Financial Text Research Articles

FinBERT: A Large Language Model for Extracting Information from Financial Text*

ABSTRACTWe develop FinBERT, a state‐of‐the‐art large language model that adapts to the finance domain. We show that FinBERT incorporates finance knowledge and can better summarize contextual information in financial texts. Using a sample of researcher‐labeled sentences from analyst reports, we document that FinBERT substantially outperforms the Loughran and McDonald dictionary and other machine learning algorithms, including naïve Bayes, support vector machine, random forest, convolutional neural network, and long short‐term memory, in sentiment classification. Our results show that FinBERT excels in identifying the positive or negative sentiment of sentences that other algorithms mislabel as neutral, likely because it uses contextual information in financial text. We find that FinBERT's advantage over other algorithms, and Google's original bidirectional encoder representations from transformers model, is especially salient when the training sample size is small and in texts containing financial words not frequently used in general texts. FinBERT also outperforms other models in identifying discussions related to environment, social, and governance issues. Last, we show that other approaches underestimate the textual informativeness of earnings conference calls by at least 18% compared to FinBERT. Our results have implications for academic researchers, investment professionals, and financial market regulators.

Research on Extraction Method of Financial Knowledge Based on How Net

Abstract In order to obtain the knowledge information of financial texts more efficiently and make the extracted information such as entity relation attribute more accurate, this paper studies the grammatical features of financial news texts and the semantic features of How Net, and puts forward the scheme of financial information extraction based on How Net. First, the phrase matching is carried out in the dictionary. Then the neural network is used for weighting, BiLSTM is used for character vector feature enhancement training, and then conditional random field (CRF) is used to complete named entity recognition, and then the relationship extraction of entity pairs from the dependency syntax is carried out to complete the research on the construction method of knowledge extraction of text in the financial field. The experimental results show that this model is superior to the other three models in entity recognition, and the overall performance is improved by about 1.2%. In relation extraction, the accuracy and recall rate of the model algorithm adopted in this paper are improved by 5% and 1.5% respectively, which shows that the improvement of the algorithm is effective.

A deep learning approach of financial distress recognition combining text

<abstract><p>The financial distress of listed companies not only harms the interests of internal managers and employees but also brings considerable risks to external investors and other stakeholders. Therefore, it is crucial to construct an efficient financial distress prediction model. However, most existing studies use financial indicators or text features without contextual information to predict financial distress and fail to extract critical details disclosed in Chinese long texts for research. This research introduces an attention mechanism into the deep learning text classification model to deal with the classification of Chinese long text sequences. We combine the financial data and management discussion and analysis Chinese text data in the annual reports of 1642 listed companies in China from 2017 to 2020 in the model and compare the effects of the data on different models. The empirical results show that the performance of deep learning models in financial distress prediction overcomes traditional machine learning models. The addition of the attention mechanism improved the effectiveness of the deep learning model in financial distress prediction. Among the models constructed in this study, the Bi-LSTM+Attention model achieves the best performance in financial distress prediction.</p></abstract>

Smart Analysis of Economics Sentiment in Spanish Based on Linguistic Features and Transformers

Texts related to economics and finances are characterized by the use of words and expressions whose meaning (and the sentiments they convey) substantially depend on the context. This poses a major challenge to Natural Language Processing tasks in general, and Sentiment Analysis in particular. For low-resource languages such as Spanish, this situation becomes even more acute. Yet, the latest advancements in the field, including word embeddings and transformers, have allowed to boost the performance of Sentiment Analysis solutions. In this work we explore the impact of the combination of different feature sets in the accuracy of Sentiment Analysis in Spanish financial texts. For this, a corpus with 15,915 tweets has been compiled and manually annotated as either positive, negative, or neutral. Then, feature sets based on contextual and non-contextual embeddings along with linguistic features were evaluated both individually and combined. The best results, with a weighted F1-score of 73.15880%, were obtained with a combination of feature sets by means of knowledge integration.

Sentiment Classification of Financial Texts for Stock Forecasting using LSTM Technique

Sentiment Classification of Financial Texts for Stock Forecasting using LSTM Technique

Constructing and analyzing domain-specific language model for financial text mining

The application of natural language processing (NLP) to financial fields is advancing with an increase in the number of available financial documents. Transformer-based models such as Bidirectional Encoder Representations from Transformers (BERT) have been successful in NLP in recent years. These cutting-edge models have been adapted to the financial domain by applying financial corpora to existing pre-trained models and by pre-training with the financial corpora from scratch. In Japanese, by contrast, financial terminology cannot be applied from a general vocabulary without further processing. In this study, we construct language models suitable for the financial domain. Furthermore, we compare methods for adapting language models to the financial domain, such as pre-training methods and vocabulary adaptation. We confirm that the adaptation of a pre-training corpus and tokenizer vocabulary based on a corpus of financial text is effective in several downstream financial tasks. No significant difference is observed between pre-training with the financial corpus and continuous pre-training from the general language model with the financial corpus. We have released our source code and pre-trained models.

A Lexicon Enhanced Collaborative Network for targeted financial sentiment analysis

The increasing interest around emotions in online texts creates the demand for financial sentiment analysis. Previous studies mainly focus on coarse-grained document-/sentence-level sentiment analysis, which ignores different sentiment polarities of various targets (e.g., company entities) in a sentence. To fill the gap, from a fine-grained target-level perspective, we propose a novel Lexicon Enhanced Collaborative Network (LECN) for targeted sentiment analysis (TSA) in financial texts. In general, the model designs a unified and collaborative framework that can capture the associations of targets and sentiment cues to enhance the overall performance of TSA. Moreover, the model dynamically incorporates sentiment lexicons to guide the sentiment classification, which cultivates the model faculty of understanding financial expressions. In addition, the model introduces a message selective-passing mechanism to adaptively control the information flow between two tasks, thereby improving the collaborative effects. To verify the effectiveness of LECN, we conduct experiments on four financial datasets, including SemEVAL2017 Task5 subset1, SemEVAL2017 Task5 subset2, FiQA 2018 Task1, and Financial PhraseBank. Results show that LECN achieves improvements over the state-of-art baseline by 1.66 p.p., 1.47 p.p., 1.94 p.p., and 1.88 p.p. in terms of F1-score. A series of further analyses also indicate that LECN has a better capacity for comprehending domain-specific expressions and can achieve the mutually beneficial effect between tasks.

Predict financial text sentiment: an empirical examination

Purpose Stock market has always been lucrative for various investors. But, because of its speculative nature, it is difficult to predict the price movement. Investors have been using both fundamental and technical analysis to predict the prices. Fundamental analysis helps to study structured data of the company. Technical analysis helps to study price trends, and with the increasing and easy availability of unstructured data have made it important to study the market sentiment. Market sentiment has a major impact on the prices in short run. Hence, the purpose is to understand the market sentiment timely and effectively. Design/methodology/approach The research includes text mining and then creating various models for classification. The accuracy of these models is checked using confusion matrix. Findings Out of the six machine learning techniques used to create the classification model, kernel support vector machine gave the highest accuracy of 68%. This model can be now used to analyse the tweets, news and various other unstructured data to predict the price movement. Originality/value This study will help investors classify a news or a tweet into “positive”, “negative” or “neutral” quickly and determine the stock price trends.

Sentiment Classification of Financial Texts for Stock Markets using LSTM Technique: A Survey

Sentiment Classification of Financial Texts for Stock Markets using LSTM Technique: A Survey

Stock Market Prediction Based on Financial News Text Mining and Investor Sentiment Recognition

The stock market is usually regarded as the bellwether of the economy, which can reflect the economic operation of a country or region. As a significant part of the financial market, the equity market plays a critical role in the financial sector. Whether in academia or investment field, stock market forecasts always excite great interest. Financial news is an important source of information in the financial market, which reflects the mood swings of investors and often goes hand in hand with the market trend. However, due to the unstructured and professional characteristics of financial news, there are challenges in accurately quantifying their emotional tendencies. This research is based on Hidden Markov Model (HMM) to segment financial news text. The recognition and classification of news emotion is carried out by bidirectional long short-term memory (BI-LSTM) algorithm, and long short-term memory(LSTM) model is trained with text emotion index and stock market transaction data to realize the prediction of stock market. The results show that BI-LSTM algorithm performs better than the emotional dictionary algorithm in emotional recognition. And the emotional index of financial news text can enhance the accuracy of stock market prediction to a certain extent. Compared with using stock market technical index and news text vector only, the prediction accuracy can be improved by about 2%.

Efficacy of time-restricted eating and behavioural economic interventions in reducing fasting plasma glucose, HbA1c and cardiometabolic risk factors compared with time-restricted eating alone or usual care in patients with impaired fasting glucose: protocol for an open-label randomised controlled trial

IntroductionImpaired fasting glucose (IFG) is a significant risk factor for diabetes mellitus. Time-restricted eating (TRE) is one type of diet showing positive effects on metabolic signal pathways. However, effects of...

Public's Mental Health Monitoring via Sentimental Analysis of Financial Text Using Machine Learning Techniques.

Public feelings and reactions associated with finance are gaining significant importance as they help individuals, public health, financial and non-financial institutions, and the government understand mental health, the impact of policies, and counter-response. Every individual sentiment linked with a financial text can be categorized, whether it is a headline or the detailed content published in a newspaper. The Guardian newspaper is considered one of the most famous and the biggest websites for digital media on the internet. Moreover, it can be one of the vital platforms for tracking the public’s mental health and feelings via sentimental analysis of news headlines and detailed content related to finance. One of the key purposes of this study is the public’s mental health tracking via the sentimental analysis of financial text news primarily published on digital media to identify the overall mental health of the public and the impact of national or international financial policies. A dataset was collected using The Guardian application programming interface and processed using the support vector machine, AdaBoost, and single layer convolutional neural network. Among all identified techniques, the single layer convolutional neural network with a classification accuracy of 0.939 is considered the best during the training and testing phases as it produced efficient performance and effective results compared to other techniques, such as support vector machine and AdaBoost with associated classification accuracies 0.677 and 0.761, respectively. The findings of this research would also benefit public health, as well as financial and non-financial institutions.

Predicting NEPSE index price using deep learning models

Stock price prediction is a prevalent research field in both industry and academia. There is a pressing demand to develop a prediction model that captures the pattern of the financial activities with high precision to make an informed decision. Stock price prediction is challenging due to the complex, incomplete, fuzzy, nonlinear, and volatile nature of financial data. However, developing a robust model is possible due to advancements in artificial intelligence, availability of large-scale data, and increased access to computational capability. This study performs a comparative analysis of three deep learning models—the Long Short-term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN)—in predicting the next day’s closing price of the Nepal Stock Exchange (NEPSE) index. A set of sixteen predictors is carefully chosen under the domain of the fundamental market data, macroeconomic data, technical indicators, and financial text data of the stock market of Nepal. The performances of employed models are compared using the standard assessment metrics—Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Correlation Coefficient (R). The experimental results show that the LSTM model architecture provides a superior fit with high prediction accuracy. Moreover, statistical evidences are presented to validate the models’ reliability and robustness.

Using sentiment analysis to study the relationship between subjective expression in financial reports and company performance.

In recent years, with the development and progress of text information research, the disclosure of non-financial and qualitative information has often be found to have an incremental function. Financial reports, including financial statements and other relevant information, provide important insights on an enterprise’s financial status, operating results, and cash flow. Faced with a large number of financial reports, readers often do not know where to start, and as financial statements are prepared based on past transactions, they cannot fully reflect the past, present, and future economic conditions of the company. Information asymmetry and uncertainty make the text mining of financial reports of great significance to enterprise stakeholders. Accordingly, this paper takes financial reports as the research object and builds a research framework on the relationship between subjective expression in financial reports and company performance. Through natural language processing, sentiment analysis, and other text-mining technologies, the paper quantifies the subjective expression in financial reports and introduces intermediaries. Variables, moderating variables, and control variables are used to construct a multiple regression model. The empirical results show that the underlying emotional tendencies in subjective expressions substantially impact on the future development of listed companies. This paper enriches understanding of the multi-dimensional relationship between financial report text and company performance, and provides ideas for further exploring this relationship. It is of great practical significance to help them make rational decisions and ensure the normal operation of the company and the preservation and appreciation of capital.

Deep-Learning-Based Financial Message Sentiment Classification in Business Management.

A deep-learning-based financial text sentiment classification method is proposed in this paper, which can provide a reference for business management. In the proposed method, domain adaptation is adopted to solve the common problem of insufficient labeled samples in the financial textual domain. Specifically, in the classification process, the seq2seq model is firstly adopted to extract the abstract from the financial message, which can reduce the influence of invalid information and speed up processing. In the process of sentiment classification, a bidirectional LSTM model is adopted for classification, which can more comprehensively make use of context information. Experiments are carried out to testify the proposed method through the open-source data set. It can be seen that the proposed method can effectively transfer from the reduced Amazon data set to the StockTwits financial text data set. Compared with the parameter-frozen-based method and the SDA-based method, the recognition rates have improved by 0.5% and 6.8%, respectively. If the target domain data set can be directly adopted for training, the recognition rate of the proposed method is higher than that of the SVM method and the LSTM method by 8.3% and 4.5%, respectively.

Textual Emotional Tone and Financial Crisis Identification in Chinese Companies: A Multi-Source Data Analysis Based on Machine Learning

Nowadays, China is faced with increasing downward pressure on its economy, along with an expanding business risk on listed companies in China. Listed companies, as the solid foundation of the national economy, once they face a financial crisis, will experience hazards from multiple perspectives. Therefore, the construction of an effective financial crisis early warning model can help listed companies predict, control and resolve their risks. Based on textual data, this paper proposes a web crawler and textual analysis, to assess the sentiment and tone of financial news texts and that of the management discussion and analysis (MD&A) section in annual financial reports of listed companies. The emotional tones of the two texts are used as external and internal information sources for listed companies, respectively, to measure whether they can improve the prediction accuracy of a financial crisis early warning model based on traditional financial indicators. By comparing the early warning effects of thirteen machine learning models, this paper finds that financial news, as external texts, can provide more incremental information for prediction models. In contrast, the emotional tone of MD&A, which can be easily modified by the management, will distort predictions. Comparing the early warning effect of machine learning models with different input feature variables, this paper also finds that DBGT, AdaBoost, random forest and Bagging models maintain stable and accurate sample recognition ability. This paper quantifies financial news texts, unraveling implied information hiding behind the surface, to further improve the accuracy of the financial crisis early warning model. Thus, it provides a new research perspective for related research in the field of financial crisis warnings for listed companies.

NEGATIVE PREFIXES IN THE LANGUAGE OF FINANCE: CAN THEY BE THOUGHT-SHAPING?

Purpose. The article highlights the research aimed at assessing correlation between the frequency of negatively prefixed terms in financial discourse and the outlook of the speakers on their economy. The article opens with a historical reference, which briefly features the problems of Sapir-Whorf theory, its achievements and the stages of its development.
 Materials and methods. The authors turn to the description of the research procedure, which involved the study of financial texts, their statistic processing, work with respondents and analysis of their feedback. With the focus on the Russian language the study nonetheless employed English texts as its aims required a comparative study.
 Results. The discussion centres on the idea that the reasons of the pessimistic view of Russians on their economy may include a language factor, notably the abundant use of negatively prefixed terms.
 Practical implications. The authors are assured that the language conceals more phenomena nourishing Russian pessimism about the economy. The study has opened new language areas that can supposedly be viewed as pessimism shaping.

A Combined Approach for Multi-Label Text Data Classification.

Automated data analysis solutions are very dependent on data and its quality. The possibility of assigning more than one class to the same data item is one of the specificities that need to be taken into account. There are no solutions, dedicated to Lithuanian text data classification that helps to assign more than one class to data item. In this paper, a new combined approach has been proposed for multilabel text data classification for text analysis. The main aim of the proposed approach is to improve the accuracy of traditional classification algorithms by incorporating the results obtained using similarity measures. The experimental investigation has been performed using the financial news multilabel text data in the Lithuanian language. Data have been collected from four public websites and classified by experts into ten classes manually, where each of the data items has no more than two classes. The results of five commonly used algorithms have been compared for dataset classification: the support vector machine, multinomial naive Bayes, k-nearest neighbours, decision trees, linear and discriminant analysis. In addition, two similarity measures have been compared: the cosine distance and the dice coefficient. Research has shown that the best results have been obtained using the cosine similarity distance and the multinomial naive Bayes classifier. The proposed approach combines the results of these two methods. Research on different cases of the proposed approach indicated the peculiarities of its application. At the same time, the combined approach allowed us to obtain a statistically significant increase in global accuracy.

Using Financial News Sentiment for Stock Price Direction Prediction

Using sentiment information in the analysis of financial markets has attracted much attention. Natural language processing methods can be used to extract market sentiment information from texts such as news articles. The objective of this paper is to extract financial market sentiment information from news articles and use the estimated sentiment scores to predict the price direction of the stock market index Standard & Poor’s 500. To achieve the best possible performance in sentiment classification, state-of-the-art bidirectional encoder representations from transformers (BERT) models are used. The pretrained transformer networks are fine-tuned on a labeled financial text dataset and applied to news articles from known providers of financial news content to predict their sentiment scores. The generated sentiment scores for the titles of the given news articles, for the (text) content of said news articles, and for the combined title-content consideration are posited against past time series information of the stock market index. To forecast the price direction of the stock market index, the predicted sentiment scores are used in a simple strategy and as features for a random forest classifier. The results show that sentiment scores based on news content are particularly useful for stock price direction prediction.

FinEAS: Financial Embedding Analysis of Sentiment

In this article, the authors introduce a new language representation model for sentiment analysis of financial text called <i>financial embedding analysis of sentiment</i> (FinEAS). The new approach is based on transformer language models that are explicitly developed for sentence-level analysis. By building upon Sentence-BERT, a sentence-level extension of vanilla BERT, the authors argue that the new approach produces sentence embeddings of higher quality that significantly improve sentence/document-level tasks such as financial sentiment analysis. Using a large-scale financial news dataset from RavenPack, they demonstrate that for financial sentiment analysis the new model outperforms several state-of-the-art models such as BERT, a bidirectional LSTM, and FinBERT, a financial-domain-specific BERT. The authors make the model code publicly available.