Sentiment and stock characteristics: Comprehensive study of individual investor influence on returns, volatility, and trading volumes

Kresta, Aleš; Xiong, Jialei; Maidiya, Bahate Article Sentiment and stock characteristics: Comprehensive study of individual investor influence on returns, volatility, and trading volumes Business Systems Research (BSR) Provided in Cooperation with: IRENET - Society for Advancing Innovation and Research in Economy, Zagreb Suggested Citation: Kresta, Aleš; Xiong, Jialei; Maidiya, Bahate (2024) : Sentiment and stock characteristics: Comprehensive study of individual investor influence on returns, volatility, and trading volumes, Business Systems Research (BSR), ISSN 1847-9375, Sciendo, Warsaw, Vol. 15, Iss. 2, pp. 67-82, https://doi.org/10.2478/bsrj-2024-0018 This Version is available at: https://hdl.handle.net/10419/318846 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ 67 Business Systems Research | Vol. 15 No. 2 |2024 Sentiment and Stock Characteristics: Comprehensive Study of Individual Investor Influence on Returns, Volatility, and Trading Volumes Aleš Kresta, Jialei Xiong, Bahate Maidiya VSB – Technical University of Ostrava, Faculty of Economics, Czech Republic Abstract Background: Traditional asset pricing models face challenges from financial anomalies, prompting exploration through behavioural finance theory. This study analyses the nuanced relationship between individual investor sentiment and key stock market variables. Objectives: To assess the impact of individual investor sentiment on stock returns, volatilities, and trading volumes using the American Association of Individual Investors (AAII) sentiment index. Methods/Approach: Using regression models, we examine the relationship between individual investor sentiment and various stock characteristics across 480 components of the Standard & Poor's 500 index. Results: We find a positive relationship between the AAII sentiment index and stock returns and a negative relationship with volatility and trading volume. Conclusions: Our study contributes to understanding the intricate role of individual investor sentiment in financial markets. Keywords: investor sentiment; stock characteristics; behavioural finance; AAII sentiment index JEL classification: G12, G14, G41 Paper type: Research article Received: Jan 23, 2024 Accepted: Jul 26, 2024 Acknowledgements: The authors express gratitude for the financial support to the Czech Science Foundation (GACR) under project no. 23-06280S, VSB – Technical University of Ostrava under project no. SP2024/047, and the European Union under the REFRESH Research Excellence for Region Sustainability and High Tech Industries project no. CZ.10.03.01/00/22_003/0000048. Citation: Kresta, A., Xiong, J., & Maidiya, B. (2024). Sentiment and Stock Characteristics: Comprehensive Study of Individual Investor Influence on Returns, Volatility, and Trading Volumes. Business Systems Research, 15(2), 67-82. DOI: doi.org/10.2478/bsrj-2024-0018 68 Business Systems Research | Vol. 15 No. 2 |2024 Introduction For many years, traditional asset pricing models have dominated the assessment of risk-return trade-offs. However, the discovery of financial anomalies suggests that the efficient market hypothesis (Fama, 1970) may be challenged from the perspective of behavioural finance theory. Since the efficient market hypothesis (EMH) does not take into account the presence of investors' idiosyncratic behaviour (Haritha & Rishad, 2020), relying solely on EMH for asset pricing may lead to distortions. Investor sentiment, on the other hand, is popular for the empirical support it provides to asset pricing, emphasizing the impact of human biases on market behaviour. Because traditional theories do not account for the impact of abnormal investor behaviour on market outcomes, behavioural finance incorporates psychological perspectives into the description of financial markets so that we can gain a better understanding of why markets may deviate from the predictions of traditional theories such as EMH. Baker and Wurgler (2007, p. 129) broadly defined investor sentiment as 'a belief about future cash flows and investment risks that is not justified by the facts at hand.' Both researchers and practitioners are interested in measuring market sentiment, reflecting the overall sentiment of market participants or their subgroups. GonzálezSánchez and Morales de Vega (2021) identified three main approaches to constructing investor sentiment indices: aggregation of market variables, investor surveys, or utilizing information from the media. Each of them has its own advantages and disadvantages. The potential drawback of constructing a sentiment index through the aggregation of market variables is that it may include unrelated information. Investor surveys, while widely used, suffer from low observation rates, usually monthly or with a lower frequency, and reliability issues when nonresponse rates are high (Sun et al., 2016). The third, rapidly evolving approach involves explaining the return on assets through textual analysis of news, but there is no clear evidence of its explanatory capacity (González-Sánchez & Morales de Vega, 2021). In this paper, we focus on the second approach – measuring investor sentiment through investor surveys. Unlike other approaches, investor surveys provide a direct measurement of investor sentiment, as they involve directly asking and observing the sentiment among investors. Notable indexes for measuring investor sentiment in the US market include the monthly University of Michigan Consumer Sentiment Index, the weekly American Association of Individual Investors (AAII) sentiment survey, and the daily Investor Intelligence and Daily Sentiment Index. We concentrate our research on the AAII sentiment survey, which has a high number of survey participants and a long history of data since its inception in 1987. The objective of this study is to explore the relationship between individual investor sentiments and characteristics of the stock market, such as stock returns, volatility, and trading volumes. The research question is whether there is any relationship between the AAII sentiment survey index and the characteristics of the stocks in the periods following the publication of the sentiment index data. According to the efficient market hypothesis, all relevant information should already be priced in, and the sentiment should have no predictive power for future returns, which is our null hypothesis. The research addressed by this study also examines what should be used for prediction – whether the absolute value of the sentiment index or its change from the previous value. We hypothesize that changes in market sentiment are better predictors of future characteristics. For example, if sentiment improves from bearish to neutral, individual investors might start buying stocks and increase their bid and ask prices. On the contrary, if sentiment worsens from bullish to neutral, individual investors could start selling stocks. In both cases, the sentiment value is the same (neutral), but 69 Business Systems Research | Vol. 15 No. 2 |2024 the actions taken by individual investors are different. Therefore, changes in the sentiment index are likely more significant than absolute values. Our results have important implications for investors. We find a positive relationship between sentiment and future returns and a negative relationship with future volatility, suggesting that sentiment could be a useful indicator in developing investment or trading strategies. The findings contribute to understanding the role of individual investor sentiment in financial markets and its implications for investment strategies. However, it is important to approach these results with caution. Although our findings indicate a relationship between sentiment and stock returns, the sentiment variable used in our study does not capture the full spectrum of influencing factors. Therefore, relying solely on sentiment indicators for investment decisions may not consistently yield high returns, and investors should consider sentiment as one of many factors in their decision-making process. Our study differs from previous studies, which primarily concentrated on a single time series, usually the market index, see (Fisher & Statman, 2006; Y.-H. Wang et al., 2006; Kurov, 2008; Jacobs, 2015; Białkowski et al., 2023). We focus on a more robust dataset comprising the component stocks of the market index. Specifically, we use components of the Standard & Poor's 500 index. This approach allows for a comprehensive analysis that considers the dynamics and interactions within a broader range of securities, providing more robust results. The remainder of the paper is structured as follows. In the next section, we provide a short review of the literature. Then we introduce the data and methods applied. In the next sections, we present the results and their discussion. The last section presents the conclusion of the paper. Literature Review The selection of stock returns, volatilities, and trading volumes as dependent variables in this study is based on their fundamental importance in financial market analysis. Stock returns are a primary measure of a stock's performance and are crucial for investors, as they directly relate to the gains or losses experienced. Volatility, on the other hand, serves as a key proxy of risk, reflecting the degree of uncertainty, with higher volatility indicating greater risk. Trading volumes provide an important measure of market activity and liquidity, and higher trading volumes typically indicate greater market interest and ease of transacting without significantly affecting stock prices. Together, these variables are crucial factors for investors, as they directly impact investment decisions (Hawaldar & Rahiman, 2019; Veld & Veld-Merkoulova, 2008). These stock characteristics are also interrelated. A fundamental principle in finance is that investors require, or expect, higher returns for undertaking higher risks (represented by volatility). Research has also examined the relationship between trading volume and returns. Chen et al. (2001) and Naik and Sethy (2022) found a positive correlation between trading volumes and stock price changes, with trading volume contributing to the return process. Naik and Sethy (2022) also highlighted the asymmetric effect of stock returns on trading volume and the positive volume-volatility relationship. However, stock returns can also be explained by other factors than trading volume and volatility. Traditional pricing models, such as the Fama-French five-factor model (Fama & French, 2015, 2017), explain the stock returns based on factors related to market return, company size, book-to-market ratio, profitability, and investment style. Macroeconomic indicators can also serve as predictors; for instance, Hjalmarsson (2010) identified the short interest rate and term spread as robust predictors in developed markets. In addition to these fundamental and macroeconomic factors, 70 Business Systems Research | Vol. 15 No. 2 |2024 technical analysis provides another approach to understanding and predicting stock returns. Technical analysts believe that all relevant information is already reflected in stock prices and that price movements follow certain patterns that can be exploited. However, there is a broad academic debate on the applicability of technical analysis, as reviewed by Park and Irwin (2007). Traditional asset pricing models face growing challenges in their explanatory power and research in behaviour finance theory has further demonstrated the role of investor sentiment in driving the stock markets. Johnson and Tversky (1983) suggested that people make risky decisions based on their sentiment state. Compared with traditional asset pricing models, one of the main arguments of behaviour finance is that imperfectly rational traders (known as noise traders) generate deviations from fundamental values (Uygur & Taş, 2014) and that these deviations can significantly affect investor behaviour and market prices (Daniel et al., 2002). Focusing on investor sentiment is particularly important because this factor can lead to market anomalies that traditional models fail to explain. In recent years, research has increasingly explored the role of sentiment in returns, volatility, and trading volumes. Although the empirical results of Lee et al. (2002) suggest that sentiment is priced in systematic risk, the excess returns are still positively correlated with changes in sentiment. Additionally, the extent of bullish (bearish) sentiment changes leads to downward (upward) corrections in volatility and higher (lower) future excess returns. Uygur and Taş (2012) demonstrated that investor sentiment has a significant positive effect on the conditional volatility of the stock market during periods of high sentiment, whereas investor sentiment has a negative effect during periods of low sentiment. Audrino et al. (2020) investigated the effect of sentiment and attention indicators on daily stock market volatility and showed that sentiment and attention variables have significant predictive power for future volatility and that the addition of the sentiment variable leads to a further decrease in the mean square prediction error, especially on days with high volatility. By distilling the sentiment of the news text, Zhang et al. (2016) found that changes in sentiment, especially those with negative views, affect volatility and volume. Studies that investigate trading volume also show a relationship between investor sentiment and trading volume. From the results of So and Lei (2011), we can understand that the increase in the volatility index (VIX) is associated with an increase in trading volume, especially during periods when the VIX is high. This suggests that the higher the VIX level, the greater the change in trading volume. This is further confirmed by Lai et al. (2014), who found a positive correlation between investor sentiment and abnormal trading volumes. However, Kim and Ryu (2021) added a nuanced perspective by noting that the term structure of the impact of sentiment on trading volume is downward sloping, suggesting that instances of sentiment-induced trading anomalies are relatively short-living. However, empirical studies that use the AAII sentiment survey as a sentiment measure are scarce. Previous studies involving data from the AAII sentiment survey have shown that sentiment-driven investors often trade based on data from the AAII sentiment survey (Chau et al., 2016). The AAII sentiment index not only affects the stock price (Bouteska, 2019) but also significantly affects both the stock return and volatility (Sayim et al., 2013). Additionally, the AAII sentiment index plays a crucial role in the performance of initial public offerings (Ibrahim & Benli, 2022). 71 Business Systems Research | Vol. 15 No. 2 |2024 Methodology The sentiment index used in this study is obtained from the Investors Intelligence Survey (American Association of Individual Investors, 2024), which is conducted by the American Association of Individual Investors (AAII). The sentiment survey collects data from individual investors on their current market outlooks and investment decisions. Weekly, participants in the sentiment survey receive an email with a straightforward question: 'Do you feel the direction of the stock market over the next six months will be up (bullish), no change (neutral), or down (bearish)?' Participants can only submit one vote and their responses are used to calculate the indices, representing the percentages of bullish, bearish, and neutral market outlooks. By bullish outlook, we refer to the expectation of the participants that the stock market will grow in value. On the other hand, the bearish outlook represents the opinion of a future decline in the value of the stock market. Neutral perspectives are neither bullish nor bearish. The sentiment index (AAII sentiment index) is then calculated as the spread between the bullish and bearish percentages of votes, ranging from –100% to 100%. Additionally, in our research, we experiment with an alternative construction of the sentiment index (the ratio of bullish to bearish) and their week-by-week differences or differences between the index and its moving average, as we believe that the change in sentiment can be more important than its absolute value. It is crucial to note that this index is based on the opinions and investment decisions of individual investors and may not always align with broader market trends or sentiments. However, the survey offers valuable information on individual investors' perspectives and can help to understand the general outlook of the market. We assume that the individual investors are usually in a long position (Visaltanachoti et al., 2007), that is, they hold the stocks. As investor sentiment measures the beliefs of the market participant, we believe that the same market participants adjust their supply and demand for the stock accordingly. When they expect the market to soar, they increase the prices for which they are willing to buy and sell (ask and bid prices) as they consider the stocks to be undervalued at the current prices. They also increase the bid price, i.e., the price for which they are willing to buy because they speculate on the price increase. However, since there is positive market sentiment, fewer market participants are willing to sell, which decreases the trading volume and volatility. At the same time, we expect that as the sentiment in the market is negative or worsens, individual investors want to sell the stocks quickly, which both decreases the prices and increases the volatility and the trading volumes at the same time. Thus, we expect a positive relationship between sentiment and future returns and a negative relationship between sentiment and future return volatility and trading volume. Furthermore, for the stocks, our dataset comprises the components of the S&P 500 index (Wikipedia, 2024), and we obtained the daily adjusted close prices and volumes from the Yahoo Finance website (Yahoo, 2024). All data were collected for the period from January 1, 2000, to December 31, 2023. We chose this period as the trade-off between the length of the data, as the longer length provides more robust results because all market phases (bullish, neutral, bearish) are present in the data, and data availability because the composition of the index is changing during the time and some companies become no longer publicly trading and the data are not available. At the same time, since the newcomers to the index do have shorter price histories, we excluded stocks with time series lengths of less than 2,000 observations, resulting in a database reduction from 502 stocks to 480 stocks. From these, for 357 stocks, we used the full history of 6,101 daily observations that covered the last 24 years, while for the rest the time series length was shorter; see Figure 1. 72 Business Systems Research | Vol. 15 No. 2 |2024 Figure 1 Boxplot of the Lengths of Time Series Applied Source: Authors’ work As the sentiment data are weekly, we need to recalculate the daily data to weekly. The sentiment data are published each Thursday, and we assume that these data can be utilized to forecast the next week's characteristics; thus, we align the sentiment indexes published on Thursday with the subsequent Thursday-to-Thursday characteristics. We employ several regression models, each estimated via the Ordinary Least Squares (OLS) method, incorporating a Newey-West heteroskedasticity and autocorrelation consistent covariance matrix (Newey & West, 1987). The general regression equation is expressed as follows, 𝐸(𝑦𝑖,𝑡)= 𝛼𝑖+ 𝛽𝑖∙ 𝑥𝑡, (1) were, y represents the chosen dependent variable, i is the index that specifies the stock, t is the time, 𝛼𝑖 and 𝛽𝑖 are regression coefficients for stock i, intercept and slope respectively, and 𝑥𝑡 is the sentiment at time t. In our research, we examine four dependent variables (Y): return, risk premium, volatility, and trading volume. We also assume different specifications of the independent variable (X): the value of the sentiment index, week-to-week differences, and the differences between the index value and its five-week moving averages for the sentiment index calculated as both the spread and the ratio of bullish and bearish percentages of votes. For all these sentiment index series, the null hypothesis of a unit root is rejected by the Augmented Dickey-Fuller unit root test at 0.01 significance level. In each model, sentiment indices are used as an independent variable, while stock returns, trading volumes, and volatilities are used as a dependent variable. The first characteristic analysed is the one-week return calculated as a percentage change of the adjusted close prices p from Thursday to next Thursday, 𝑟𝑖,𝑡 =𝑝𝑖,𝑡 𝑝𝑖,𝑡−1 − 1. (2) According to the CAPM model, the returns can be divided into the risk-free rate and the risk premium. Thus, we also focus on one-week risk premiums, calculated as one-week returns minus the risk-free rates obtained from French (2024). The third dependent variable under consideration is the volatility of the returns. As it is not directly observable in the market, we estimate it ex-post from the returns using the generalized autoregressive conditional heteroskedasticity (GARCH) model (Bollerslev, 1986), specifically, we assume GARCH(1,1) specification: 𝑟𝑖,𝑡 = 𝜇𝑖+ σ𝑖,𝑡 ∙ 𝜖𝑖,𝑡, (3) 𝜎𝑖,𝑡 2= 𝜔𝑖+ 𝑎𝑖∙ 𝜎𝑖,𝑡−1 2+ 𝑏𝑖∙ 𝜖𝑡−𝑗 2, (4) 73 Business Systems Research | Vol. 15 No. 2 |2024 where 𝜇𝑖 is the mean return of the ith stock, 𝜎𝑖.𝑡 is the standard deviation (volatility) for the ith stock modelled by the GARCH model and 𝜖𝑖,𝑡 is a white noise. Parameters 𝜔𝑖, 𝑎𝑖 and 𝑏𝑖 need to be estimated. Positive variance is ensured if 𝜔𝑖> 0, 𝑎𝑖≥ 0, and 𝑏𝑖≥ 0, and the model is stationary if 𝑎𝑖+ 𝑏𝑖< 1 . The fourth dependent variable considered is the volume in US dollars traded in one week from Thursday to Thursday. In all characteristics, we align the newly announced values of the sentiment indices at time t, with the characteristics of the following week, that is, return, premium, volatility, and trading volume in the period from t to t+1. Results In this section, we present the results of the estimated regression models (1), which are carried out for the 480 component stocks of the S&P 500 index. The reported results include the number of stocks for which the estimated parameters are considered statistically significant at significance levels of 10%, 5% and 1% by means of the t-test and box plots of the parameter values. First, we focus on returns, where the dependent variable in Equation (1) is the weekly return. Table 1 illustrates the number of stocks for which the estimated parameters are statistically significant. As can be seen, the sentiment index calculated as the spread is a better predictor than the index calculated as the ratio. In fact, when using the spread, the slope is statistically significant in the case of 242 stocks out of 480, that is, for around half of the stocks, compared to only 107 in the case of ratio. When we transform the independent variable to differences, either week-to-week or valueto-average, the number of statistically significant parameters increases. We can observe that for 390 stocks from 480, the difference between the index calculated as spread and its previous month's average is statistically significant in predicting the future one-week return. Figure 2 shows the box plots of the parameters for all stocks. As can be seen, for all dependent variables except the ratio, both the intercept and the slope are positive in most of the stocks. We can conclude that there is a positive relationship between sentiment and subsequent one-week returns. Therefore, investors can use sentiment as a predictor of the return next week. Considering the best model (difference of spread to its MA), the median values can be interpreted as follows. The expected value of the weekly return from Thursday to Thursday is 0.33% (intercept) plus 4% (slope) for every 100 bps of the difference between the sentiment index value and its average in the previous four weeks. Alternatively, we can annualize the returns for better comparability. Then, the expected value of the next-week return is 16.96% p.a. plus 2.44% p.a. for every 1 bps of the difference between the sentiment value and its average in the previous four weeks. Table 1 Quantities of Statistically Significant Parameters of Regression of Return on Sentiment Independent variable 10% significance level 5% significance level 1% significance level Intercept Slope Intercept Slope Intercept Slope Spread 254 375 175 339 63 242 Spread diff. 404 443 343 420 202 349 Spread diff. MA 406 447 344 439 202 390 Ratio 48 296 27 223 3 107 Ratio diff. 403 289 342 225 201 129 Ratio diff. MA 403 369 342 333 202 225 Source: Authors’ work 74 Business Systems Research | Vol. 15 No. 2 |2024 Figure 2 Parameter Values of Return Regressions Note: Outliers positioned significantly far from the median have been excluded for the sake of clarity. Source: Authors’ work The same conclusions can be drawn when analysing the premiums, that is when we subtract the risk-free rate from the returns; see Table 2 and Figure 3. The median value of the intercept is 15.66% p.a. and the median value of the slope is 2.44% p.a. for every 1 bps of the difference between the sentiment value and its average in the previous four weeks. We can see that the slope has not changed while the intercept has changed by 1.3 % p.a., roughly equalling the risk-free return during the analysed period. Table 2 Quantities of Statistically Significant Parameters of Regression of Premium on Sentiment Independent variable 10% significance level 5% significance level 1% significance level Intercept Slope Intercept Slope Intercept Slope Spread 206 373 142 338 43 238 Spread diff. 371 443 300 420 160 348 Spread diff. MA 374 447 302 439 161 390 Ratio 54 286 27 216 6 103 Ratio diff. 369 289 300 225 160 129 Ratio diff. MA 370 369 300 333 160 225 Source: Authors’ work Figure 3 Parameter Values of Premium Regressions Note: Outliers positioned significantly far from the median have been excluded for the sake of clarity. Source: Authors’ work 81 Business Systems Research | Vol. 15 No. 2 |2024 41. Veld, C., & Veld-Merkoulova, Y. V. (2008). The risk perceptions of individual investors. Journal of Economic Psychology, 29(2), 226–252. https://doi.org/10.1016/j.joep.2007.07.001 42. Visaltanachoti, N., Lu, L., & (Robin) Luo, H. (2007). Holding periods, illiquidity and disposition effect in the Chinese stock markets. Applied Financial Economics, 17(15), 1265–1274. https://doi.org/10.1080/09603100600905053 43. Wang, W., Su, C., & Duxbury, D. (2022). The conditional impact of investor sentiment in global stock markets: A two-channel examination. Journal of Banking & Finance, 138, 106458. https://doi.org/10.1016/j.jbankfin.2022.106458 44. Wang, Y.-H., Keswani, A., & Taylor, S. J. (2006). The relationships between sentiment, returns and volatility. International Journal of Forecasting, 22(1), 109–123. https://doi.org/10.1016/j.ijforecast.2005.04.019 45. Wikipedia. (2024). List of S&P 500 companies. In Wikipedia. https://en.wikipedia.org/w/index.php?title=List_of_S%26P_500_companies&oldid=1193 421148 46. Yahoo. (2024). Yahoo Finance—Stock Market Live, Quotes, Business & Finance News. https://finance.yahoo.com/ 47. Zhang, J. L., Härdle, W. K., Chen, C. Y., & Bommes, E. (2016). Distillation of News Flow Into Analysis of Stock Reactions. Journal of Business & Economic Statistics, 34(4), 547– 563. https://doi.org/10.1080/07350015.2015.1110525 82 Business Systems Research | Vol. 15 No. 2 |2024 About the authors Aleš Kresta serves as an Associate Professor in the Department of Finance at the Faculty of Economics, VSB—Technical University of Ostrava, where he earned his Ph.D. in Finance. His primary research interests encompass risk estimation, backtesting, portfolio optimisation, financial time series modelling, and soft computing within the realm of quantitative finance. Actively involved in numerous scientific projects, he can be contacted at ales.kr[email protected]. Jialei Xiong is a Ph.D. student in the Department of Finance at the Faculty of Economics, VSB – Technical University of Ostrava. Her research and dissertation focus on issues related to portfolio optimisation, specifically emphasising sentiment measures. The author can be contacted at jialei.xion[email protected]. Bahate Maidiya, is a PhD student in the Department of Finance at the Faculty of Economics, VSB – Technical University of Ostrava. She directs her research toward biases in investment decision-making. The author can be contacted at bahate.maidiy[email protected].