Russian Federation
The article examines the architectural and methodological limitations of language models (LLM) in the field of financial analytics. The features of the architecture of LLM models are analyzed and their inconsistency with the properties of financial data is shown. Based on modern research, LLM and ensemble algorithms are compared at different sample sizes. It has been found that LLM models face problems with large and unbalanced amounts of data, unlike ensemble approaches.
LLM, Transformers, neural networks, ensemble algorithms, ROC-AUC, Accuracy, F-1 score, XGBoost, CatBoost, Random Forest, SMOTE
1. Hollmann N., Myuller S., Puruker L. i dr. Tochnye prognozy na osnove nebol'shih dannyh s ispol'zovaniem tablichnoy modeli // Nature 637. 2025. C. 319-326. DOI:https://doi.org/10.1038/s41586-024-08328-6.
2. Kostreva M., Furman O., Furman R., Tomchak S., Zieba M. Polezny li bazovye modeli dlya prognozirovaniya bankrostva? // 39-ya konferenciya po neyronnym sistemam obrabotki inofrmacii. NeurlPS 2025. Seminar: Generativnyy II v finansah. 2025. DOI:https://doi.org/10.48550/arXiv.2511.16375.
3. Sumin, V. I. Ispol'zovanie situacionnogo modelirovaniya v razrabotke sistem prinyatiya resheniy dlya slozhnyh organizacionnyh sistem / V. I. Sumin, A. S. Kravchenko, A. V. Tolkachev // Modelirovanie sistem i processov. – 2024. – T. 17, № 3. – S. 71-79. – DOIhttps://doi.org/10.12737/2219-0767-2024-69-77. – EDN FBBHJO.
4. Maktum T., Pulgram N., Chandgradkar V., Pathak P. Solanki A. Mashinnoe obuchenie dlya prognozirovaniya bankrotstva v sfere korporativnyh finansov s ispol'zovaniem ponyatnyy metodov II // Informatica 49. C. 15-26. 2025. DOI:https://doi.org/10.31449/inf.v49i15.6745.
5. Lokanan M.E., Ramzan S. Prognozirovanie finansovyh problem kompaniy, kotiruyuschihsya na birzhe TSX, s ispol'zovaniem algoritmov mashinnogo obucheniya // Front Artif Intell v. 7. 2024. DOI:https://doi.org/10.3389/frai.2024.1466321.
