Modelling Prices of Cryptocurrencies on Centralized Exchanges Using Econometric Methods and Machine Learning

In recent years, cryptocurrencies have attracted the attention of investors, traders and economists around the world due to the uniqueness and rapid expansion of this economic sphere, and the potential for future technological development of the electronic finance market. Due to the lack of a control mechanism, less reliability and security of the decentralized part of the cryptocurrency market, more than 80% of the trading volume of cryptocurrencies falls on centralized exchanges. Quite a large number of works in the world focus on the analysis of one specific token, completely ignoring the study of several tokens at the same time or focusing the analysis on a specific method of econometrics or machine learning. This study attempts to correct a number of inaccuracies in these studies and uses a broader range of statistical methods for analysis, and does not focus on a single coin. As a sample, data was taken from the Binance website with the following metrics: date, opening price, closing price, highest price for the period, trading volume for the period from 10/16/2020 to 03/13/2024. The closing price is used as the main indicator. The interval of datasets received from the exchange website is 1 day, 1 hour, 4 hours, 5, 15 and 30 minutes, but the range of one day is mainly used. Based on these data, a number of models were built; MSE, MAE, RMSE, and MAPE were used as the final metrics for comparison.