20.500.12556/RUL-177390 Ovrednotenje temeljnih modelov za napovedovanje kratkoročnih gibanj na delniških trgih Evaluation of foundation models for short term stock-market forecasting Napovedovanje gibanja cen delnic je kompleksen problem zaradi nestacionarnosti in visoke stopnje šuma v finančnih časovnih vrstah. To diplomsko delo sistematično ovrednoti učinkovitost sodobnih temeljnih modelov, ki temeljijo na arhitekturi Transformer, za napovedovanje kratkoročnih cenovnih gibanj. V dvofaznem eksperimentu smo primerjali modela Chronos in TimesFM na znotrajdnevnih podatkih enajstih volatilnih delnic pri 5-minutnih, 15-minutnih in 1-urnih intervalih. Na podlagi rezultatov je bil za podrobnejšo analizo na 5-minutnem intervalu izbran model Chronos, ki smo ga ovrednotili v dveh načinih: brez finega uravnavanja (zero-shot) in s finim uravnavanjem (fine-tuned), ter ga primerjali z modeloma ARIMA in naključnim sprehodom. Rezultati kažejo, da se fino uravnani model Chronos izkaže kot najstabilnejši, saj ima najožji razpon napak, čeprav po mediani napake ne presega vedno močnega izhodiščnega modela naključnega sprehoda. Kljub temu kvalitativna analiza pokaže, da je model izjemno uspešen pri zajemanju jasnih trendov v obdobjih nižje volatilnosti, kjer dosega visoko smerno natančnost, medtem ko, pričakovano, odpove pri napovedovanju ekstremnih, nepredvidljivih dogodkov. Delo potrjuje potencial arhitekture Transformer, ki ponuja bolj konsistentne napovedi od klasičnih pristopov, in nakazuje na obstoj predvidljivih vzorcev v določenih tržnih pogojih. Forecasting stock price movements is a complex problem due to the non-stationarity and high level of noise in financial time series. This thesis systematically evaluates the effectiveness of modern foundation models based on the Transformer architecture for predicting short-term price movements. In a two-phase experiment, the Chronos and TimesFM models were compared on intraday data from eleven volatile stocks at 5-minute, 15-minute, and 1-hour intervals. Based on the results, the Chronos model was selected for a more detailed analysis at a 5-minute interval, evaluated in both zero-shot and fine-tuned modes, and compared against the ARIMA and random walk baseline models. The results show that the fine-tuned Chronos model proves to be the most stable, exhibiting the narrowest error distribution, although its median error does not always surpass that of the strong random walk baseline. Nevertheless, qualitative analysis demonstrates that the model is exceptionally successful at capturing clear trends during periods of lower volatility, where it achieves high directional accuracy, while, as expected, failing to predict extreme, unpredictable events. This work confirms the potential of the Transformer architecture, which offers more consistent predictions than classical approaches, and indicates the existence of predictable patterns under certain market conditions. napovedovanje časovnih vrst finančni trgi delnice arhitektura Transformer temeljni modeli Chronos time series forecasting financial markets stocks Transformer architecture foundation models Chronos true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2025-12-22 14:05:07 2025-12-22 14:05:17 2026-02-04 14:56:43 0000-00-00 00:00:00 2025 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 62925 267340035 Valand_Michael_-_Ovrednotenje_temeljnih_modelov_za_napovedovanje_kratkorocnih_gibanj_na_delniski.pdf Valand_Michael_-_Ovrednotenje_temeljnih_modelov_za_napovedovanje_kratkorocnih_gibanj_na_delniski.pdf 1 5F37D5AEFAE793A8968978D152FDBA1A 2ad648422f8691110dc85548348f00db1495a95d340cc92ecdbfd79425f330df 69e51f85-df36-11f0-9328-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=224418 Fakulteta za elektrotehniko 0 0 0