11.57)
⊿Turnover	−0.497 ***	−0.240 ***	−0.067	−0.743 ***	−0.261 **	0.043
	(−5.36)	(−2.67)	(−0.26)	(−6.17)	(−2.53)	(0.16)
Analyst	0.363 ***	0.321 ***	2.042 ***	0.462 ***	0.356 ***	1.767 ***
	(2.70)	(3.17)	(5.56)	(3.50)	(2.86)	(4.17)
Cprice	−0.225	−0.180	−0.877	−0.561 *	−0.315	−2.197 **
	(−0.96)	(−0.78)	(−1.24)	(−1.73)	(−0.98)	(−2.44)
Pret12	0.016	0.022	−0.200 **	0.043	0.008	−0.243 **
	(0.55)	(0.77)	(−2.28)	(1.14)	(0.20)	(−2.33)
Amihud’s Liquidity	18.316 ***	8.043 **	12.415	18.882 ***	12.242 ***	38.175 ***
	(3.80)	(2.47)	(1.54)	(3.11)	(2.91)	(3.92)
Volatility	−0.294 ***	−0.147	−1.085 ***	−0.296 **	−0.130	−0.937 ***
	(−2.88)	(−1.50)	(−4.35)	(−2.12)	(−1.01)	(−3.12)
Constant	0.286	1.371	19.962 ***	2.760	4.370	50.660 ***
	(0.19)	(0.86)	(3.85)	(0.98)	(1.46)	(5.21)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	111,232	109,936	94,679	111,232	109,936	94,679
R2	0.126	0.123	0.131	0.201	0.187	0.197
Panel B: Excluding Stocks with Dividends Announcements
Media	0.107 *	0.208 ***	1.416 ***	0.144 **	0.265 ***	1.349 ***
	(1.76)	(3.86)	(5.75)	(2.13)	(4.12)	(4.78)
Asset	0.030	−0.042	−0.598 ***	0.018	−0.110	−0.784 **
	(0.49)	(−0.67)	(−3.62)	(0.17)	(−1.02)	(−2.03)
BM	−0.149	−0.046	−0.508	0.149	0.289 **	2.951 ***
	(−1.07)	(−0.34)	(−1.27)	(1.24)	(2.52)	(10.39)
⊿Turnover	−0.495 ***	−0.170 *	0.260	−0.775 ***	−0.215 **	0.369
	(−4.85)	(−1.84)	(0.91)	(−6.12)	(−2.01)	(1.33)
Analyst	0.417 ***	0.343 ***	2.144 ***	0.487 ***	0.418 ***	1.820 ***
	(3.30)	(3.45)	(5.56)	(4.09)	(3.47)	(4.10)
Cprice	−0.154	−0.093	−0.980	−0.479	−0.247	−2.479 ***
	(−0.67)	(−0.41)	(−1.29)	(−1.49)	(−0.76)	(−2.62)
Pret12	0.030	0.035	−0.153 *	0.058	0.031	−0.169
	(1.04)	(1.23)	(−1.72)	(1.50)	(0.79)	(−1.60)
Amihud’s Liquidity	17.529 ***	8.654 ***	17.065 **	19.311 ***	12.936 ***	39.231 ***
	(3.66)	(2.98)	(2.27)	(3.24)	(3.37)	(4.54)
Volatility	−0.296 ***	−0.092	−1.051 ***	−0.296 **	−0.057	−0.836 ***
	(−2.90)	(−0.91)	(−4.01)	(−2.09)	(−0.43)	(−2.79)
Constant	0.289	0.319	16.455 ***	2.736	3.777	43.876 ***
	(0.18)	(0.21)	(3.32)	(0.91)	(1.25)	(4.82)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	111,837	110,194	94,035	111,837	110,194	94,035
R2	0.127	0.124	0.131	0.202	0.188	0.194
Panel C: Excluding Stocks with Merger and Acquisition Announcements
Media	0.107	0.194 ***	1.284 ***	0.154 **	0.288 ***	1.216 ***
	(1.60)	(3.25)	(5.74)	(2.08)	(4.12)	(4.76)
Asset	0.017	−0.030	−0.540 ***	0.004	−0.109	−0.682 *
	(0.26)	(−0.48)	(−3.14)	(0.04)	(−1.02)	(−1.71)
BM	−0.202	−0.139	−0.643	0.077	0.181	2.774 ***
	(−1.47)	(−1.02)	(−1.57)	(0.65)	(1.55)	(9.15)
⊿Turnover	−0.501 ***	−0.243 ***	−0.025	−0.772 ***	−0.268 ***	0.032
	(−4.78)	(−2.72)	(−0.09)	(−5.82)	(−2.68)	(0.12)
Analyst	0.385 ***	0.430 ***	2.234 ***	0.485 ***	0.479 ***	1.973 ***
	(3.16)	(4.48)	(5.68)	(3.95)	(3.77)	(4.59)
Cprice	−0.220	−0.167	−0.757	−0.543 *	−0.312	−2.187 **
	(−0.95)	(−0.73)	(−1.04)	(−1.68)	(−0.97)	(−2.39)
Pret12	0.038	0.061 **	−0.063	0.071 *	0.053	−0.091
	(1.35)	(2.08)	(−0.67)	(1.83)	(1.31)	(−0.80)
Amihud’s Liquidity	17.560 ***	8.003 **	23.291 **	18.001 ***	12.505 ***	47.976 ***
	(3.70)	(2.24)	(2.55)	(3.05)	(2.97)	(4.45)
Volatility	−0.320 ***	−0.110	−1.224 ***	−0.308 **	−0.082	−0.939 ***
	(−3.18)	(−1.08)	(−5.06)	(−2.22)	(−0.62)	(−3.19)
Constant	0.767	0.929	18.841 ***	3.005	4.570	45.840 ***
	(0.51)	(0.56)	(3.60)	(1.06)	(1.51)	(4.65)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	103,588	102,226	87,769	103,588	102,226	87,769
R2	0.130	0.127	0.135	0.207	0.194	0.201
Panel D: Applying All Three Filters in Panels A–C
Media	0.136 *	0.269 ***	1.299 ***	0.176 **	0.350 ***	1.216 ***
	(1.70)	(3.81)	(4.75)	(2.08)	(4.21)	(3.77)
Asset	0.024	0.007	−0.399 **	0.008	−0.077	−0.622
	(0.38)	(0.09)	(−2.05)	(0.07)	(−0.68)	(−1.52)
BM	−0.222	−0.151	−0.686	0.085	0.187	2.763 ***
	(−1.54)	(−1.02)	(−1.54)	(0.66)	(1.49)	(8.69)
⊿Turnover	−0.514 ***	−0.167	0.391	−0.804 ***	−0.184	0.352
	(−4.68)	(−1.60)	(1.18)	(−6.14)	(−1.52)	(1.13)
Analyst	0.341 **	0.284 ***	1.810 ***	0.450 ***	0.360 ***	1.620 ***
	(2.58)	(2.71)	(3.80)	(3.48)	(2.83)	(3.10)
Cprice	−0.285	−0.160	−1.238	−0.593 *	−0.267	−2.870 ***
	(−1.18)	(−0.68)	(−1.65)	(−1.78)	(−0.81)	(−3.10)
Pret12	0.016	0.035	−0.145	0.050	0.024	−0.189
	(0.53)	(1.18)	(−1.48)	(1.22)	(0.60)	(−1.66)
Amihud’s Liquidity	15.445 ***	7.409 *	21.849 *	15.135 ***	11.882 **	46.427 ***
	(3.32)	(1.89)	(1.92)	(2.63)	(2.61)	(3.61)
Volatility	−0.337 ***	−0.110	−1.189 ***	−0.355 **	−0.100	−0.992 ***
	(−3.09)	(−1.01)	(−3.99)	(−2.41)	(−0.72)	(−2.87)
Constant	0.297	0.457	12.569 **	2.909	4.690	38.188 ***
	(0.18)	(0.28)	(2.25)	(1.08)	(1.51)	(3.92)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	77,942	77,076	66,643	77,942	77,076	66,643
R2	0.142	0.139	0.147	0.217	0.203	0.211

We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 5. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

below gives information on the results of the robustness tests; be it by earnings announcements, dividends announcements, or merger or acquisition announcements, the positive effect of media coverage on stock returns is not driven.

5.2. Eliminating the Effect of the Financial Crisis in 2008

In 2007, the Shanghai Stock Exchange (SSE) composite index witnessed a noticeable rise from 2715.72 points at the beginning of the year to 6092.06 points on 16 October 2007. Due to the shock of the financial crisis in 2008, the SSE composite index had plunged dramatically to 1706.70 points by 14 November 2008. Therefore, the period between 2007 and 2008 experienced a significant fluctuation that we are concerned about the influence of the particular period on our results. To address this concern, we drop observations in the period from 2007 to 2008.

Table 6. Robustness tests: eliminating the effect of the financial crisis in 2008.

	DGTW [t+1]	DGTW [t+2]	DGTW [t+3, t+12]	Stkret [t+1]	Stkret [t+2]	Stkret [t+3,t+12]
Media	0.149 **	0.195 ***	1.332 ***	0.202 ***	0.273 ***	1.323 ***
	(2.38)	(3.29)	(5.22)	(2.79)	(4.00)	(4.41)
Asset	−0.013	−0.041	−0.829 ***	0.012	−0.070	−0.198
	(−0.22)	(−0.70)	(−4.73)	(0.11)	(−0.64)	(−0.45)
BM	−0.079	−0.057	0.486	0.175	0.258 **	3.551 ***
	(−0.66)	(−0.48)	(1.23)	(1.51)	(2.28)	(11.21)
⊿Turnover	−0.463 ***	−0.206 **	0.157	−0.682 ***	−0.259 ***	0.204
	(−4.83)	(−2.60)	(0.55)	(−5.29)	(−2.77)	(0.71)
Analyst	0.387 **	0.364 ***	2.705 ***	0.469 ***	0.468 ***	2.506 ***
	(2.57)	(3.28)	(6.69)	(3.24)	(3.26)	(5.36)
Cprice	−0.075	−0.100	−1.057	−0.290	−0.183	−1.991 *
	(−0.39)	(−0.47)	(−1.23)	(−1.01)	(−0.60)	(−1.79)
Pret12	0.036	0.065 *	0.069	0.076 *	0.069	0.124
	(1.12)	(1.93)	(0.74)	(1.68)	(1.47)	(1.18)
Amihud’s Liquidity	18.525 ***	9.986 ***	13.915 **	18.019 ***	13.282 ***	35.233 ***
	(3.46)	(2.96)	(2.00)	(2.79)	(2.96)	(4.02)
Volatility	−0.301 ***	−0.090	−1.281 ***	−0.325 **	−0.069	−1.285 ***
	(−2.88)	(−0.89)	(−5.51)	(−2.12)	(−0.50)	(−4.81)
Constant	0.678	1.130	29.807 ***	1.507	3.231	38.229 ***
	(0.48)	(0.83)	(6.95)	(0.49)	(1.07)	(3.38)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	104,780	103,035	85,306	104,780	103,035	85,306
R2	0.117	0.115	0.116	0.190	0.179	0.181

We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 6. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

illustrates the reexamined results that are consistent with the baseline results.

6. Further Explanation and Exploration

To further confirm that the positive effect of media coverage on stock returns in China’s stock market is consistent with Barber and Odean’s [5] “attention-driven buying” hypothesis, we need to further test the following three issues. First, whether stocks with higher media coverage at the current month have persistently higher investor attention (ATT) in the following months. Second, whether investor attention (ATT) has a positive effect on stock returns. Third, whether the impact of media coverage will significantly decrease after controlling for the effect of investor attention (ATT).

In

Table 7. Media Coverage and investor attention.

	ATT [t+1]	ATT [t+2]	ATT [t+3]	ATT [t+4]	ATT [t+5]	ATT [t+6]	ATT [t+12]
Media	0.432 ***	0.432 ***	0.430 ***	0.431 ***	0.425 ***	0.419 ***	0.375 ***
	(13.59)	(14.17)	(14.12)	(14.43)	(14.10)	(14.08)	(11.76)
Asset	0.384 ***	0.475 ***	0.422 ***	0.402 ***	0.401 ***	0.431 ***	0.465 ***
	(19.62)	(4.98)	(10.42)	(15.75)	(13.21)	(6.84)	(4.16)
BM	0.068 *	0.068 *	0.074 **	0.077 **	0.080 **	0.082 **	0.087 **
	(1.91)	(1.86)	(2.01)	(2.08)	(2.22)	(2.29)	(2.28)
⊿Turnover	−0.052 **	−0.062 **	−0.069 **	−0.052 *	−0.048 *	−0.053 **	−0.049 *
	(−2.10)	(−2.37)	(−2.49)	(−1.90)	(−1.79)	(−2.14)	(−1.84)
Analyst	0.026	0.027	0.033 *	0.028	0.028	0.028	0.022
	(1.47)	(1.54)	(1.86)	(1.60)	(1.50)	(1.38)	(0.95)
Cprice	0.131 ***	0.134 ***	0.143 ***	0.149 ***	0.159 ***	0.162 ***	0.193 ***
	(3.24)	(3.30)	(3.51)	(3.74)	(4.13)	(4.18)	(4.62)
Pret12	−0.040	−0.017	−0.024 **	−0.024 **	−0.036	−0.048	−0.045
	(−1.08)	(−0.81)	(−2.38)	(−2.04)	(−1.49)	(−1.29)	(−1.07)
Amihud’s Liquidity	−12.943 ***	−12.518 ***	−12.467 ***	−12.247 ***	−12.819 ***	−12.848 ***	−11.624 ***
	(−9.39)	(−11.21)	(−11.77)	(−10.66)	(−10.10)	(−9.65)	(−8.64)
Volatility	−0.013	−0.035	−0.032 *	−0.033 *	−0.045 **	−0.057 ***	−0.066 **
	(−0.74)	(−1.67)	(−1.84)	(−1.83)	(−2.50)	(−3.04)	(−2.57)
Constant	−2.402 ***	−4.233 **	−3.036 ***	−2.543 ***	−2.320 ***	−2.825 **	−3.786
	(−4.91)	(−2.08)	(−3.46)	(−4.12)	(−3.27)	(−2.11)	(−1.62)
Industry	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	61,573	60,170	58,801	57,457	56,120	54,794	47,093
R2	0.285	0.286	0.287	0.290	0.292	0.294	0.310

We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 7. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

, we test the effect of media coverage of investor attention after controlling for the effects of control variables in Table 4. The regression results show that stocks with higher media coverage at the current month have significantly higher investor attention in the following months persistently during a year.

Table 8. Regression analysis: media coverage, investor attention and stock returns.

	DGTW [t+1]	DGTW [t+2]	DGTW [t+3, t+12]	Stkret [t+1]	Stkret [t+2]	Stkret [t+3, t+12]
Media	−0.101	−0.022	0.439	−0.124	0.025	−0.140
	(−1.19)	(−0.27)	(0.94)	(−1.23)	(0.28)	(−0.30)
ATT [t+1]	0.346 ***			0.380 ***
	(5.08)			(5.27)
ATT [t+2]		0.344 ***			0.365 ***
		(4.74)			(4.71)
ATT [t+3, t+12]			1.762 *			1.563 *
			(1.91)			(1.96)
Asset	0.299	−0.144	−1.102 ***	0.214	−0.402 **	−4.481 ***
	(1.14)	(−1.44)	(−2.78)	(0.51)	(−2.23)	(−7.24)
BM	−0.245	−0.061	0.007	0.092	0.344 *	3.640 ***
	(−1.00)	(−0.25)	(0.01)	(0.48)	(1.78)	(8.22)
⊿Turnover	−0.528 ***	−0.250 *	−0.453	−0.841 ***	−0.309 *	−0.157
	(−3.47)	(−1.74)	(−1.02)	(−4.56)	(−1.93)	(−0.38)
Analyst	0.313 ***	0.339 ***	1.205 ***	0.307 ***	0.254 **	0.480
	(3.49)	(3.30)	(3.62)	(3.04)	(2.20)	(1.42)
Cprice	−0.368	−0.265	0.207	−0.876 *	−0.591	−3.257 **
	(−1.02)	(−0.71)	(0.21)	(−1.80)	(−1.14)	(−2.35)
Pret12	0.014	0.020	−0.446 ***	0.023	−0.033	−0.637 ***
	(0.40)	(0.39)	(−2.83)	(0.49)	(−0.52)	(−3.90)
Amihud’s Liquidity	29.700 ***	16.856 **	47.899 ***	30.871 ***	26.798 ***	96.804 ***
	(3.39)	(2.55)	(2.79)	(2.67)	(3.12)	(5.42)
Volatility	−0.197	−0.228	−1.092 *	−0.301	−0.224	−0.693
	(−1.27)	(−1.43)	(−1.69)	(−1.57)	(−1.16)	(−1.13)
Constant	−10.663*	1.295	8.790	−4.281	11.234 **	125.202 ***
	(−1.85)	(0.54)	(0.96)	(−0.46)	(2.10)	(12.53)
Industry	YES	YES	YES	YES	YES	YES
Obs.	61,334	59,809	45,398	61,334	59,809	45,398
R2	0.193	0.165	0.181	0.264	0.232	0.263

We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 8. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

further suggests that investor attention has a positive and significant effect on stock returns. This result is consistent with either Da, Engelberg, and Gao [22] or Ying, Kong and Luo [31], arguing that higher investor attention can generate a greater buying pressure and thus raise stock prices and the short-term holding period returns. Meanwhile, we find that the media effect is no longer significant after controlling for the measure of investor attention, which is consistent with our earlier hypothesis that the investor attention is the intermediate channel and the key mechanism supporting the positive media effect on stock returns in China’s stock market. Therefore, we can conclude that relationship between media coverage and stock returns in Chinese stock market is consistent with the “attention-driven buying” theory advanced by Barber and Odean [5], but contradicts with the “risk compensation” theory proposed by Fang and Peress [3].

According to the existing literature [30,31,32], growth stocks with a higher price to equity ratio tend to have attracted more investor attention. To verify the validity of “attention-driven buying” theory to explain our main empirical results, we further test how the effect of media coverage on stock returns differ in stocks with different implied growth level measured by price to equity ratio. Consistent with our expectation, the regression results in

Table 9. Additional test: media coverage, implied growth and stock return.

	DGTW [t+1]	DGTW [t+2]	DGTW [t+3, t+12]	Stkret [t+1]	Stkret [t+2]	Stkret [t+3, t+12]
Media	−0.281 **	−0.066	0.095	−0.418 ***	−0.034	−1.349 ***
	(−2.47)	(−0.60)	(0.21)	(−3.33)	(−0.29)	(−2.79)
Media*Growth	0.144 ***	0.101 ***	0.528 ***	0.194 ***	0.112 ***	1.021 ***
	(3.46)	(2.62)	(3.84)	(4.42)	(2.71)	(6.37)
Growth	0.869 ***	0.761 ***	2.624 ***	0.962 ***	0.853 ***	2.556 ***
	(12.53)	(11.26)	(9.63)	(11.68)	(10.45)	(8.18)
Asset	0.011	−0.054	−0.743 ***	−0.009	−0.139	−1.004 ***
	(0.18)	−0.86)	(−4.98)	(−0.09)	(−1.34)	(−2.77)
BM	3.181 ***	2.822 ***	9.473 ***	3.879 ***	3.513 ***	13.201 ***
	(11.13)	(9.34)	(9.04)	(11.43)	(10.67)	(11.62)
⊿Turnover	−0.514 ***	−0.250 ***	−0.211	−0.775 ***	−0.289 ***	−0.135
	(−5.84)	(−2.83)	(−0.83)	(−6.83)	(−2.94)	(−0.54)
Analyst	0.364 ***	0.329 ***	2.062 ***	0.437 ***	0.372 ***	1.743 ***
	(2.78)	(3.46)	(6.39)	(3.50)	(3.20)	(4.66)
Cprice	−0.204	−0.146	−0.554	−0.533 *	−0.276	−1.728 *
	(−0.91)	(−0.64)	(−0.79)	(−1.68)	(−0.86)	(−1.91)
Pret12	−0.077 ***	−0.030	−0.373 ***	−0.062	−0.048	−0.419 ***
	(−2.73)	(−1.06)	(−4.27)	(−1.65)	(−1.28)	(−4.28)
Amihud’s Liquidity	13.059 ***	4.992 *	5.501	13.993 **	8.654 **	29.237 ***
	(2.90)	(1.71)	(0.77)	(2.35)	(2.41)	(3.44)
Volatility	−0.338 ***	−0.171 *	−1.130 ***	−0.338 **	−0.148	−0.872 ***
	(−3.49)	(−1.72)	(−4.48)	(−2.46)	(−1.15)	(−3.01)
Constant	0.720	2.773 *	20.523 ***	3.238	5.866 **	49.525 ***
	(0.47)	(1.75)	(4.33)	(1.10)	(2.02)	(5.25)
Industry	Yes	Yes	Yes	Yes	Yes	Yes
Obs.	128,846	127,030	108,569	128,846	127,030	108,569
R2	0.142	0.137	0.150	0.220	0.204	0.218

Growth represents the implied growth measured by price to equity ratio in the current month. We employ the cross-sectional regression method proposed by Fama and MacBeth (1973) in Table 4. T-values are in parentheses. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% level, respectively.

show that the positive affect of current month’s media coverage on stock returns in the following months significantly increase with the level of price to equity ratio.

7. Conclusions

Using data of news reports from China’s major newspapers during the period of 2001 to 2011, we ran cross-sectional regressions to reexamine the relationship between media coverage and stock returns. The results show that in China’s stock market stocks with higher media coverage have higher holding period returns than those with lower media coverage at least in the following year. This result is robust after excluding the observations with events of earnings announcements, dividends announcements and merger and acquisition announcements, as well as after excluding the observations in the financial crisis from 2007 to 2008. To some extent, our findings are consistent with the “attention-driven buying” theory advocated by Barber and Odean [5], but contradict the “risk compensation” theory suggested by Fang and Peress [3]. Particularly, unlike the findings that media coverage can push up the prices of stocks in the short terms, but such pressure subsequently reverses, we find stocks with higher media coverage have higher holding period returns than those with lower media coverage at least in the following one year in China A-shares market dominated by individual investors.

We also find that stocks with higher media coverage at the current month have significantly higher investor attention in the following year than those with lower media coverage. At the same time, investor attention has a direct and significant positive impact on stock returns, while the media effect is no longer significant after controlling for investor attention. These results indicate that in China’s stock market, the buying pressure effect of investor attention is the intermediate channel and the key mechanism causing the positive relationship between media coverage and stock returns.

Our impressive results using Chinese market data show that the effect of media coverage on stock returns depends on the type of investors. Media coverage can have a significant and positive influence on stock returns in these markets dominated by individual investors or immature investors, while there is an opposite effect in mature markets or markets dominated by institutional investors.

The limitations of this paper lie in at least two aspects. Firstly, due to the data availability of investor attention, we had to limit our sample up to 2011. After 2011, there might be significant changes in the structure and maturity of the investors in the Chinese stock market. How this change will affect the relation between media coverage and stock returns might be an interesting topic for future research. Secondly, besides the investor attention channel, there might be other channels for media coverage to affect stock returns. We also leave the tests on other transmission channels for future research.