Investigating the Relationship Between Romanian Financial News and Closing Prices from the Bucharest Stock Exchange

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5130–5136 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5130 Investigating the Relationship Between Romanian Financial News and Closing Prices from the Bucharest Stock Exchange Ioan-Bogdan Iordache1,2, Ana Sabina Uban1,2, Catalin Stoean1,3, Liviu P. Dinu1,2 1Human Language Technology Research Center, University of Bucharest, Bucharest, Romania, 2Faculty of Mathematics and Computer Science, University of Bucharest, Bucharest, Romania, 3Faculty of Sciences, University of Craiova, Craiova, Romania, iordache.bogdan1998@gmail.com, auban@fmi.unibuc.ro, catalin.stoean@inf.ucv.ro, ldinu@fmi.unibuc.ro Abstract A new data set is gathered from a Romanian financial news website for the duration of four years. It is further refined to extract only information related to one company by selecting only paragraphs and even sentences that referred to it. The relation between the extracted sentiment scores of the texts and the stock prices from the corresponding dates is investigated using various approaches like the lexicon-based Vader tool, Financial BERT, as well as Transformer-based models. Automated translation is used, since some models could be only applied for texts in English. It is encouraging that all models, be that they are applied to Romanian or English texts, indicate a correlation between the sentiment scores and the increase or decrease of the stock closing prices. Keywords: sentiment analysis, low resource languages, transformers, stock exchange 1. Introduction The prediction of the stock prices and, more recently, the one of the cryptocurrencies, represents an important challenge for economists and machine learning scien- tists. The efforts are generally focused on designing models that identify specific patterns that determine the changes in the stock price market and accordingly pre- dict the value for the next hour, day, month or even year. The prediction is performed for each company separately and, at a time t, it is often based on the price stock from the previous N days, as well as for the num- ber of transactions that were achieved, the amount of shares, the minimum and maximum prices for them in that period, the opening price and so on (Rouf et al., 2021; Gandhmal and Kumar, 2019; Shah et al., 2019). Relying the prediction only on previous information about the transactions of stocks is clearly not enough. A model that seems to perform well for the stocks of a certain company for predicting its closing price for a period of time does not lead to similarly good out- puts on a different time frame. Additionally, a model that appears to be suitable for one company lead to modest results for a different one. In conclusion, de- signing a model that is based only on numerical infor- mation about transactions resembles to a Sisyphean ef- fort. In the current work, we focus on only one com- pany and we analyse the information that appears in the press regarding that company for verifying the sen- timents extracted from the texts and the closing prices of the company. Plus, we use texts in Romanian, fact that further complicates the task, since …

Content truncated to 500 words