Since the 1980s, economists have used Vector Autoregressive (VAR) models for financial time series forecasting, for their simplicity and causal interpretability.This thesis investigates the performance of VAR models enhanced with ridge and lasso regularization methods for predicting cryptocurrency prices, focusing on high-frequency data. The study compares traditional VAR models with their regularized counterparts, VARX-Lasso and VARX-Ridge, in forecasting returns in volatile and speculative cryptocurrency markets. The empirical analysis is conducted using Bitcoin and a selection of other cryptocurrencies, exploring the impact of these regularization techniques on prediction accuracy. This research contributes to the limited literature on high-frequency cryptocurrency forecasting, offering insights into the dynamic relationships between crypto assets.