Artificial Intelligence Models and Cryptocurrency Price Volatility: A Predictive Econometric Approach to Bitcoin and Ethereum

Abstract

This study aims to analyze the determinants of digital asset pricing by assessing the explanatory and predictive roles of generative artificial intelligence tools and financial analytics platforms relative to macroeconomic variables and traditional financial market dynamics. Particular emphasis is placed on examining whether Large Language Models (LLMs) possess an independent structural capacity to influence cryptocurrency market volatility. The study focuses on a sample comprising Bitcoin and Ethereum over the period from March 10, 2016, to March 14, 2026. Methodologically, it employs the Random Forest algorithm, a machine learning technique, to extract the relative importance of explanatory variables using the Percentage Increase in Mean Squared Error (%IncMSE) metric. The findings reveal that cryptocurrencies have become increasingly integrated into the global financial system, with financial stress indicators, energy markets, and energy costs emerging as the dominant drivers of price fluctuations. The results also indicate a limited structural impact of generative AI, suggesting that its role is primarily interpretive and reactive rather than a direct catalyst of market shocks. In contrast, on-chain analytics platforms demonstrate stronger explanatory power, as smart liquidity tracking through Nansen plays a more prominent role in explaining Ethereum dynamics, while aggregate supply indicators from Glassnode exert greater influence on Bitcoin as a monetary asset. Furthermore, the results show that Ethereum exhibits higher informational efficiency and a shorter price memory than Bitcoin, reflecting a faster process of shock absorption and price adjustment.