As an important component of the financial market, the stock market’s fluctuations are influenced by multiple factors. Among them, investor sentiment, as one of the important factors affecting the fluctuations of the stock market, has received extensive attention in recent years (
As big data and artificial intelligence technologies have advanced at a rapid pace, deep learning and text mining techniques have offered fresh perspectives and innovative approaches for conducting research on investor sentiment (
Numerous research endeavors have delved into examining how investor sentiment influences stock returns. In the early stages, these studies predominantly relied on data sources like questionnaires or market indexes, and they uncovered a definite connection between investor sentiment and stock returns (
Over recent years, as big data and artificial intelligence technologies have advanced, an increasing number of studies have started to employ deep learning and text mining methods for the purpose of analyzing investor sentiment (
Behavioral finance incorporates psychological principles into financial studies, overcoming the constraints imposed by the “rational person” hypothesis in conventional finance. It posits that investors are not fully rational and are swayed by elements like cognitive distortions and emotional shifts during their decision-making process. Investor sentiment, as an important research content of behavioral finance, reflects investors’ optimistic or pessimistic attitude towards the market. This emotional state can distort investors’ judgment of stock value, causing stock prices to deviate from their intrinsic value. For instance, overconfident investors might overestimate the future returns of stocks, thereby driving up stock prices. However, overly pessimistic investors may underestimate the value of stocks, triggering a decline in stock prices. The theory of behavioral finance provides an important theoretical basis for understanding the impact of investor sentiment on stock returns and reveals the mechanism of investors’ irrational behavior in the stock market.
The market efficiency theory posits that within an efficient market, stock prices are capable of completely mirroring all accessible information. Nevertheless, the presence of investor sentiment might undermine the market’s efficiency. When investor sentiment tends to be broadly optimistic, a substantial influx of funds surges into the market, propelling stock prices higher. At this time, stock prices may be higher than their intrinsic value, and the market shows an overreaction. Conversely, when investors are pessimistic, a large amount of funds flows out, stock prices fall, and stock prices may be lower than their intrinsic value, resulting in an insufficient market response. The market inefficiency stemming from investor sentiment gives rise to an intricate connection between stock returns and sentiment, thus offering a theoretical foundation for exploring the nonlinear influence of investor sentiment on stock returns. The herd effect refers to the fact that when information is uncertain, investors tend to imitate others’ investment decisions instead of making investments based on their own independent analysis. In the stock market, when the sentiment of some investors changes, it may trigger the herd behavior of other investors, leading to the rapid spread and amplification of market sentiment. For instance, when some well-known investors or institutions show optimism about a certain stock and purchase it in large quantities, other investors may follow suit and buy it, driving up the stock price significantly. The herd effect makes the impact of investor sentiment on stock returns collective and contagious, further intensifying the volatility and nonlinearity of the stock market.
As the Internet has become more widespread and social media has advanced, a vast quantity of textual data has surfaced, furnishing abundant data resources for research into investor sentiment (
Stock market data encompasses a wide range of elements, such as share prices, trading volumes, price-to-earnings ratios, and price-to-book ratios, among others. These data points are capable of mirroring the overall functioning of the stock market as well as the performance of individual stocks. Stock prices are the most direct indicator reflecting stock returns, while trading volume can reflect the market’s activity level and investors’ participation. Assessment metrics like the price-earnings ratio and the price-to-book ratio can assist investors in figuring out whether the stock price is fair or reasonable. Stock market data of different types are interrelated and jointly constitute an important data basis for studying the relationship between investor sentiment and stock returns. Through the comprehensive analysis of these data, a more comprehensive understanding of the influence mechanism of investor sentiment on stock returns can be achieved.
To thoroughly examine the connection between investor sentiment and stock returns, it is essential to combine and perform a correlation analysis on both text data and stock market data. For the purpose of analyzing how shifts in sentiment influence stock returns, the investor sentiment details within the text data must align with the return information present in the stock market data. During the process of data integration, issues such as the temporal synchronization of data and the consistency of data need to be considered. Correlation analysis can explore the potential relationship between investor sentiment and stock returns through statistical methods, machine learning algorithms, etc. For instance, the distribution characteristics of stock returns under different emotional states can be analyzed, or the correlation between emotional changes and fluctuations in stock returns can be explored. Through data integration and correlation analysis, the nonlinear impact of investor sentiment on stock returns can be revealed more accurately.
An emotional dictionary is a kind of dictionary that contains both positive and negative words. By calculating the quantity and proportion of positive and negative words in the text, the emotional score of the text can be obtained. In investor sentiment analysis, a specialized financial sentiment dictionary can be constructed, including both positive and negative terms related to the stock market. Then, sentiment analysis is conducted on the preprocessed text data to calculate the sentiment score of each text, and thereby obtain the investor sentiment indicators at each time point. This approach is straightforward and easy to grasp, yet it might be constrained by the extent of coverage and the level of accuracy offered by the sentiment dictionary.
Machine learning methods can automatically identify the sentiment tendencies in text by training models. Frequently employed machine learning algorithms encompass Support Vector Machine (SVM), Naive Bayes, Random Forest, and the like. In the context of investor sentiment analysis, labeled textual data can be utilized to train machine learning models, empowering them to precisely differentiate between texts with positive, negative, and neutral tones. Subsequently, the trained model is applied to carry out sentiment classification on fresh textual data, and the percentage of each text type is computed to serve as an addition to the sentiment index. This method can automatically learn the emotional features in the text, with high accuracy and flexibility, but it requires a large amount of labeled data for model training.
Deep learning methods have achieved remarkable results in text sentiment analysis. Widely adopted deep learning models comprise Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), along with their variations (like LSTM, GRU), and so on. These models are capable of handling long sequence text data and capturing semantic information and emotional features in the text. In investor sentiment analysis, deep learning models can be used to perform sentiment classification or sentiment score prediction on text data to obtain more accurate sentiment indicators. Deep learning methods have strong learning ability and generalization ability, but they require a large amount of computing resources and time for model training.
Through statistical analysis of the extracted investor sentiment indicators, one can understand the distribution characteristics, mean values, variances and other statistics of the sentiment indicators. These statistics can reflect the overall level and fluctuations of investor sentiment. For instance, a higher average value of sentiment indicators may indicate that the overall market sentiment is relatively optimistic, while a larger variance may suggest that market sentiment fluctuates more intensely.
Through the application of the correlation analysis approach, we can delve into the connection between investor sentiment indicators and stock returns. By computing the correlation coefficient between these sentiment indicators and stock returns, we can initially ascertain whether a linear relationship exists between the two. Nevertheless, given that the link between investor sentiment and stock returns might be nonlinear, correlation analysis merely offers a preliminary insight. Further analysis can adopt nonlinear correlation coefficients or visualization methods, such as scatter plots and heat maps, to present the relationship between the two more intuitively.
Investor sentiment is dynamically changing and its changes may be influenced by multiple factors. By analyzing the dynamic changes of emotional indicators, one can understand the changing trends and cyclical characteristics of emotions. For example, the time series analysis method can be utilized to model the emotional indicators and predict the future changing trend of emotions. Additionally, we can analyze how sentiment indicators evolve across various market phases (like bull markets and bear markets) to investigate the connection between investor sentiment and market trends.
Investor sentiment’s influence on stock returns might exhibit a threshold effect. In other words, once the sentiment indicator hits a specific level, the impact of sentiment on stock returns will undergo a notable change. For instance, when market sentiment is extremely optimistic, investors may overbuy stocks, driving up stock prices significantly. When market sentiment is extremely pessimistic, investors may sell stocks excessively, causing a significant drop in stock prices. This threshold effect can be verified by setting different emotional thresholds and analyzing the impact of emotions on stock returns within different threshold intervals.
Shifts in investor sentiment might not be instantly mirrored in stock returns; instead, they could exhibit a certain delay or lag effect. That is, the current emotional state may have an impact on future stock returns. This lag effect might be due to the fact that the decision-making process of investors requires a certain amount of time, or the dissemination and digestion of market information also takes a certain amount of time. By scrutinizing the time-series connection between sentiment indicators and stock returns, we can delve into whether a lag effect exists and gauge its intensity.
The influence of investor sentiment on stock returns might exhibit an asymmetric impact. In other words, the degree to which positive (optimistic) and negative (pessimistic) sentiments impact stock returns could vary. For instance, pessimism may have a greater negative impact on stock returns because investors are more likely to engage in panic selling when they are pessimistic. The existence of the asymmetric effect can be verified by comparing the changes in stock returns under optimistic and pessimistic sentiments.
The psychological factors of investors are one of the important reasons that lead to the nonlinear impact of investor sentiment on stock returns. Psychological factors such as investors’ cognitive biases and emotional fluctuations can affect their interpretation of market information and decision-making behaviors. For example, investors may be overly confident and ignore risks when they are optimistic. When pessimistic, one may be overly fearful and magnify the risks. The effect of this psychological factor can lead to the influence of emotions on stock returns, showing nonlinear characteristics.
The operating mechanism of the stock market can also affect the nonlinear impact of investor sentiment on stock returns. For instance, factors such as market liquidity and transaction costs can influence investors’ trading behaviors. When market liquidity is poor, changes in investors’ sentiment may be more likely to cause significant fluctuations in stock prices. When transaction costs are high, changes in investors’ sentiment may have a relatively small impact on stock returns. In addition, market regulatory policies, information disclosure systems, etc. will also have an impact on investors’ emotions and behaviors, and thereby affect stock returns.
External disturbances, like alterations in the macroeconomic landscape, policy modifications, and unforeseen occurrences, can exert a substantial influence on investor sentiment and the stock market. These external factors might cause abrupt shifts in investor sentiment, subsequently prompting irrational volatility in the stock market.
For example, when an economic crisis erupts, it can result in highly pessimistic investor sentiment and spark a stock market plunge. Conversely, the implementation of beneficial policies could foster optimistic investor sentiment and propel stock prices higher.
The impact of these external shocks will amplify the nonlinear effect that investor sentiment has on stock returns.
Investors should recognize the nonlinear impact of investor sentiment on stock returns and enhance their risk awareness. When market sentiment is extremely optimistic, be vigilant against possible market bubbles and avoid excessive investment. When market sentiment is extremely pessimistic, one should remain calm and avoid blind selling. At the same time, the investment portfolio should be reasonably adjusted according to the changes in sentiment indicators to reduce investment risks.
Due to the nonlinear and uncertain impact of investor sentiment on stock returns, short-term speculative behavior often faces greater risks. Therefore, investors should establish the concept of long-term investment, pay attention to the fundamentals and long-term value of stocks, and avoid being affected by short-term emotional fluctuations. Through long-term investment, the impact of emotional fluctuations on investment returns can be smoothed out, achieving stable appreciation of assets.
Diversified investment is one of the effective strategies to reduce investment risks. Investors can diversify the impact of investor sentiment on a single stock or industry by investing in stocks of different industries and regions, as well as allocating to other asset classes (such as bonds, funds, etc.). Diversified investment can reduce the overall risk of the investment portfolio and enhance the stability of investment returns.
This research carries out a comprehensive examination of the nonlinear influence of investor sentiment on stock returns, leveraging deep learning and text mining approaches. Behavioral finance, market efficiency theory, and herd effect theory collectively offer a robust theoretical foundation for investigating the connection between investor sentiment and stock returns. Behavioral finance elucidates how investors’ irrational behaviors shape their judgments of stock value and contribute to price fluctuations. Market efficiency theory posits that investor sentiment can potentially give rise to market inefficiency, thereby establishing an intricate relationship between stock returns and sentiment. The herd effect underscores the collective and contagious nature of investor sentiment, exacerbating the volatility and nonlinearity of the stock market. These theories are mutually complementary and together establish a theoretical framework for comprehending the nonlinear impact of investor sentiment on stock returns.
With the proliferation of the Internet and social media, text data has emerged as a rich and up-to-date source for researching investor sentiment. Stock market data encompasses various dimensions, including stock prices, trading volumes, price-earnings ratios, and price-to-book ratios, offering a holistic view of the stock market’s operations. By integrating and conducting correlation analysis on text data and stock market data, we can more precisely uncover the nonlinear impact of investor sentiment on stock returns. When it comes to extracting investor sentiment indicators, the sentiment dictionary method is straightforward and easy to understand. However, its effectiveness is constrained by the dictionary’s coverage and accuracy. Machine learning methods are capable of automatically learning emotional features, boasting high accuracy and flexibility, yet they demand a substantial amount of labeled data. Deep learning methods exhibit strong learning and generalization capabilities, but they necessitate significant computing resources and time for model training. Each extraction method has its own strengths and weaknesses, and the appropriate method can be chosen based on research requirements and data availability.
This study, by employing deep learning and text mining techniques, delves deeply into the nonlinear impact of investor sentiment on stock returns, equipping investors with more sophisticated sentiment analysis tools and a scientific basis for making investment decisions. Nevertheless, this study is not without limitations, such as constraints related to data sources and the complexity of model construction. Future research endeavors can further broaden data sources, refine the model structure, enhance the accuracy and reliability of the research, and offer more valuable insights for investors.