The world of finance is about to change. Having a significant impact on other commercial sectors, the stock market serves as a major mechanism and key factor for companies to increase capital. With American stock market assets which should reach 40 billions of dollars in 2025, or 1.5 times the country's GDP, it represents an important part of the global economy, highlighting its important position as a reference for the American economic landscape.
Two eminent scientists of the South Dakota State University – Kaiqun Fu, an assistant professor experienced in the Department of Electric and Computer Science, with the graduate student researcher, Yanxiao Bai, presented a very interesting artificial intelligence model called “Alerta -Net”. This deep neural network has a unique advantage by combining macroeconomic indicators, search engine data and a social media context. This raises the question: can AI tools be the key to predicting the movements of shares and market volatility? Let us take a closer look at these innovative studies and their potential impact on the financial landscape.
Their research focuses on the actions of Blue Chip, which reflect a wider stock market dynamic. Blue Chip's actions are actions issued by financially solid and well -established companies with an impeccable reputation.
Researchers have selected 41 Blue Chip shares in 10 sectors according to the global standard of classification of industry (GICS) to study the financial market. Each of these actions is considered to be an investment perspective by Moody's and S&P. Given the difficulties known to predict the action courses, scientists have decided to use Blue Chip actions to predict the movements of the prices of future actions and the trend of volatility.
Two main methodologies prevail in stock market studies: technical and fundamental analysis. Technical analysis uses the prices of previous actions to predict future trends. However, its high dependence on historical data can sometimes miss sudden market changes due to unexpected events. Assuming uniformly rational market behavior, this methodology can sometimes create a random echo chamber, which makes commercial signals isolated from the real economic context.
The fundamental analysis, on the other hand, integrates both the characteristics of the prices and external information, including data from social networks and search engines. This approach demonstrates better precision in the forecast of the closure of the S&P 500 index when integration of Twitter data in its model. Although these data sources often reflect not only the financial market but also important economic indicators, predominant research in fundamental analysis tends to underline the financial market, by neglecting the symbiotic relationship between the global economy and the stock market. In addition, existing models are mainly focused on forecasting changes in trends, often neglecting the importance of the scale of these changes. But in the field of stock behavior, the scale of these changes is important.
In their article, published On the Arxiv preparation server, researchers offer a new model that combines the two approaches. Alerta -Net – Attentional time distance aware of recurring neural networks. The proposed framework makes it possible to combine data from social networks, macroeconomic indicators and information from search engines to predict the movements of stock prices and volatility.
Consider the general architecture of the alert-net. The data input and pre -treatment layer converts temporal and textual information into dense vectors. Then, the temporal layer sensitive to the distance has a recurring representation which identifies the hidden dependencies in the current data of the actions according to the information passed. The context of the distance matrix then integrates these historical dependencies into a sequence of characteristics. And finally, the forecasting layer generates movement forecasts of actions and volatility adjusted to time for the next time interval, thus providing a complete and coherent system for predicting actions.
Alerta-Net architecture is designed to predict movement and volatility on day T. The data input and pre -treatment phase extracts the textual information from Twitter and converts it into feeling. Alerta-Net then uses these feeling scores as well as other features to make predictions based on temporal distance.
In order to validate the effectiveness of the proposed model, experiences and comparisons have been made on a real data set. The data set provides feeling scores for 41 actions from Blue Chip and combines three main components: price data, Twitter feeling data and macroeconomic data.
Alerta-Net's performance has been assessed by comparison with DP-MSTM, a well-known stock prediction network based on financial data. The other benchmarks used in the study include strengthening the extreme gradient, the LSTM based on attention and the GRU. The results are defined in terms of precision and the Matthews correlation coefficient. Given that data points involving the action changes in the action of more than 5% were only a small part of the data set, it was decided to use the zone under the ROC curve (AUC) as a performance measure to obtain a more reliable and realistic forecast. The results are presented in the table below.
With Alerta-Net, it becomes possible not only to predict the movements of equity prices, but also to effectively obtain information on the volatility of the stock markets. This allows you to anticipate all unusual stock market fluctuations in the future.
Alerta-Net has proven itself in the recognition of dynamic, temporal and remote relationships integrated in various hidden states. Using the movement's movements the same day, the model considerably increases its precision in the forecast of the volatility of the stock markets.
Overall, Alerta-Net, a deep generative neural network architecture, has demonstrated the effectiveness of the combination of search engine data, macroeconomic indicators and social media data to try to predict the movements of actions and volatility. In future studies, researchers plan to improve precision by integrating several sources of text and audio, including profits and financial statements.
