Investor Sentiment and Stock Return: Evidence from Chinese Stock Market

Transcription

1 Investor Sentiment and Stock Retrn: Evidence from Chinese Stock Market Feng Jnwen and Li Xinxin School of Economics and Management, Nanjing University of Science and Technology, Nanjing Jiangs, China * Corresponding athor: Feng Jnwen, School of Economics and Management, Nanjing University of Science and Technology, Nanjing Jiangs 10094, China, Tel: ; @qq.com Received: Janary 16, 017; Accepted: Janary 7, 017; Pblished: Febrary 04, 017 Abstract This paper selects the trading volme data from Janary 010 to December 014 to stdy investor sentiment and regard the market daily earnings ratio of Shanghai Stock Exchange A-shares index as stock retrn. Using ARMA model to research its own characteristics of investor sentiment, then se VAR model and Granger casality test, implse response fnction and variance decomposition to research the relationship between investor sentiment and the market retrns, thereafter empirical reslts show that two series have Granger casal relationship exist between each other. Then the heteroscedastic EGARCH model tests the impact of the retrns on investor sentiment, and the bad news in the market can case greater volatility than good news ; Finally, the EGARCH-M heteroscedastic model shows that the impact of investor sentiment to retrns exist risk premim phenomenon. Keywords: Investor sentiment; Stock retrn; Granger casality test; Heteroscedastic model Introdction The central premise of traditional financial theory is that investors are completely rational. In 1950s, the research field of psychology began to expand to cognitive psychology and social psychology. In this context, behavioral finance theory and experimental economics began to rise. They were based on the real psychological activities and decision-making process of the investors, and they fond that the decision-making process is actally the process of risk verss retrn on the psychological. China's stock market development time is short, instittion bilding is not sond. As an important part of behavioral finance theory, research on interactive relationship between investor sentiment and stock retrn helps improve or existing behavioral finance theory system, and frther revealing the inherent relationship between the stock market investor behavior and decisions related to stock market data and stock market gains. It can provide a more abndant fondation for the research on asset price formation, and provide a sefl theoretical spport to control or financial markets. The research ideas of this paper are as follows: firstly, the paper reviews the existing research on investor sentiment and stock market retrns of domestic and foreign, and then introdces the data and model tools sed in this paper. Then, introdce the research methods, inclding the ARMA model, VAR model and EGARCH model. Finally, empirical analysis is carried ot Citation: Feng Jnwen,Li Xinxin. Investor sentiment and stock retrn: evidence from Chinese stock market. Biotechnol Ind J. 017;13(1): Trade Science Inc. 1

2 Febrary-017 to stdy the dynamic relationship between investor sentiment and stock market retrns and the mtal inflence of the two in or contry. The innovation of this paper is that the stdy of the impact of investor sentiment and stock retrn in China and abroad are mostly selected only one aspect, this paper will be more than one model for the integration of research, the content is rich and the sample data are p to date. Literatre Review Smmary of foreign research literatre The empirical research on the investor sentiment and retrn of foreign research literatre mainly adopt the extended ARCH model and the extended GARCH model to stdy the volatility of retrn rate. Nelson [1] overcomes the drawback of GARCH model, and proposes a more simple and flexible asymmetric ARCH dynamic model. Lee et al. [] sed the GARCH model to analyze the impact of investor sentiment on market volatility, and fond that the decline of investor sentiment cold lead to a more intense volatility in the market. Baker and Wrgler [3] define investor sentiment as a pessimistic or optimistic view of the ftre price of the stock, or a speclative tendency, which is driven by the investment demand and decision-making will have an impact on the stock price, and also confirmed that the trading volme can be sed as an effective indicator of investor sentiment. Gerhard Kling and Lei Gao [4] select China's Shanghai stock exchange daily data to stdy the Chinese instittional investors, fond that in the short term, instittional investor sentiment has a positive feedback effect, bt there is no association in the long term. At the same time, their research reveals the volatility of investor sentiment in the system response by sing the EGARCH model, the decline in investor sentiment will strengthen the market volatility and instability, bt did not find its inflence on predicting the market expected earnings. Review of domestic literatre Rao Ylei et al. [5] choose CCTV BSI index to do the empirical research and they fond there is no significant relationship between CCTV BSI index and the expected stock retrn. Bt Wang Meijin et al. [6] also se the BSI index to indicate the investor sentiment, the empirical reslts show that investor sentiment can not only significantly affect the stock market retrn, bt also can make a significant negative correction to the volatility of earnings. Cheng Kn et al. [7] select the "good light index" to constrct the investor sentiment index which released by Dynamic analysis of stock market. The empirical reslts show that the intermediate sentiment index is the Granger case of the stock market retrns, and the two does not exist mtal feedback effect. At the same time, the reslts of implse response and variance decomposition show that the impact of the medim-term investor sentiment on the stock market volatility is stronger than the short-term investor sentiment. Li Xiao et al. [8] select the GARCH model to condct research, and EGARCH (3,1) model was sed to stdy the volatility of stock index in Shenzhen stock market at last, and the relevant policy recommendations were pt forward. Hang Ganyao [9] selected Shanghai secrities composite index to fit and forecast accracy of for GARCH models on the volatility of the stock market to make a comparative stdy. Li Xindan et al. [10] sed the same index and fond that ARCH effect existed in investor sentiment, at the same time investor sentiment has a certain degree of spillover effect" on the stock market retrns. Yang Sh e and Zhang Qiang [11] selected the discont of closed-end fnd investors, growth rate and trnover rate to constrct comprehensive sentiment index of investor sentiment, based on the noise trading theory model (DSSW model) proposed by Delong in 1990, sing GARCH-M (1,1) to stdy whether Chinese investor sentiment has

3 Febrary-017 inflence on the stock market retrn and the empirical reslts showed that instittional investor sentiment is the main factor affecting the stock price, while individal investors are not significant, and knocked down the inflence of individal investor sentiment on small cap stocks. Yang Yang and Wan DiFang et al. [1] select BSI index and trnover rate to reflect investor sentiment, analyzed investor sentiment and stock market retrn and its volatility in different market state, empirical findings showed that stock retrn has significant impact on investor sentiment, not vice versa; and investor sentiment has asymmetric impact on stock retrn dring the bear market, bt the bll market does not exist. Chi Lix et al. [13] select the fnd's capital flows to constrct the investor sentiment index, sing the panel data model to stdy the relationship between investor sentiment and stock retrns. The empirical reslts showed that extreme emotions in China's stock market has a niqe ability to predict, the impact of investor sentiment is asymmetric, and reasonable sentiment index is condcive to or secrities investors and reglators to make better decisions. Xia Fang [14] combined investor sentiment with the behavior of managers to stdy the problem of the stock mispricing. Wang Liping [15] choose the macro point of view, sing principal component analysis method to constrct a comprehensive investor sentiment, introdcing the expected theory, sing linear regression model to stdy the impact of interest rate srprises, investor sentiment on the stock index retrns, and fond that investor sentiment can enhance the impact of policy variables on the stock index retrns. Recently, Zhang Wei et al. [16] examined the relations between the daily happiness sentiment extracted from Twitter and the stock market performance in 11 international stock markets. By partitioning this happiness sentiment into qintiles from the least to the happiest days, they first shown that the contemporary correlation coefficients between happiness sentiment and index retrn in the 4 and most-happiness sbgrops were higher than that in least, and 3-happiness sbgrops. Secondly, the happiness sentiment can provide additional explanatory power for index retrn in the most-happiness sbgrop. Thirdly, the daily happiness can granger-case the changes in index retrn for the majority of stock markets. Forthly, they fond that the index retrn and the range-based volatility of the most-happiness sbgrop are larger than those of other sbgrops. These reslts highlight the important role of social media in stock market. Zhang Yongjie et al. [17] advocated the provincial TV adience rating as the novel proxy for the provincial investor sentiment (PIS) and investigate its relation with stock retrns. The empirical reslts firstly shown that the PIS is positively related to stock retrns. Secondly, they provided 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 co movement. To sm p, all these findings spport the role of the non-traditional information sorces in nderstanding the anomalies in stock market. Sample selection and data sorces Investor sentiment is not only an immediate reflection of the crrent stock market retrns, bt also an expectation of ftre market prices. According to index data sorces, it is generally divided into two categories, one is throgh direct qestionnaire srvey related investors to obtain direct indicators, sch as "good light index" which released by Dynamic analysis of stock market ; the other is indirect indicator of investor sentiment change by collecting relevant financial data from the financial markets. Direct indicators are sbject to the constraints of their own circmstances, the investigation object is limited. Coverage is small and cannot garantee its athenticity. This paper selects trading volme as measre index has two advantages: first of all, the main one of the direct performance of the investors' psychological decision-making behavior is 3

4 Febrary-017 the change of the trading volme; secondly, the trading volme as a daily data has continity, which has obvios advantages in the stdy of the emotional trend of investors. This paper selects the change of trading volme between Janary 010 to December 014 to express the investor sentiment. The calclation formla is as follows: Investor sentiment change = (crrent trading volme-volme of the previos period) /volme of the previos period IS t = (TV -TV t t-1 )/TV t-1 Among them, IS t represents the investor sentiment change, TV t represents the trading volme of period t, TV t-1 represents the trading volme of period t-1. This paper selects Shanghai Stock Exchange A shares index to represent the market retrn, the calclation formla is as follows: R t = Ln(P t/pt-1 ) Among them, R t represents the logarithmic retrn of the Shanghai Stock Exchange A shares index at the t day; P t represents date closing price of the Shanghai Stock Exchange A shares index at the t day; P t-1 represents date closing price of the Shanghai Stock Exchange A shares index at the t-1 day. The data sorce is RESSET database, and the analysis tool is Eviews 4.0 and Excel 007. Research Design Data characteristics of investor sentiment TABLE 1 reports the nmerical characteristics of investor sentiment. Investor sentiment time series skewness is , which is in a left skewed distribtion; Krtosis is 7.069, which has obvios peak characteristics. So, the investor sentiment seqence has obvios leptokrtic heavy tail featre, and most of the financial data is similar. The Jarqe-Bera statistic rejected the normal distribtion hypothesis at the level of TABLE 1. Nmerical Characteristics of Investor Sentiment. Variate Mean Median Minimm Maximm Standard Deviation Skewness Krtosis Jarqe-Bera statistic IS Note: The significance level of JB statistic is Model bilding Firstly, considering the mtal inflence between the investor sentiment and the stock index retrn, this paper selects the vector ato regressive (VAR) model to analyze the dynamic relationship. The VAR model with a lag phase k of N variables is expressed as follows: 4

5 Febrary-017 Yt c Yt Y Y, ~ D 0, Among them, Y t ( y, t, y, t, yn, t,), c ( c1, c, cn ) 1 1 t k tk t t 1 j 11, j 1, j N1, j 1, j, j N, j 1N, j N, j NN, j j 1, k, t ( 1, N), Y t is N 1 order time colmn vector, and c is N 1 order constant colmn vector. 1 k are all N N order parameter matrix, t ~ D(0, ) is N 1 order random error colmn vector. This paper constrcts the two element VAR model of investor sentiment and stock index, according to the AIC, HQIC and SBIC criterion, this paper chooses the lag period for the three orders. The reslts of the VAR model's stationary test are all located in the nit circle, so the VAR model is stable, which provides a basis for the following Granger casality test, implse response fnction analysis and variance decomposition. Secondly, this paper stdies the asymmetric impact of stock market retrns on investor sentiment and the impact of investor sentiment volatility on the investor's income. Based on the test reslts of VAR model, the model sed in this paper is bilt with ARMA model and EGARCH model. ARMA (p, q) model is the ato-regressive moving average model, and the general ARMA (p, q) can be expressed as the following form: y t c y y 1 t - 1 t - pyt - p t 1 t - 1 t - q t - q Among them, t is the white noise process for the variance, c is the constant term, j and j represents the ato regression coefficient and the moving average respectively. ARMA model can better nderstand the data characteristics of the investors' sentiment, and it is the basis of the stdy of the mtal inflence of investor sentiment and stock index retrns. This paper ses the system estimation method proposed by Boakes and Jenkins (Box-Jenkins, 1976) and determine the ARMA (1,1) model. ARMA (1,1) model is sed to analyze the ARCH-LM test, and it is fond that there is a significant heteroscedastic in the investor sentiment series, at the same time, the non symmetry of the financial time series is considered, so this paper ses one of the asymmetric GARCH model, EGARCH model to do the frther research. The conditional variance eqation of the EGARCH model is expressed as follows: ln( ) t q j1 ln( t1 ) p i1 i t i t i E t i t i r k1 k t k t k 5

6 Febrary-017 The left side of the eqation is the logarithm of the conditional variance, which means that the leverage effect is exponential, so the forecast vale of the conditional variance mst be non-negative. The existence of the leverage effect can be tested by the hypothesis of γ<0. Compare with the GARCH model, there is no limit to this eqation, and the process is more simple and flexible. Empirical Reslts and Analysis Time series stationary test From the theory of econometrics, the false regression problem cold be generated by sing the non-stationary economic variables to establish the regression model, so it is necessary to condct time series stationary test. This paper selects ADF(Agmented Dickey-Fller) test method and reslts are shown in TABLE, t statistics vale is less than the reslts of nit root test on every significance level, so it can be sed to determine the seqence of investor sentiment seqence IS and the logarithm of the daily yield seqence R is a stationary time series. TABLE. ADF nit root test reslts. Objects ADF Test Statistic 1% Significance Level 5% Significance Level 10% Significance Level Throgh Test (Y/N) IS Y R Y Granger casality test Granger casality test can analyze the casal relationship between two stationary time series X and Y. Test methods are as follows: To estimate the crrent vale of Y is the vale of the degree of its own lag can be explained, and then verify that throgh the introdction of hysteresis vale whether the seqence X can improve the degree of Y is explained. If it is, then the seqence X is the Granger reason of Y and the lag coefficient of X has a significant statistical characteristics. Granger casality test reslts of investor sentiment and market retrns are shown in TABLE 3. TABLE 3. Granger casality test reslts. Nll Hypothesis ᵡStatistic P Vale Conclsion R is not the Granger reason of IS Reject the nll hypothesis IS is not the Granger reason of R Reject the nll hypothesis As can be seen from the table, there is a two-way Granger casality between the retrn rate of stock index R and investor sentiment IS, and R has a more significant. impact on investor sentiment IS. Implse response fnction analysis and variance decomposition 6

7 Febrary-017 The implse response fnction describes the impact of a VAR model endogenos variables to other endogenos variables broght and analyzes the effects when an error term changes, or the model is affected by the dynamic effects of a certain impact on the system. And the variance decomposition is to analyze the contribtion of each strctre impact on the internal variable, and the importance of the impact of different strctres is frther evalated. FIG. 1 shows the response of the stock index retrn to the impact of investor sentiment. As can be seen from the figre, the stock index retrn rate to the investor sentiment implse response reached the maximm in the forth period, at this time, when the investor sentiment appeared 1 percentage points of the positive impact, will make the stock market yield increased by percentage points, followed by a rapid redction in the impact, and showed signs of convergence to zero. In TABLE 4 of the variance analysis reslts, the contribtion of investor sentiment after the first six stabilized and the impact of the stock market's contribtion rate is 9.88%. FIG. 1. R response fnction for IS. TABLE 4. Reslts of stock index retrn variance decomposition (%). Period S. E. IS R FIG. describes the response of the investor sentiment to the stock market shock. When earnings appear one percentage point positive impact, the investor sentiment in the crrent period did not respond, and rise to a maximm positive response rapidly after one period, indicating that the response of investor s sentiment has lagged behind. In the third phase, a relatively 7

8 Febrary-017 weak reverse response, followed by a rapid redction of the impact, and showing signs of convergence of the zero effect. In the analysis of variance reslts in TABLE 5, the contribtion of stock index retrn to stability after the fifth period and the impact of the stock market's contribtion rate is 8.70%. FIG.. IS response fnction for R. TABLE 5. Reslts of investor sentiment variance decomposition (%). Period S.E. IS R Asymmetric effect of stock market retrns on investor sentiment The reslts of VAR model test fond that two-stage investor retrns lagged effect on investor sentiment is not significant, bt the first order is significant, so the stock market retrns research model of investor sentiment shold contain the first-order and the EGARCH model and the conditional variance accordingly constrcted as follows: ISt c b1rt br t 1 ln( t) ln( dis t 1 t 1 ) t 1 t 1 t 1 t t 1 t 1 8

9 Febrary-017 Among them, c is the constant term, b 1 and b are the affecting coefficients of the investor retrn. The above formla, only considering the effect of investor retrns, it shold also take into accont changes in investor sentiment itself, nder the conditions of constant variance, adding expression to adjst the model change in investor sentiment itself as follows: ISt c b1rt br t 1 ln( t) ln( dis t 1 t 1 ) t 1 t 1 t 1 t t 1 t 1 The EGARCH model and the adjsted EGARCH model are estimated respectively, and the reslts are shown in TABLE 6. TABLE 6. Reslts of the EGARCH model estimation. Coefficients The adjsted EGARCH The simple EGARCH model model c (0.0000) (0.0006) b (0.0000) (0.0000) b (0.0000) (0.0000) d (0.0000) θ (0.0000) ω (0.0035) (0.0000) β (0.019) (0.0000) α (0.0140) (0.0001) γ (0.0039) (0.1008) R Note: The vales in parentheses are showing significant levels, R represents statistic model fit test. As can be seen by the table, the adjsted R has improved significantly. In the adjsted EGARCH model, γ is significantly negative, which shows that the stock retrn has asymmetrical impact on investor sentiment, "bad news" can prodce a greater volatility than the same amont of "good news", negative expected retrn on the impact of sentiment generated by investor sentiment is greater than the positive expected rate of retrn, it is also the performance of nder-reaction and over-reaction. The impact of investor sentiment volatility on stock retrn In the reslts of the VAR model, there is no significant effect of investor sentiment on the stock retrn of the first order to the three order, so in the stdy of the impact of investor sentiment on the investment income model, only containing the crrent 9

10 Febrary-017 term, after taking into accont the change in investor sentiment itself, EGARCH-M (1,1) model expression adjsted as follows: ln t ln t1 t1 t1 t1 t1 Among them, C 1 is a constant term, b 3 is the inflence coefficient of investor sentiment, estimating EGARCH-M (1,1) model and adjsted EGARCH-M (1,1) model respectively, the reslts are shown in TABLE 7: TABLE 7. Reslts of EGARCH model estimation. Coefficients The adjsted EGARCH model The simple EGARCH model c (0.0000) (0.0000) b (0.0000) (0.0000) (0.0000) (0.0000) d (0.0000) (0.0000) ω (0.0001) (0.003) β (0.0000) (0.0000) α (0.0003) 0.134(0.0004) γ (0.059) (0.860) R Note: The vales in parentheses are showing significant levels, R represents statistic model fit test As can be seen by the table, significant level of each statistic in the adjsted model has improved significantly, and fit statistics R also increased. The risk premim is significant, which shows that the change of investor sentiment has a significant positive correlation with the earnings of investors, this is the same as Wang Mei jin(004) 's research, also shows that investors in the stock market volatility reqires higher retrns. Rt c1 b3* ISt t 1 d1ist 11t 1 t Conclsion and Revelation This paper selects the trading volme data from Janary 010 to December 014 to stdy investor sentiment and regard the market daily earnings ratio of Shanghai Stock Exchange A-shares index as stock retrn, sing ARMA model to research its own characteristics of investor sentiment, integrating the VAR model, expanded heteroscedastic EGARCH model and EGARCH-M heteroscedastic models to do the empirical test, reslts are as follows: Firstly, in the dynamic relationship between investor sentiment and market retrns, investor sentiment and market retrns have Granger casal relationship exist 10

11 Febrary-017 between each other, and the effect of market retrn on investor sentiment is more significant. In a positive impact on investor sentiment, the stock market retrns slowly to generate the same response; and in the positive impact of market retrns, investor sentiment rapidly generate the same response and reach the maximm the second period, which shows that the emotional response of investors in China is lagging behind. Secondly, the impact of stock market retrns on investor sentiment is asymmetric, that is, the negative market expectations will prodce a greater investor sentiment volatility than the positive expectations, and also shows that China's investors have a loss aversion effect. Finally, investor sentiment has a significant positive correlation with stock retrn, and the "risk premim" effect of the stock market is also existed in or contry. REFERENCES 1. Daniel NB. Conditional Heterosdasticity in Asset Retrns: A New Approach. Econmonetrica. 1991;59: Lee WY, Jiang CX, Indro DC, et al. Stock market volatility, excess retrn, and the role of investor sentiment. J Bank Financ. 00; Baker M, Wrgler J. Investor Sentiment and the Cross-Section of Stock Retrns. J Financ. 006;61: Kling G, Lei Gao. Chinese Instittional Investors Sentiment. 008;18: Ylei R, Dafeng L. Behavioral Finance. Shanghai. Shanghai University of Finance and Economics Press. 003; Meijin W, Jianjn S. Stock Market Retrns. Volatility and the Role of Investor Sentiment in China. Econom Res J. 004;10: Kn C, Renhe L. A Stdy of the Interactional Relationship between the Investor Sentiment and the Stock Markets. Shanghai J Econom. 005;11: Xiao L, Yiming L. The application of GARCH family model in the story -Stdy on Shenzhen Component Index Volatility. Technoeconomics Manag Res. 005;5: Ganyao H, Ri C. Non-linear stdy of stock market volatility in China-Compare findings from GARCH Models. Technoeconomics and Management Research. 007;: Chaohi W, Xindan L. Investor sentiment volatility and stock retrns in China. Jornal of Ningbo University. Hmanities and Social Sciences Edition. 008;11: Qiang Z, Sh e Y. Noise trading, investor sentiment volatility and stock retrns. Sys Eng Theory Prac. 009;3: Yang Y, Difang W. Relationship among investor sentiment. Stock market retrn and volatility in different market states. Sys Eng. 010;1: Lix C, Xintian Z. A stdy on the relationship between investor sentiment and the stock market retrns in China- Based on Panel Data Model. Manag Rev. 011;6: Fang X. Research on investment-cash flow sensitivity from the respective of mispricing. Sys Eng. 01;7: Liping W. The effect of interest srprise on stock index retrns-based on investor sentiment. Jornal of Taiyan Normal University. 014;9: Zhang W, Li X, Shen D, et al. Daily happiness and stock retrns: Some international evidence. Physica A: Statistical Mechanics and its Applications. 016;460: Zhang Y, Zhang Y, Shen D, et al. Investor sentiment and stock retrns: Evidence from provincial TV adience rating in China. Physica A: Statistical Mechanics and its Applications. 017;466: