ANALYSIS OF MODELS OF WEB-BASED ANALYTICAL SYSTEMS FOR FORECASTING THE INTER-EXCHANGE VALUE OF DIGITAL CRYPTOASSETS

Abstract

The article presents a comprehensive analysis of existing models and methods for forecasting the inter-exchange value of digital cryptoassets. The evolution of forecasting approaches is studied – from simple algorithms based solely on historical data to complex multifactor models using artificial intelligence. Particular attention is paid to analysing the effectiveness of recurrent neural networks and long short-term memory (LSTM) models in forecasting cryptocurrency markets. Specific examples of the implementation of forecasting systems are considered, particularly the works of Ognjen Gatalo, Marco Santos, Derk Zomer and Frederic Riverall, and their advantages and limitations are analysed. An experiment on using an LSTM model for forecasting the bitcoin exchange rate is covered in detail, demonstrating the potential of neural networks in identifying hidden market patterns. An innovative approach based on the analysis of the news background and its correlation with the price dynamics of cryptocurrencies is also studied. The results of Daniel Chen’s research on multicriteria statistical analysis of the cryptocurrency market, including the study of the relationship between market capitalisation and various metrics of cryptocurrency popularity in social networks, are presented. The main problems and limitations of existing forecasting models are identified, particularly their low efficiency during periods of high volatility and the difficulty of considering external factors of influence. The necessity of an integrated approach to forecasting, combining the analysis of technical, fundamental and social factors, is substantiated. The article identifies promising areas for further research in the field of forecasting the value of cryptoassets, including improving methods for integrating heterogeneous data and developing algorithms that are more resistant to market fluctuations, as well as exploring the possibilities of using transformers and other modern neural network architectures to improve the accuracy of long-term forecasts.