- Email: [email protected]
Investor sentiment and stock returns: Evidence from provincial TV audience rating in China
Physica A 466 (2017) 288–294
Contents lists available at ScienceDirect
Physica A journal homepage: www.elsevier.com/locate/physa
Investor sentiment and stock returns: Evidence from provincial TV audience rating in China Yongjie Zhang a,b , Yuzhao Zhang a , Dehua Shen a,b,∗ , Wei Zhang a,c a
College of Management and Economics, Tianjin University, Tianjin, 300072, PR China
b
China Center for Social Computing and Analytics (CCSCA), Tianjin University, Tianjin, 300072, PR China
c
Key Laboratory of Computation and Analytics of Complex Management Systems (CACMS), Tianjin, 300072, PR China
highlights • • • •
Provincial investor sentiment is positively related to stock returns. The provincial correlation coefficient is larger than the cross-provincial correlation coefficient. There exists home bias in Chinese stock market. Provincial investor sentiment can explain the provincial comovement.
article
info
Article history: Received 12 May 2016 Received in revised form 24 August 2016 Available online 24 September 2016 Keywords: TV audience rating Provincial investor sentiment Home bias Provincial comovement Internet information
abstract In this paper, we advocate the provincial TV audience rating as the novel proxy for the provincial investor sentiment (PIS) and investigate its relation with stock returns. The empirical results firstly show that the PIS is positively related to stock returns. Secondly, we provide direct evidence on the existence of home bias in China by observing that the provincial correlation coefficient is significantly larger than the cross-provincial correlation coefficient. Finally, the PIS can explain a large proportion of provincial comovement. To sum up, all these findings support the role of the non-traditional information sources in understanding the ‘‘anomalies’’ in stock market. © 2016 Elsevier B.V. All rights reserved.
1. Introduction The role that investor sentiment plays in asset pricing has been highlighted for a long history in financial economics. The earliest discussion could broadly trace back to Keynes’s ‘‘animal spirits’’, which claims that prices move in a way unrelated to fundamentals [1]. The academic enthusiasm has been aroused by Refs. [2–6], who clearly illustrate that irrational noise traders (with erroneous psychological beliefs and diverse biases) could not be offset by limited arbitrageurs and therefore they have material impact on stock prices. To empirically investigate this issue, various investor sentiment proxies have been put forward to investigate the contemporaneous correlations between investor sentiment and market-wide variables e.g., stock returns, volatility and liquidity as well as time series predictive power for these variables. Among these, two distinct streams of literature stand out. The first stream refers to the continuous observations of overly forecasts of cash flow or investors’ state of mind revealed by survey on individual investor or consumer confidence [7–12], sentient extracted from the online message boards [13–16] as well as the market-wide variables, including net mutual
∗
Corresponding author at: College of Management and Economics, Tianjin University, Tianjin, 300072, PR China. E-mail address: [email protected] (D. Shen).
http://dx.doi.org/10.1016/j.physa.2016.09.043 0378-4371/© 2016 Elsevier B.V. All rights reserved.
Y. Zhang et al. / Physica A 466 (2017) 288–294
289
fund redemptions, dividend premium, bid–ask spreads and average first-day returns on IPOs [17–20]. The rationale of these studies is that some proxies are essentially correlated to investor sentiment or served as the reflections of investor behavior. For example, the linguistic analysis of the online postings in message boards could be expounded as the individual investor sentiment towards certain firms. However, this stream of literature has the disadvantage of endogenous problem. Because the online postings and changes in market variables, e.g., dividend premium and bid–ask spreads, could be the results of the trading activities. It is inappropriate to ‘‘test a theory that is about from inputs to outputs with an output measure’’ as claimed by Ref. [21]. Besides, the participation rate of the survey is not sufficient enough and participants have little incentives to response to survey questions truthfully. The second stream of the literature mainly relies on some sentiment-altering events, which can lead to sentiment changes in a large number of investor and thus influence the asset prices. Therefore this stream of sentiment proxies is discrete variable, including the game results of sport event [22,23], TV series finales [24] as well as the aviation disasters [25]. This stream of sentiment proxies outperforms the first in the sense that it avoids the endogenous problem as well as provides more objective emotions. Admittedly, these discrete proxies only allow us to investigate the price changes around the special events, but fail to provide further illustrations on time series anomalies. In this paper, we aim to bridge the gap between the above-mentioned two streams of literature by advocating the provincial TV audience rating as the novel proxy for the provincial investor sentiment (hereafter, PIS). This PIS proxy has the advantage of coming from a large number of potential investor, avoiding the endogenous problem as well as serving as continuous observations. The rationales of this interpretation are in the following three aspects. Firstly, drawing on the literature on economic psychology, sentiment can affect financial decision-making through determining the risk perception as well as the information processing behavior of investors when forming expectations [26,27]. Secondly, given the high coverage rate, the TV audience rating represents the sentiment from hundreds of millions of the TV watcher. Besides, as is shown in Ref. [24], TV series final can eventually affect investors’ demand for risky assets through connecting with reviewer’s emotions. Thirdly, Internet information has been extensively employed in econophysics [28–40]. In particular, the utilization of search engine [31–33,35,38,39], online news [29,37,40] as well as the microblogging [36] has profoundly reshaped our understanding of complex financial economic systems. With this novel PIS proxy, we contribute to the existing literature in two aspects. On one hand, we can directly measure the correlation coefficients between PIS and its corresponding as well as stock returns of others provincial. The distinct differences in the correlation coefficients provide direct evidence on the existence of home bias in Chinese stock market. On the other hand, the PIS can explain a large proportion of provincial comovement and the results are robust to alternative proxies of PIS across different models. The rest of this paper is organized as follows. We put forward the hypotheses in Section 2. Section 3 describes the data. Section 4 performs the empirical analysis and illustrates the results. Robustness test is given in Section 5. Section 6 concludes. 2. Hypotheses The daily PIS proxy allows us to directly examine the behavioral predictions on the contemporaneous relations between investor sentiment and stock returns. Existing literature based on the proxies constructed by Internet information has documented a positively contemporaneous relation to stock returns [15,16], showing that the highest (lowest) sentiment corresponds to positive (negative) stock returns in the market index. In that sense, we put forward the most straightforward hypothesis on the positively contemporaneous relations between PIS and corresponding stock returns. Hypothesis 1. The PIS has a positively contemporaneous relation to stock returns. The home bias phenomenon has long been one of the most intriguing puzzles in financial economics, which could be understood as investors construct their portfolios with disproportionate amount of assets on local (domestic) equities in spite of the benefits of diversifications [41–43]. Besides, several findings emphasize the crucial role of geographic location in financial decision-making and asset prices [44–47]. In that sense, if there exists home bias puzzle in Chinese stock market, the correlation coefficient between PIS and corresponding stock returns (provincial correlation coefficient) should be larger than the correlation coefficient between PIS and stock returns of other provinces (cross-provincial correlation coefficient). For the same reason, there should be no significant differences in provincial correlation coefficient and cross-provincial correlation coefficient if there is no home bias puzzle in Chinese stock market. The following two alternative hypotheses are naturally derived. Hypothesis 2a. The provincial correlation coefficient is larger than the cross-provincial correlation coefficient. Hypothesis 2b. There are no significant differences in provincial correlation coefficient and cross-provincial correlation coefficient. In a recent work by Wongchoti and Wu [48], they show that stock returns exhibit provincial comovement after controlling for the market and industry effect. This location-based movement cannot be explained by fundamental factors and is pronounced in firm with large number of local investors [49]. In that sense, if PIS represents the sentiment from local investors, it should display some explanatory power for the observed provincial comovement. Therefore, we put forward the hypothesis on the explanation of PIS on provincial comovement. Hypothesis 3. The PIS can explain the provincial comovement.
290
Y. Zhang et al. / Physica A 466 (2017) 288–294 Table 1 Provinces, municipalities and firms. This table reports the average number of firms during the sample period as the TV audience rating. Because some firms undergo trading halt, the average number is non-integer. The municipalities are Beijing, Shanghai, Tianjin and Chongqing. Provinces (Municipalities)
Firms
Provinces (Municipalities)
Firms
Anhui Beijing Fujian Guangdong Guangxi Guizhou Hebei Henan Heilongjiang Hubei Hunan
76.42 207.10 83.14 369.36 27.77 19.19 46.75 63.46 30.46 80.26 66.59
Jilin Jiangsu Jiangxi Liaoning Shandong Shanghai Sichuan Tianjin Yunnan Zhejiang Chongqing
36.21 228.12 31.11 66.61 147.98 223.97 84.16 36.67 26.19 239.11 35.22
3. Data description We combine several datasets in this study. The capital data comes from the China Stock Market & Accounting Research Database (CSMAR Database), including individual stock returns after the reimbursement, total capitalization, industry classification codes and market returns (Shanghai Shenzhen CSI 300 Index, Shanghai Composite Index, Shenzhen Composite Index and Shanghai Shenzhen Composite Index). The audience rating data is from CSM Media Research (http://www.csm.com.cn/), in which TV audience measurement network provides data that represents the viewing of 1.27 billion people in China. Therefore, the audience rating data are reliable and representative in the sense of large size of the sample. Given the limited access to the database, we obtain the audience rating of 22 provinces and municipalities (Beijing, Shanghai, Tianjin and Chongqing) over the period from 1 April 2013 to 3 April 2014. The GDP data is from the National Bureau of Statistics of China (http://www.stats.gov.cn/). We use the yearly GDP in 2013 to calculate the GDP-based investor sentiment in Section 4. Besides, in order to calculate the abnormal returns, we extend the sample period of the capital data to 1 year before the beginning of the audience rating, i.e., from 1 April 2012 to 3 April 2014. Table 1 reports the provinces and the corresponding average number of the firms in the sample period as the audience rating. The average number of the firms is calculated as the daily firms without trading halt during the sample. All the firms account for more than 90% of the listed firms in both Shanghai and Shenzhen stock exchanges. Therefore, our sample is a good representation of Chinese stock market. 4. Empirical analysis and results We obtain the province code from the CSMAR Database and separate the stocks into different provinces. For each given trading day, we use the average return of all the listed firms in certain province as the mean of stock return for the province. In order to observe the differences in a clear-cut way, we classify the investor sentiment into quartiles from the lowest to the highest days, i.e., the lowest-subgroup, the second-subgroup, the third-subgroup and the highest-subgroup. To avoid the spurious relations between TV audience rating and stock returns, e.g., the weekend effect, we do not use the raw value. In particular, we regress the raw TV audience rating on weekday dummies and keep the residuals as the PIS for a certain province. Fig. 1 gives an illustration of the Jiangxi province. As is shown, the PIS in Jiangxi province is significantly different from its raw TV audience rating. Besides, following [50], we use three different models to calculate the abnormal stock returns for all the provinces. Table 2 reports the results of the (abnormal) stock returns in the highest and lowest-subgroups. We find that the mean of the stock returns and the abnormal returns in the highest subgroup are significant larger than that in the lowest subgroup at 1% level. These results suggest that the PIS is positively related to stock returns, which is in accord with the sentiment extracted from Yahoo! Finance message board and Twitter [14,51]. Abnormal return 1 (Mean adjusted return) ARi,t = Ri,t − RI RI =
1
−26
100 t =−125
(1) Ri,t
(2)
where Ri,t is the daily return for province i at time t. Daily return for a certain province is calculated as the average return of all the listed firms in the province excluding the firms with trading halt; RI is the simple average of Ri,t of province i in the [−125, −26] estimation period. Abnormal return 2 (Market adjusted return) ARi,t = Ri,t − Rm,t where Rm,t is the return on equally weighted daily return at time t.
(3)
Y. Zhang et al. / Physica A 466 (2017) 288–294
291
Panel A: TV audience rating.
Panel B: PIS. Fig. 1. An illustration of TV audience rating and PIS. Table 2 PIS and stock returns. This table reports the PIS and stock return in quartiles. We calculate mean of the corresponding (abnormal) stock returns in the lowest and highest-subgroups, respectively. We perform the two-sample t-test to detect the significant differences between highest and lowest-subgroups. The t-values in the parentheses denote the significances. Subgroups
Mean of return
Abnormal return 1
Abnormal return 2
Abnormal return 3
Lowest Highest
−0.0002
−0.0016
0.0017 0.0019*** (t-value = 11.88)
0.0002 0.0018*** (t-value = 10.98)
0.0006 0.0018 0.0012*** (t-value = 18.22)
0.0008 0.0019 0.0011*** (t-value = 16.76)
Highest–Lowest ***
Denotes the significant at 1% level.
Abnormal return 3 (OLS market model) ARi,t = Ri,t − αi − βi Rm,t
(4)
where αi and βi are estimated parameters from the CAPM model with the moving estimation window of [−125, −26]. We further calculate the provincial and cross-provincial correlation coefficients with the Kendall correlation analysis [51]. The Kendall correlation analysis has the advantage of not making assumptions about the probability distributions of the variables. To depict the time-varying characteristic of the correlation coefficients, we employ the rolling windows estimation method to calculate the Kendall correlation coefficients. Admitted, the major disadvantage of this method is that the choice of the window length is arbitrary. In the first analysis, we employ the window length of 50 and it gives 193 observations. Fig. 2 illustrates that the provincial correlation coefficient (0.0822) is significant larger than that of the cross-provincial correlation coefficient (0.0350) at 1% level with t-value = 28.97 and p-value = 0.0000.1 The larger provincial correlation coefficient suggests that there exists home bias in Chinese stock market. To empirically investigate the explanatory power of the PIS on provincial comovement, we need to calculate the provincial comovement controlling for market returns and industry returns [48]. The following regression models are employed: Ri,t = α + β1 RP ,t + β2 RM ,t + β3 RIND,t + εi,t
(5)
where Ri,t denotes the daily return of stock i on date t , RP ,t denotes the equally-weighted daily return in the corresponding province, RM ,t denotes the daily return on the market portfolio, RIND,t denotes the corresponding daily return on the industry that stock i belongs to. In each regression model, stock i is excluded in the construction of the provincial stock index (RP ,t ) to avoid the spurious correlation. The null hypothesis that there is no provincial comovement implies β1 = 0. For robustness test, we construct four alternative models with different measurements of RM ,t . Models (from (1) to (4)) in Table 3 represent
1 In fact, we also perform the Pearson and Spearman correlation analysis. They all show that the provincial correlation coefficients are significantly larger than the cross-provincial correlation coefficients.
292
Y. Zhang et al. / Physica A 466 (2017) 288–294
Fig. 2. Provincial and cross-provincial Kendall correlation coefficients. Table 3 Explain the provincial comovement. This table reports the explanatory power of the weighted PISs on the provincial comovement based on model (6). In the cross-sectional analysis, the PISs exhibit strong explanatory power for the provincial comovement across all models with alternative proxies. Values in the parentheses are the corresponding p-values of the t-statistics. Model
(1)
(2)
(3)
(4)
7.4382 2.5805** (0.0179) 21.23%
6.3655 1.9827* (0.0613) 12.24%
12.92 3.0131*** (0.0069) 27.78%
4.9262 2.9354*** (0.0082) 26.62%
4.8664 2.9174*** (0.0085) 26.34%
5.4255 3.161*** (0.0049) 29.99%
7.8085 3.0553*** (0.0062) 28.41%
9.237 3.007*** (0.0070) 27.69%
9.231 3.0385*** (0.0065) 28.16%
9.717 3.0273*** (0.0067) 28.00%
1.5562 4.4393*** (0.0026) 34.02%
Panel A: Capitalization-based PIS β 7.1027 2.4040** t-statistics (0.0260) Adjusted R2 18.54% Panel B: GDP-based PIS
β
t-statistics Adjusted R2 Panel C: Firm-based PIS
β
t-statistics Adjusted R2 * ** ***
Denotes the significant at 10% level. Denotes the significant at 5% level. Denotes the significant at 1% level.
the models with Shanghai Shenzhen CSI 300 Index, Shanghai Composite Index, Shenzhen Composite Index and Shanghai Shenzhen Composite Index, respectively. pc m = α + β Sentiment m + εm
(6)
where pc m is the provincial comovement for province m, i.e., the mean of the β1 in certain province of model (5), Sentiment m represents the three different constructions of PIS, i.e., the capitalization-based PIS, the GDP-based PIS and the firm-based PIS. Specifically, for a certain province, we have the yearly GDP, number of the firms and their total capitalization. We calculate the product term of the average PIS during the sample period and the yearly GDP as the GDP-based PIS. Model (6) is the cross-sectional regression and all the variables pass through the Augmented Dickey–Fuller test confirming the stationary property. Table 3 reports our main results for Hypothesis 3 with the cross-sectional regression model (6). As is shown, all the PISs (Panel A, Panel B and Panel C) can explain large proportions of the provincial comovement. In particular, the firm-based PIS exhibits the best explanatory power. All these findings uphold Hypothesis 3. 5. Robustness Since we have employed three alternative models to calculate the abnormal returns for Hypothesis 1 and constructed three distinct measurements of PIS across four models for Hypothesis 3, we mainly focus on the robustness of Hypothesis 2
Y. Zhang et al. / Physica A 466 (2017) 288–294
293
Table 4 Correlation coefficients with alternative choices of window length. This table reports the provincial and cross-provincial correlation coefficients with alternative choices of window length. Observation is the number of the time-varying correlation coefficients. PCC denotes the provincial correlation coefficient and CCC denotes the cross-provincial correlation coefficient. PCC–CCC denotes the differences. We perform the two-sample t-test to detect the significant differences. Values in the parentheses are the corresponding pvalues of the t-statistics. Variables
Observation PCC CCC PCC–CCC ***
Alternative choices of window length 40
45
50
55
60
203 0.0922 0.0351 0.0571*** (0.0000)
198 0.0871 0.0353 0.0518*** (0.0000)
193 0.0822 0.0350 0.0472*** (0.0000)
188 0.0773 0.0347 0.0425*** (0.0000)
183 0.0724 0.0342 0.0382*** (0.0000)
Denotes the significant at 1% level.
in this section. In particular, we consider alternative choices of the window length to calculate the provincial and crossprovincial correlation coefficients. Table 4 reports the empirical results of the provincial and cross-provincial correlation coefficients with 40, 45, 50, 55 and 60 trading days of window length. As is shown, the provincial correlation coefficient is significant larger than the cross-provincial correlation coefficient at 1% level. All these results further prove that there exists home bias in Chinese stock market. 6. Conclusions In this paper, we advocate the provincial TV audience rating as the novel proxy for PIS and investigate its relation with stock returns. This novel investor sentiment proxy overcomes the endogenous problem and reveals the state of mind from hundreds of millions investors in China. By dividing the PIS into subgroups and observing the corresponding (abnormal) stock returns, we find that the PIS is positively related to stock returns. Besides, given the nature of PIS representing the local investor sentiment, we provide direct evidence on the existence of home bias. Finally, we show that the PIS can explain a large proportion of provincial comovement with the cross-sectional analysis. All these findings highlight the role of the non-traditional information sources in understanding the ‘‘anomalies’’ in stock market. However, the predictive ability of the PIS on stock returns is unveiled in this study. We leave this for future research. Acknowledgments We thank Tony He and Youwei Li for helpful comments and discussions during their visiting at Tianjin University. This work is supported by the National Natural Science Foundation of China (71271144, 71320107003 and 71532009) and Core Projects in Tianjin Education Bureaus Social Science Program (2011ZD008 and 2014ZD13). References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26]
J.M. Keynes, The General Theory of Employment, Interest and Money, Macmillan, London, 1936. F. Black, Noise, J. Finance 41 (1986) 529–543. H. Hong, J.C. Stein, A unified theory of underreaction, momentum trading, and overreaction in asset markets, J. Finance 54 (1999) 2143–2184. J.B. De Long, A. Shleifer, L.H. Summers, R.J. Waldmann, Noise trader risk in financial markets, J. Polit. Econ. 98 (1990) 703–738. K. Daniel, D. Hirshleifer, A. Subrahmanyam, Investor psychology and security market under- and overreactions, J. Finance 53 (1998) 1839–1885. N. Barberis, A. Shleifer, R. Vishny, A model of investor sentiment, J. Financ. Econ. 49 (1998) 307–343. G.W. Brown, M.T. Cliff, Investor sentiment and the near-term stock market, J. Empir. Finance 11 (2004) 1–27. G.W. Brown, M.T. Cliff, T. Investor, Sentiment and asset valuation, J. Bus. 78 (2005) 405–440. K.L. Fisher, M. Statman, Investor sentiment and stock returns, Financ. Anal. J. 56 (2000) 16–23. M. Lemmon, E. Portniaguina, Consumer confidence and asset prices: Some empirical evidence, Rev. Financ. Stud. 19 (2006) 1499–1529. M.E. Solt, M. Statman, How useful is the sentiment index? Financ. Anal. J. 44 (1988) 45–55. W.Y. Lee, C.X. Jiang, D.C. Indro, Stock market volatility, excess returns, and the role of investor sentiment, J. Banking Finance 26 (2002) 2277–2299. R. Tumarkin, R.F. Whitelaw, News or noise? Internet postings and stock prices, Financ. Anal. J. 57 (2001) 41–51. S.-H. Kim, D. Kim, Investor sentiment from Internet message postings and the predictability of stock returns, J. Econ. Behav. Organ. 107 (2014) 708–729. Part B. S.R. Das, M.Y. Chen, Yahoo! for Amazon: Sentiment extraction from small talk on the web, Manage. Sci. 53 (2007) 1375–1388. W. Antweiler, M.Z. Frank, Is all that talk just noise? The information content of Internet stock message boards, J. Finance 59 (2004) 1259–1294. M. Baker, J. Wurgler, Investor sentiment and the cross-section of stock returns, J. Finance 61 (2006) 1645–1680. M. Baker, J. Wurgler, Investor sentiment in the stock market, J. Econ. Perspect. 21 (2007) 129–152. M. Baker, J.C. Stein, Market liquidity as a sentiment indicator, J. Financ. Mark. 7 (2004) 271–299. R. Neal, S.M. Wheatley, Do measures of investor sentiment predict returns? J. Finan. Quant. Anal. 33 (1998) 523–547. L. Qiu, I. Welch, Investor sentiment measures, National Bureau of Economic Research, 2006. A. Edmans, D. GarcÍA, Ø. Norli, Sports sentiment and stock returns, J. Finance 62 (2007) 1967–1998. S.-C. Chang, S.-S. Chen, R.K. Chou, Y.-H. Lin, Local sports sentiment and returns of locally headquartered stocks: A firm-level analysis, J. Empir. Finance 19 (2012) 309–318. G.M. Lepori, Investor mood and demand for stocks: Evidence from popular TV series finales, J. Econ. Psychol. 48 (2015) 33–47. G. Kaplanski, H. Levy, Sentiment and stock prices: The case of aviation disasters, J. Financ. Econ. 95 (2010) 174–201. E.J. Johnson, A. Tversky, Affect, generalization, and the perception of risk, J. Pers. Soc. Psychol. 45 (1983) 20–31.
294
Y. Zhang et al. / Physica A 466 (2017) 288–294
[27] N. Schwarz, G.L. Clore, Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states, J. Pers. Soc. Psychol. 45 (1983) 513–523. [28] B. Edelman, Using Internet data for economic research, J. Econ. Perspect. 26 (2012) 189–206. [29] D. Shen, W. Zhang, X. Xiong, X. Li, Y. Zhang, Trading and non-trading period Internet information flow and intraday return volatility, Physica A 451 (2016) 519–524. [30] H.S. Moat, C. Curme, A. Avakian, D.Y. Kenett, H.E. Stanley, T. Preis, Quantifying wikipedia usage patterns before stock market moves, Sci. Rep. 3 (2013) 1801. [31] L. Kristoufek, Can Google Trends search queries contribute to risk diversification? Sci. Rep. 3 (2013) 2713. [32] T. Preis, D. Reith, H.E. Stanley, Complex dynamics of our economic life on different scales: insights from search engine query data, Phil. Trans. R. Soc. A 368 (1933) (2010) 5707–5719. [33] T. Preis, H.S. Moat, H.E. Stanley, Quantifying trading behavior in financial markets using Google Trends, Sci. Rep. 3 (2013) 1684. [34] T. Preis, H.S. Moat, H.E. Stanley, S.R. Bishop, Quantifying the advantage of looking forward, Sci. Rep. 2 (2012) 350. [35] W. Zhang, D. Shen, Y. Zhang, X. Xiong, Open source information, investor attention, and asset pricing, Ecol. Modell. 33 (2013) 613–619. [36] X. Jin, D. Shen, W. Zhang, Has microblogging changed stock market behavior? Evidence from China, Physica A 452 (2016) 151–156. [37] Y. Zhang, L. Feng, X. Jin, D. Shen, X. Xiong, W. Zhang, Internet information arrival and volatility of SME PRICE INDEX, Physica A 399 (2014) 70–74. [38] Z. Da, J. Engelberg, P. Gao, In search of attention, J. Finance 66 (2011) 1461–1499. [39] T. Vakrman, L. Kristoufek, Underpricing, underperformance and overreaction in initial public offerings: Evidence from investor attention using online searches, SpringerPlus 4 (2015) 1–11. [40] Y. Zhang, W. Song, D. Shen, W. Zhang, Market reaction to Internet news: Information diffusion and price pressure, Econ. Modell. 56 (2016) 43–49. [41] I. Cooper, E. Kaplanis, Home bias in equity portfolios, inflation hedging, and international capital market equilibrium, Rev. Financ. Stud. 7 (1994) 45–60. [42] K.R. French, J.M. Poterba, Investor diversification and international equity markets, Amer. Econ. Rev. 81 (1991) 222–226. [43] L.L. Tesar, I.M. Werner, Home bias and high turnover, J. Int. Money Finance 14 (1995) 467–492. [44] J.D. Coval, T.J. Moskowitz, Home bias at home: Local equity preference in domestic portfolios, J. Finance 54 (1999) 2045–2073. [45] M. Grinblatt, M. Keloharju, How distance, language, and culture influence stockholdings and trades, J. Finance 56 (2001) 1053–1073. [46] T. Loughran, P. Schultz, Liquidity: Urban versus rural firms, J. Financ. Econ. 78 (2005) 341–374. [47] Z. IvkoviĆ, S. Weisbenner, Local does as local is: information content of the geography of individual investors’ common stock investments, J. Finance 60 (2005) 267–306. [48] U. Wongchoti, F. Wu, Provincial co-movement in Chinese stock returns, Appl. Financ. Econ. Lett. 4 (2008) 171–176. [49] C. Pirinsky, Q. Wang, Does corporate headquarters location matter for stock returns? J. Finance 61 (2006) 1991–2015. [50] S.J. Brown, J.B. Warner, Using daily stock returns: The case of event studies, J. Financ. Econ. 14 (1985) 3–31. [51] W. Zhang, X. Li, D. Shen, A. Teglio, Daily happiness and stock returns: Some international evidence, Physica A 460 (2016) 201–209.
Contents lists available at ScienceDirect
Physica A journal homepage: www.elsevier.com/locate/physa
Investor sentiment and stock returns: Evidence from provincial TV audience rating in China Yongjie Zhang a,b , Yuzhao Zhang a , Dehua Shen a,b,∗ , Wei Zhang a,c a
College of Management and Economics, Tianjin University, Tianjin, 300072, PR China
b
China Center for Social Computing and Analytics (CCSCA), Tianjin University, Tianjin, 300072, PR China
c
Key Laboratory of Computation and Analytics of Complex Management Systems (CACMS), Tianjin, 300072, PR China
highlights • • • •
Provincial investor sentiment is positively related to stock returns. The provincial correlation coefficient is larger than the cross-provincial correlation coefficient. There exists home bias in Chinese stock market. Provincial investor sentiment can explain the provincial comovement.
article
info
Article history: Received 12 May 2016 Received in revised form 24 August 2016 Available online 24 September 2016 Keywords: TV audience rating Provincial investor sentiment Home bias Provincial comovement Internet information
abstract In this paper, we advocate the provincial TV audience rating as the novel proxy for the provincial investor sentiment (PIS) and investigate its relation with stock returns. The empirical results firstly show that the PIS is positively related to stock returns. Secondly, we provide direct evidence on the existence of home bias in China by observing that the provincial correlation coefficient is significantly larger than the cross-provincial correlation coefficient. Finally, the PIS can explain a large proportion of provincial comovement. To sum up, all these findings support the role of the non-traditional information sources in understanding the ‘‘anomalies’’ in stock market. © 2016 Elsevier B.V. All rights reserved.
1. Introduction The role that investor sentiment plays in asset pricing has been highlighted for a long history in financial economics. The earliest discussion could broadly trace back to Keynes’s ‘‘animal spirits’’, which claims that prices move in a way unrelated to fundamentals [1]. The academic enthusiasm has been aroused by Refs. [2–6], who clearly illustrate that irrational noise traders (with erroneous psychological beliefs and diverse biases) could not be offset by limited arbitrageurs and therefore they have material impact on stock prices. To empirically investigate this issue, various investor sentiment proxies have been put forward to investigate the contemporaneous correlations between investor sentiment and market-wide variables e.g., stock returns, volatility and liquidity as well as time series predictive power for these variables. Among these, two distinct streams of literature stand out. The first stream refers to the continuous observations of overly forecasts of cash flow or investors’ state of mind revealed by survey on individual investor or consumer confidence [7–12], sentient extracted from the online message boards [13–16] as well as the market-wide variables, including net mutual
∗
Corresponding author at: College of Management and Economics, Tianjin University, Tianjin, 300072, PR China. E-mail address: [email protected] (D. Shen).
http://dx.doi.org/10.1016/j.physa.2016.09.043 0378-4371/© 2016 Elsevier B.V. All rights reserved.
Y. Zhang et al. / Physica A 466 (2017) 288–294
289
fund redemptions, dividend premium, bid–ask spreads and average first-day returns on IPOs [17–20]. The rationale of these studies is that some proxies are essentially correlated to investor sentiment or served as the reflections of investor behavior. For example, the linguistic analysis of the online postings in message boards could be expounded as the individual investor sentiment towards certain firms. However, this stream of literature has the disadvantage of endogenous problem. Because the online postings and changes in market variables, e.g., dividend premium and bid–ask spreads, could be the results of the trading activities. It is inappropriate to ‘‘test a theory that is about from inputs to outputs with an output measure’’ as claimed by Ref. [21]. Besides, the participation rate of the survey is not sufficient enough and participants have little incentives to response to survey questions truthfully. The second stream of the literature mainly relies on some sentiment-altering events, which can lead to sentiment changes in a large number of investor and thus influence the asset prices. Therefore this stream of sentiment proxies is discrete variable, including the game results of sport event [22,23], TV series finales [24] as well as the aviation disasters [25]. This stream of sentiment proxies outperforms the first in the sense that it avoids the endogenous problem as well as provides more objective emotions. Admittedly, these discrete proxies only allow us to investigate the price changes around the special events, but fail to provide further illustrations on time series anomalies. In this paper, we aim to bridge the gap between the above-mentioned two streams of literature by advocating the provincial TV audience rating as the novel proxy for the provincial investor sentiment (hereafter, PIS). This PIS proxy has the advantage of coming from a large number of potential investor, avoiding the endogenous problem as well as serving as continuous observations. The rationales of this interpretation are in the following three aspects. Firstly, drawing on the literature on economic psychology, sentiment can affect financial decision-making through determining the risk perception as well as the information processing behavior of investors when forming expectations [26,27]. Secondly, given the high coverage rate, the TV audience rating represents the sentiment from hundreds of millions of the TV watcher. Besides, as is shown in Ref. [24], TV series final can eventually affect investors’ demand for risky assets through connecting with reviewer’s emotions. Thirdly, Internet information has been extensively employed in econophysics [28–40]. In particular, the utilization of search engine [31–33,35,38,39], online news [29,37,40] as well as the microblogging [36] has profoundly reshaped our understanding of complex financial economic systems. With this novel PIS proxy, we contribute to the existing literature in two aspects. On one hand, we can directly measure the correlation coefficients between PIS and its corresponding as well as stock returns of others provincial. The distinct differences in the correlation coefficients provide direct evidence on the existence of home bias in Chinese stock market. On the other hand, the PIS can explain a large proportion of provincial comovement and the results are robust to alternative proxies of PIS across different models. The rest of this paper is organized as follows. We put forward the hypotheses in Section 2. Section 3 describes the data. Section 4 performs the empirical analysis and illustrates the results. Robustness test is given in Section 5. Section 6 concludes. 2. Hypotheses The daily PIS proxy allows us to directly examine the behavioral predictions on the contemporaneous relations between investor sentiment and stock returns. Existing literature based on the proxies constructed by Internet information has documented a positively contemporaneous relation to stock returns [15,16], showing that the highest (lowest) sentiment corresponds to positive (negative) stock returns in the market index. In that sense, we put forward the most straightforward hypothesis on the positively contemporaneous relations between PIS and corresponding stock returns. Hypothesis 1. The PIS has a positively contemporaneous relation to stock returns. The home bias phenomenon has long been one of the most intriguing puzzles in financial economics, which could be understood as investors construct their portfolios with disproportionate amount of assets on local (domestic) equities in spite of the benefits of diversifications [41–43]. Besides, several findings emphasize the crucial role of geographic location in financial decision-making and asset prices [44–47]. In that sense, if there exists home bias puzzle in Chinese stock market, the correlation coefficient between PIS and corresponding stock returns (provincial correlation coefficient) should be larger than the correlation coefficient between PIS and stock returns of other provinces (cross-provincial correlation coefficient). For the same reason, there should be no significant differences in provincial correlation coefficient and cross-provincial correlation coefficient if there is no home bias puzzle in Chinese stock market. The following two alternative hypotheses are naturally derived. Hypothesis 2a. The provincial correlation coefficient is larger than the cross-provincial correlation coefficient. Hypothesis 2b. There are no significant differences in provincial correlation coefficient and cross-provincial correlation coefficient. In a recent work by Wongchoti and Wu [48], they show that stock returns exhibit provincial comovement after controlling for the market and industry effect. This location-based movement cannot be explained by fundamental factors and is pronounced in firm with large number of local investors [49]. In that sense, if PIS represents the sentiment from local investors, it should display some explanatory power for the observed provincial comovement. Therefore, we put forward the hypothesis on the explanation of PIS on provincial comovement. Hypothesis 3. The PIS can explain the provincial comovement.
290
Y. Zhang et al. / Physica A 466 (2017) 288–294 Table 1 Provinces, municipalities and firms. This table reports the average number of firms during the sample period as the TV audience rating. Because some firms undergo trading halt, the average number is non-integer. The municipalities are Beijing, Shanghai, Tianjin and Chongqing. Provinces (Municipalities)
Firms
Provinces (Municipalities)
Firms
Anhui Beijing Fujian Guangdong Guangxi Guizhou Hebei Henan Heilongjiang Hubei Hunan
76.42 207.10 83.14 369.36 27.77 19.19 46.75 63.46 30.46 80.26 66.59
Jilin Jiangsu Jiangxi Liaoning Shandong Shanghai Sichuan Tianjin Yunnan Zhejiang Chongqing
36.21 228.12 31.11 66.61 147.98 223.97 84.16 36.67 26.19 239.11 35.22
3. Data description We combine several datasets in this study. The capital data comes from the China Stock Market & Accounting Research Database (CSMAR Database), including individual stock returns after the reimbursement, total capitalization, industry classification codes and market returns (Shanghai Shenzhen CSI 300 Index, Shanghai Composite Index, Shenzhen Composite Index and Shanghai Shenzhen Composite Index). The audience rating data is from CSM Media Research (http://www.csm.com.cn/), in which TV audience measurement network provides data that represents the viewing of 1.27 billion people in China. Therefore, the audience rating data are reliable and representative in the sense of large size of the sample. Given the limited access to the database, we obtain the audience rating of 22 provinces and municipalities (Beijing, Shanghai, Tianjin and Chongqing) over the period from 1 April 2013 to 3 April 2014. The GDP data is from the National Bureau of Statistics of China (http://www.stats.gov.cn/). We use the yearly GDP in 2013 to calculate the GDP-based investor sentiment in Section 4. Besides, in order to calculate the abnormal returns, we extend the sample period of the capital data to 1 year before the beginning of the audience rating, i.e., from 1 April 2012 to 3 April 2014. Table 1 reports the provinces and the corresponding average number of the firms in the sample period as the audience rating. The average number of the firms is calculated as the daily firms without trading halt during the sample. All the firms account for more than 90% of the listed firms in both Shanghai and Shenzhen stock exchanges. Therefore, our sample is a good representation of Chinese stock market. 4. Empirical analysis and results We obtain the province code from the CSMAR Database and separate the stocks into different provinces. For each given trading day, we use the average return of all the listed firms in certain province as the mean of stock return for the province. In order to observe the differences in a clear-cut way, we classify the investor sentiment into quartiles from the lowest to the highest days, i.e., the lowest-subgroup, the second-subgroup, the third-subgroup and the highest-subgroup. To avoid the spurious relations between TV audience rating and stock returns, e.g., the weekend effect, we do not use the raw value. In particular, we regress the raw TV audience rating on weekday dummies and keep the residuals as the PIS for a certain province. Fig. 1 gives an illustration of the Jiangxi province. As is shown, the PIS in Jiangxi province is significantly different from its raw TV audience rating. Besides, following [50], we use three different models to calculate the abnormal stock returns for all the provinces. Table 2 reports the results of the (abnormal) stock returns in the highest and lowest-subgroups. We find that the mean of the stock returns and the abnormal returns in the highest subgroup are significant larger than that in the lowest subgroup at 1% level. These results suggest that the PIS is positively related to stock returns, which is in accord with the sentiment extracted from Yahoo! Finance message board and Twitter [14,51]. Abnormal return 1 (Mean adjusted return) ARi,t = Ri,t − RI RI =
1
−26
100 t =−125
(1) Ri,t
(2)
where Ri,t is the daily return for province i at time t. Daily return for a certain province is calculated as the average return of all the listed firms in the province excluding the firms with trading halt; RI is the simple average of Ri,t of province i in the [−125, −26] estimation period. Abnormal return 2 (Market adjusted return) ARi,t = Ri,t − Rm,t where Rm,t is the return on equally weighted daily return at time t.
(3)
Y. Zhang et al. / Physica A 466 (2017) 288–294
291
Panel A: TV audience rating.
Panel B: PIS. Fig. 1. An illustration of TV audience rating and PIS. Table 2 PIS and stock returns. This table reports the PIS and stock return in quartiles. We calculate mean of the corresponding (abnormal) stock returns in the lowest and highest-subgroups, respectively. We perform the two-sample t-test to detect the significant differences between highest and lowest-subgroups. The t-values in the parentheses denote the significances. Subgroups
Mean of return
Abnormal return 1
Abnormal return 2
Abnormal return 3
Lowest Highest
−0.0002
−0.0016
0.0017 0.0019*** (t-value = 11.88)
0.0002 0.0018*** (t-value = 10.98)
0.0006 0.0018 0.0012*** (t-value = 18.22)
0.0008 0.0019 0.0011*** (t-value = 16.76)
Highest–Lowest ***
Denotes the significant at 1% level.
Abnormal return 3 (OLS market model) ARi,t = Ri,t − αi − βi Rm,t
(4)
where αi and βi are estimated parameters from the CAPM model with the moving estimation window of [−125, −26]. We further calculate the provincial and cross-provincial correlation coefficients with the Kendall correlation analysis [51]. The Kendall correlation analysis has the advantage of not making assumptions about the probability distributions of the variables. To depict the time-varying characteristic of the correlation coefficients, we employ the rolling windows estimation method to calculate the Kendall correlation coefficients. Admitted, the major disadvantage of this method is that the choice of the window length is arbitrary. In the first analysis, we employ the window length of 50 and it gives 193 observations. Fig. 2 illustrates that the provincial correlation coefficient (0.0822) is significant larger than that of the cross-provincial correlation coefficient (0.0350) at 1% level with t-value = 28.97 and p-value = 0.0000.1 The larger provincial correlation coefficient suggests that there exists home bias in Chinese stock market. To empirically investigate the explanatory power of the PIS on provincial comovement, we need to calculate the provincial comovement controlling for market returns and industry returns [48]. The following regression models are employed: Ri,t = α + β1 RP ,t + β2 RM ,t + β3 RIND,t + εi,t
(5)
where Ri,t denotes the daily return of stock i on date t , RP ,t denotes the equally-weighted daily return in the corresponding province, RM ,t denotes the daily return on the market portfolio, RIND,t denotes the corresponding daily return on the industry that stock i belongs to. In each regression model, stock i is excluded in the construction of the provincial stock index (RP ,t ) to avoid the spurious correlation. The null hypothesis that there is no provincial comovement implies β1 = 0. For robustness test, we construct four alternative models with different measurements of RM ,t . Models (from (1) to (4)) in Table 3 represent
1 In fact, we also perform the Pearson and Spearman correlation analysis. They all show that the provincial correlation coefficients are significantly larger than the cross-provincial correlation coefficients.
292
Y. Zhang et al. / Physica A 466 (2017) 288–294
Fig. 2. Provincial and cross-provincial Kendall correlation coefficients. Table 3 Explain the provincial comovement. This table reports the explanatory power of the weighted PISs on the provincial comovement based on model (6). In the cross-sectional analysis, the PISs exhibit strong explanatory power for the provincial comovement across all models with alternative proxies. Values in the parentheses are the corresponding p-values of the t-statistics. Model
(1)
(2)
(3)
(4)
7.4382 2.5805** (0.0179) 21.23%
6.3655 1.9827* (0.0613) 12.24%
12.92 3.0131*** (0.0069) 27.78%
4.9262 2.9354*** (0.0082) 26.62%
4.8664 2.9174*** (0.0085) 26.34%
5.4255 3.161*** (0.0049) 29.99%
7.8085 3.0553*** (0.0062) 28.41%
9.237 3.007*** (0.0070) 27.69%
9.231 3.0385*** (0.0065) 28.16%
9.717 3.0273*** (0.0067) 28.00%
1.5562 4.4393*** (0.0026) 34.02%
Panel A: Capitalization-based PIS β 7.1027 2.4040** t-statistics (0.0260) Adjusted R2 18.54% Panel B: GDP-based PIS
β
t-statistics Adjusted R2 Panel C: Firm-based PIS
β
t-statistics Adjusted R2 * ** ***
Denotes the significant at 10% level. Denotes the significant at 5% level. Denotes the significant at 1% level.
the models with Shanghai Shenzhen CSI 300 Index, Shanghai Composite Index, Shenzhen Composite Index and Shanghai Shenzhen Composite Index, respectively. pc m = α + β Sentiment m + εm
(6)
where pc m is the provincial comovement for province m, i.e., the mean of the β1 in certain province of model (5), Sentiment m represents the three different constructions of PIS, i.e., the capitalization-based PIS, the GDP-based PIS and the firm-based PIS. Specifically, for a certain province, we have the yearly GDP, number of the firms and their total capitalization. We calculate the product term of the average PIS during the sample period and the yearly GDP as the GDP-based PIS. Model (6) is the cross-sectional regression and all the variables pass through the Augmented Dickey–Fuller test confirming the stationary property. Table 3 reports our main results for Hypothesis 3 with the cross-sectional regression model (6). As is shown, all the PISs (Panel A, Panel B and Panel C) can explain large proportions of the provincial comovement. In particular, the firm-based PIS exhibits the best explanatory power. All these findings uphold Hypothesis 3. 5. Robustness Since we have employed three alternative models to calculate the abnormal returns for Hypothesis 1 and constructed three distinct measurements of PIS across four models for Hypothesis 3, we mainly focus on the robustness of Hypothesis 2
Y. Zhang et al. / Physica A 466 (2017) 288–294
293
Table 4 Correlation coefficients with alternative choices of window length. This table reports the provincial and cross-provincial correlation coefficients with alternative choices of window length. Observation is the number of the time-varying correlation coefficients. PCC denotes the provincial correlation coefficient and CCC denotes the cross-provincial correlation coefficient. PCC–CCC denotes the differences. We perform the two-sample t-test to detect the significant differences. Values in the parentheses are the corresponding pvalues of the t-statistics. Variables
Observation PCC CCC PCC–CCC ***
Alternative choices of window length 40
45
50
55
60
203 0.0922 0.0351 0.0571*** (0.0000)
198 0.0871 0.0353 0.0518*** (0.0000)
193 0.0822 0.0350 0.0472*** (0.0000)
188 0.0773 0.0347 0.0425*** (0.0000)
183 0.0724 0.0342 0.0382*** (0.0000)
Denotes the significant at 1% level.
in this section. In particular, we consider alternative choices of the window length to calculate the provincial and crossprovincial correlation coefficients. Table 4 reports the empirical results of the provincial and cross-provincial correlation coefficients with 40, 45, 50, 55 and 60 trading days of window length. As is shown, the provincial correlation coefficient is significant larger than the cross-provincial correlation coefficient at 1% level. All these results further prove that there exists home bias in Chinese stock market. 6. Conclusions In this paper, we advocate the provincial TV audience rating as the novel proxy for PIS and investigate its relation with stock returns. This novel investor sentiment proxy overcomes the endogenous problem and reveals the state of mind from hundreds of millions investors in China. By dividing the PIS into subgroups and observing the corresponding (abnormal) stock returns, we find that the PIS is positively related to stock returns. Besides, given the nature of PIS representing the local investor sentiment, we provide direct evidence on the existence of home bias. Finally, we show that the PIS can explain a large proportion of provincial comovement with the cross-sectional analysis. All these findings highlight the role of the non-traditional information sources in understanding the ‘‘anomalies’’ in stock market. However, the predictive ability of the PIS on stock returns is unveiled in this study. We leave this for future research. Acknowledgments We thank Tony He and Youwei Li for helpful comments and discussions during their visiting at Tianjin University. This work is supported by the National Natural Science Foundation of China (71271144, 71320107003 and 71532009) and Core Projects in Tianjin Education Bureaus Social Science Program (2011ZD008 and 2014ZD13). References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26]
J.M. Keynes, The General Theory of Employment, Interest and Money, Macmillan, London, 1936. F. Black, Noise, J. Finance 41 (1986) 529–543. H. Hong, J.C. Stein, A unified theory of underreaction, momentum trading, and overreaction in asset markets, J. Finance 54 (1999) 2143–2184. J.B. De Long, A. Shleifer, L.H. Summers, R.J. Waldmann, Noise trader risk in financial markets, J. Polit. Econ. 98 (1990) 703–738. K. Daniel, D. Hirshleifer, A. Subrahmanyam, Investor psychology and security market under- and overreactions, J. Finance 53 (1998) 1839–1885. N. Barberis, A. Shleifer, R. Vishny, A model of investor sentiment, J. Financ. Econ. 49 (1998) 307–343. G.W. Brown, M.T. Cliff, Investor sentiment and the near-term stock market, J. Empir. Finance 11 (2004) 1–27. G.W. Brown, M.T. Cliff, T. Investor, Sentiment and asset valuation, J. Bus. 78 (2005) 405–440. K.L. Fisher, M. Statman, Investor sentiment and stock returns, Financ. Anal. J. 56 (2000) 16–23. M. Lemmon, E. Portniaguina, Consumer confidence and asset prices: Some empirical evidence, Rev. Financ. Stud. 19 (2006) 1499–1529. M.E. Solt, M. Statman, How useful is the sentiment index? Financ. Anal. J. 44 (1988) 45–55. W.Y. Lee, C.X. Jiang, D.C. Indro, Stock market volatility, excess returns, and the role of investor sentiment, J. Banking Finance 26 (2002) 2277–2299. R. Tumarkin, R.F. Whitelaw, News or noise? Internet postings and stock prices, Financ. Anal. J. 57 (2001) 41–51. S.-H. Kim, D. Kim, Investor sentiment from Internet message postings and the predictability of stock returns, J. Econ. Behav. Organ. 107 (2014) 708–729. Part B. S.R. Das, M.Y. Chen, Yahoo! for Amazon: Sentiment extraction from small talk on the web, Manage. Sci. 53 (2007) 1375–1388. W. Antweiler, M.Z. Frank, Is all that talk just noise? The information content of Internet stock message boards, J. Finance 59 (2004) 1259–1294. M. Baker, J. Wurgler, Investor sentiment and the cross-section of stock returns, J. Finance 61 (2006) 1645–1680. M. Baker, J. Wurgler, Investor sentiment in the stock market, J. Econ. Perspect. 21 (2007) 129–152. M. Baker, J.C. Stein, Market liquidity as a sentiment indicator, J. Financ. Mark. 7 (2004) 271–299. R. Neal, S.M. Wheatley, Do measures of investor sentiment predict returns? J. Finan. Quant. Anal. 33 (1998) 523–547. L. Qiu, I. Welch, Investor sentiment measures, National Bureau of Economic Research, 2006. A. Edmans, D. GarcÍA, Ø. Norli, Sports sentiment and stock returns, J. Finance 62 (2007) 1967–1998. S.-C. Chang, S.-S. Chen, R.K. Chou, Y.-H. Lin, Local sports sentiment and returns of locally headquartered stocks: A firm-level analysis, J. Empir. Finance 19 (2012) 309–318. G.M. Lepori, Investor mood and demand for stocks: Evidence from popular TV series finales, J. Econ. Psychol. 48 (2015) 33–47. G. Kaplanski, H. Levy, Sentiment and stock prices: The case of aviation disasters, J. Financ. Econ. 95 (2010) 174–201. E.J. Johnson, A. Tversky, Affect, generalization, and the perception of risk, J. Pers. Soc. Psychol. 45 (1983) 20–31.
294
Y. Zhang et al. / Physica A 466 (2017) 288–294
[27] N. Schwarz, G.L. Clore, Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states, J. Pers. Soc. Psychol. 45 (1983) 513–523. [28] B. Edelman, Using Internet data for economic research, J. Econ. Perspect. 26 (2012) 189–206. [29] D. Shen, W. Zhang, X. Xiong, X. Li, Y. Zhang, Trading and non-trading period Internet information flow and intraday return volatility, Physica A 451 (2016) 519–524. [30] H.S. Moat, C. Curme, A. Avakian, D.Y. Kenett, H.E. Stanley, T. Preis, Quantifying wikipedia usage patterns before stock market moves, Sci. Rep. 3 (2013) 1801. [31] L. Kristoufek, Can Google Trends search queries contribute to risk diversification? Sci. Rep. 3 (2013) 2713. [32] T. Preis, D. Reith, H.E. Stanley, Complex dynamics of our economic life on different scales: insights from search engine query data, Phil. Trans. R. Soc. A 368 (1933) (2010) 5707–5719. [33] T. Preis, H.S. Moat, H.E. Stanley, Quantifying trading behavior in financial markets using Google Trends, Sci. Rep. 3 (2013) 1684. [34] T. Preis, H.S. Moat, H.E. Stanley, S.R. Bishop, Quantifying the advantage of looking forward, Sci. Rep. 2 (2012) 350. [35] W. Zhang, D. Shen, Y. Zhang, X. Xiong, Open source information, investor attention, and asset pricing, Ecol. Modell. 33 (2013) 613–619. [36] X. Jin, D. Shen, W. Zhang, Has microblogging changed stock market behavior? Evidence from China, Physica A 452 (2016) 151–156. [37] Y. Zhang, L. Feng, X. Jin, D. Shen, X. Xiong, W. Zhang, Internet information arrival and volatility of SME PRICE INDEX, Physica A 399 (2014) 70–74. [38] Z. Da, J. Engelberg, P. Gao, In search of attention, J. Finance 66 (2011) 1461–1499. [39] T. Vakrman, L. Kristoufek, Underpricing, underperformance and overreaction in initial public offerings: Evidence from investor attention using online searches, SpringerPlus 4 (2015) 1–11. [40] Y. Zhang, W. Song, D. Shen, W. Zhang, Market reaction to Internet news: Information diffusion and price pressure, Econ. Modell. 56 (2016) 43–49. [41] I. Cooper, E. Kaplanis, Home bias in equity portfolios, inflation hedging, and international capital market equilibrium, Rev. Financ. Stud. 7 (1994) 45–60. [42] K.R. French, J.M. Poterba, Investor diversification and international equity markets, Amer. Econ. Rev. 81 (1991) 222–226. [43] L.L. Tesar, I.M. Werner, Home bias and high turnover, J. Int. Money Finance 14 (1995) 467–492. [44] J.D. Coval, T.J. Moskowitz, Home bias at home: Local equity preference in domestic portfolios, J. Finance 54 (1999) 2045–2073. [45] M. Grinblatt, M. Keloharju, How distance, language, and culture influence stockholdings and trades, J. Finance 56 (2001) 1053–1073. [46] T. Loughran, P. Schultz, Liquidity: Urban versus rural firms, J. Financ. Econ. 78 (2005) 341–374. [47] Z. IvkoviĆ, S. Weisbenner, Local does as local is: information content of the geography of individual investors’ common stock investments, J. Finance 60 (2005) 267–306. [48] U. Wongchoti, F. Wu, Provincial co-movement in Chinese stock returns, Appl. Financ. Econ. Lett. 4 (2008) 171–176. [49] C. Pirinsky, Q. Wang, Does corporate headquarters location matter for stock returns? J. Finance 61 (2006) 1991–2015. [50] S.J. Brown, J.B. Warner, Using daily stock returns: The case of event studies, J. Financ. Econ. 14 (1985) 3–31. [51] W. Zhang, X. Li, D. Shen, A. Teglio, Daily happiness and stock returns: Some international evidence, Physica A 460 (2016) 201–209.