Hybrid financial modelling approaches: Integrating classical stochastic models with machine learning

The aim of this study was to assess the potential of large language models in improving the accuracy of financial forecasts and market stability in transition economies, in particular in Azerbaijan. The methodology identified the essence of hybrid approaches to financial modelling that combine classical econometric tools, such as AutoRegressive Integrated Moving Average and Vector Autoregression, with deep learning algorithms. As a result, it was found that the trading volume on the Baku Stock Exchange increased from AZN 37 to AZN 71 billion in 2024, the gross domestic product increased from AZN 72.4 to AZN 126.3 billion, and the inflation rate decreased from 13.9% in 2022 to 2.2% in 2024. The share of non-cash transactions exceeded 90%, and the time for interbank settlements was reduced to 5-20 seconds due to the introduction of the Azerbaijan Interbank Payment System. The information and communication technologies development index increased by 7%, and the Open Data Index increased from 56 to 61 points, creating a foundation for effective digital analytics. At the same time, the study showed that the effectiveness of large language models was limited by low transparency of financial reporting, the presence of only 26 companies listed on the stock exchange, and a Corruption Perceptions Index score of 22-23. The practical significance of this study is that its results can be used by financial analysts, government agencies, and banking regulators in Azerbaijan to develop strategies for digital transformation of the financial sector and implement analytical systems based on large language models