Abstract
Recent advances in natural language processing have contributed to the development of market sentiment measures through text content analysis in news providers and social media. The effectiveness of these sentiment variables depends on the implemented techniques and the type of source on which they are based. In this paper, we investigate the impact of the release of public financial news on the S&P 500. Using automatic labeling techniques based on either stock index returns or dictionaries, we apply a classification problem based on long short-term memory neural networks to extract alternative proxies of investor sentiment. Our findings provide evidence that there exists an impact of those sentiments in the market in a 20-min time frame. We find that dictionary-based sentiment provides meaningful results that outperform those based on stock index returns, which partly fails in the mapping process between news and financial returns.
Similar content being viewed by others
Notes
According to the authors, net buying pressure can be defined as “the difference between the number of buyer-initiated trades and the number of seller-initiated trades calibrated from the bid-ask quotes”.
To avoid the look-ahead bias, the testing set is both out-of-sample and out-of-time in order to ensure that the training and validation sets contain observations earlier in time than the observations in the testing set.
Further information can be found at https://www.merriam-webster.com/words-at-play/longest-words-ever.
The integer values are discussed in the Sect. 4.1 on the classification results of the stock index returns approach.
An alternative approach would be to use value thresholds for returns, denoted as \(t_1<0\) and \(t_2>0\), by imposing, for example, \(t_1 = -t_2\) or possibly \(t_1 \ne -t_2\). Although using value thresholds for positive and negative returns may appear to be a simple and immediate solution for categorization, it raises several issues related to selecting appropriate threshold values. The selection of threshold magnitudes requires careful consideration specific to the financial asset being studied and the data frequency, which ultimately involves analyzing distributional features. The choice of value thresholds should be specific to the dataset, making it necessary to analyze each case carefully. The only exception to this would be in the case of positive/negative categorizations, where a single threshold centered at zero can be used to identify the two classes.
Full results are available upon request to the authors.
GARCH type models have been successfully applied in measuring the relationship between investor sentiment and stock market returns-volatility (i.e., Rupande et al. 2019).
Full results are available upon request to the authors.
Additionally, we have repeated the analysis of the changes in the VIX which represents the forward-looking volatility of the S&P 500. Results are included in Appendix 1 and are qualitatively similar.
The “correct” class of each article is the one that has been previously assigned through automatic labeling.
We thank an anonymous Reviewer for this beneficial reference.
References
Atkins A, Niranjan M, Gerding E (2018) Financial news predicts stock market volatility better than close price. J Finance Data Sci 4:120–137
Behrendt S, Schmidt A (2018) The twitter myth revisited: intraday investor sentiment, twitter activity and individual-level stock return volatility. J Banking Finance 96:355–367
Black F (1976) Studies of stock price changes. In: Proceeding of the 1976 meetings of the American Statistical Association, pp 177–181
Caporin M, Poli F (2017) Building news measures from textual data and an application to volatility forecasting. Econometrics 5:1–46
Chen S, Guo Z, Zhao X (2021) Predicting mortgage early delinquency with machine learning methods. Eur J Oper Res 290:358–372
Chung S-L, Hung C-H, Yeh C-Y (2012) When does investor sentiment predict stock returns? J Empir Financ 19:217–240
Costola M, Iacopini M, Santagiustina CRMA (2020) Google search volumes and the financial markets during the COVID-19 outbreak. Finance Res Lett 42:101884
Feng L, Fu T, Shi Y (2022) How does news sentiment affect the states of Japanese stock return volatility? Int Rev Financ Anal 84:102267
Frugier A (2016) Returns, volatility and investor sentiment: evidence from European stock markets. Res Int Bus Financ 38:45–55
Garcia D (2013) Sentiment during recessions. J Financ 68:1267–1300
Groß-Klußmann A, Hautsch N (2011) When machines read the news: using automated text analytics to quantify high frequency news-implied market reactions. J Empir Financ 18:321–340
Hajek P, Myskova R, Olej V (2021) Predicting stock return volatility using sentiment analysis of corporate annual reports. In: The essentials of machine learning in finance and accounting. Routledge, pp 75–96
Harrison J (2022) R-package ‘rselenium’. https://github.com/ropensci/RSelenium
Harvard University (1960) General inquirer. http://www.wjh.harvard.edu/~inquirer/
Henry E (2008) Are investors influenced by how earnings press releases are written? J Bus Commun 45:363–407
Houlihan P, Creamer GG (2017) Can sentiment analysis and options volume anticipate future returns? Comput Econ 50:669–685
Huang X, Zhang W, Tang X, Zhang M, Surbiryala J, Iosifidis V, Liu Z, Zhang J (2021) Lstm based sentiment analysis for cryptocurrency prediction. In: International conference on database systems for advanced applications. Springer, pp 617–621
Iacopini M, Santagiustina CR (2021) Filtering the intensity of public concern from social media count data with jumps. J Roy Stat Soc Ser A (Stat Soc)
Jiang GJ, Tian YS (2005) The model-free implied volatility and its information content. Rev Financ Stud 18:1305–1342
Jin Z, Yang Y, Liu Y (2020) Stock closing price prediction based on sentiment analysis and lstm. Neural Comput Appl 32:9713–9729
Karpathy A (2015) CS231n convolutional neural networks for visual recognition. Linear classification: Support Vector Machine, Softmax classifier. http://cs231n.github.io/linear-classify/#softmax
Li X, Xie H, Chen L, Wang J, Deng X (2014) News impact on stock price return via sentiment analysis. Knowl Based Syst 69:14–23
Liu Y, Qin Z, Li P, Wan T (2017) Stock volatility prediction using recurrent neural networks with sentiment analysis. In: International conference on industrial. Springer, Engineering and Other Applications of Applied Intelligent Systems, pp 192–201
Loughran T, McDonald B (2011) When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. J Financ 66:35–65
Loughran T, McDonald B (2015) The use of word lists in textual analysis. J Behav Financ 16:1–11
Mandal PK, Mahto R (2019) Deep cnn-lstm with word embeddings for news headline sarcasm detection. In: 16th International conference on information technology-new generations (ITNG 2019). Springer, pp. 495–498
Mangee N (2018) Stock returns and the tone of marketplace information: Does context matter? J Behav Financ 19:396–406
Medhat W, Hassan A, Korashy H (2014) Sentiment analysis algorithms and applications: a survey. Ain Shams Eng J 5:1093–1113
Nigam K, Lafferty J, McCallum, A (1999) Using maximum entropy for text classification. In: IJCAI-99 workshop on machine learning for information filtering, Stockholom, Sweden, vol. 1, pp 61–67
Pennington J, Socher R, Manning CD (2014) Glove: global vectors for word representation. In: Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pp 1532–1543
Renault T (2017) Intraday online investor sentiment and return patterns in the us stock market. J Banking Finance 84:25–40
Rupande L, Muguto HT, Muzindutsi P-F (2019) Investor sentiment and stock return volatility: evidence from the Johannesburg stock exchange. Cogent Econ Finance 7:1–16
Schumaker RP, Chen H (2009) Textual analysis of stock market prediction using breaking financial news: the azfin text system. ACM Trans Inf Syst (TOIS) 27:1–19
Shi Y, Ho K-Y, Liu W-M (2016) Public information arrival and stock return volatility: evidence from news sentiment and Markov regime-switching approach. Int Rev Econ Finance 42:291–312
Sohangir S, Wang D, Pomeranets A, Khoshgoftaar TM (2018) Big data: deep learning for financial sentiment analysis. J Big Data. https://doi.org/10.1186/s40537-017-0111-6
Souma W, Vodenska I, Aoyama H (2019) Enhanced news sentiment analysis using deep learning methods. J Comput Soc Sci 2:33–46
Uysal AK, Gunal S (2014) The impact of preprocessing on text classification. Inf Process Manag 50:104–112
Vicari M, Gaspari M (2021) Analysis of news sentiments using natural language processing and deep learning. AI Soc 36:931–937
Wan X, Yang J, Marinov S, Calliess J-P, Zohren S, Dong X (2021) Sentiment correlation in financial news networks and associated market movements. Sci Rep 11:1–12
Wang G, Wang T, Wang B, Sambasivan D, Zhang Z, Zheng H, Zhao BY (2015) Crowds on Wall Street: Extracting value from social investing platforms. In: Proceedings of the 18th ACM conference on computer supported cooperative work and social computing. ACM, pp 17–30
Wang C, Wang T, Yuan C, Rong JY (2022) Learning to trade on sentiment. J Econ Finance 46:308–323
Wickham H (2016) R-package ‘rvest’, p 156. https://cran.r-project.org/web/packages/rvest/rvest.pdf
Xing FZ, Cambria E, Welsch RE (2018) Natural language based financial forecasting: a survey. Artif Intell Rev 50:49–73
Yadav R, Kumar AV, Kumar A (2019) News-based supervised sentiment analysis for prediction of futures buying behaviour. IIMB Manag Rev 31:157–166
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank very much the Associate Editor and two anonymous Reviewers for their valuable suggestions and insightful remarks. We also thank Luca Coraggio of the University of Naples, Tomasz Gubiec of the University of Warsaw, Dian Kusumaningrum of Prasetiya Mulya University, Giovanni Zambruno of the University of Milano-Bicocca, and the other participants of the 2nd One-Day Workshop on Machine Learning for Finance, held at the Ca’ Foscari University of Venice in 2020.
A: Stock index returns approach using the VIX
A: Stock index returns approach using the VIX
In this appendix, we replicate the stock index returns approach using the VIX as the matching variable. The VIX is considered a superior predictor of historical volatility since it is based on option prices that reflect the future expectations of market participants (see, for instance, Jiang and Tian 2005).
Due to shock asymmetry, volatility is usually higher when the S&P 500 Index returns are negative and might be lower when the S&P 500 Index returns are positive. Following the analysis in the main text, we do not impose any asymmetry in the weighting structure and therefore, process news for the VIX following the same method as for returns. As discussed in the paper, news articles were classified as positive in case of log returns higher than the 55th percentile, negative in case of log returns lower than the 45th percentile, and neutral otherwise. The results are presented in Table 12. The period considered and number of articles analyzed were the same as for stock returns. Table 13 shows the accuracy of the results for the four models according to the time windows and the presence of the lagging method. Finally, Table 14 shows the four sentiment variables built on the basis of the classification methods presented in Table 13.
The estimates for the EGARCH model are presented in Tables 15 and 16 for \(S_{vix20}\) and \(S_{vix20lag}\), respectively. In both cases, we found similar evidence as for the sentiment built on the S&P 500. For instance, \(\lambda _m\) is significant and very close to zero in both \(S_{vix20}\) and \(S_{vix20lag}\) in seven and nine of the ten cases, respectively. In addition, in this case, there is a discordant sign among the different versions of the sentiment, confirming that the approach based on stock index returns is highly sensitive to the initial settings. Analogously, \(\lambda _m\) is significant in six and nine of ten cases for \(S_{vix20}\) and \(S_{vix20lag}\), respectively. As the tables shows, the coefficient exhibits different signs according to the different versions of sentiment. Also in the paper, we conclude that stock index returns do not represent a reliable approach since it fails in the mapping process between news and financial returns.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Anese, G., Corazza, M., Costola, M. et al. Impact of public news sentiment on stock market index return and volatility. Comput Manag Sci 20, 20 (2023). https://doi.org/10.1007/s10287-023-00454-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10287-023-00454-2