Machine Learning for Cryptocurrency Momentum Trading

Summary

The document describes a daily classification approach for detecting positive momentum, negative momentum, or normal conditions in Bitcoin, Ethereum, and Litecoin. It uses historical prices and technical indicators as features, then compares classifiers including support vector machines, naive Bayes, K-nearest neighbors, logistic regression, and random forests to predict the next day’s momentum state and direction. The proposed strategy uses those predictions to guide trades, aiming to adapt momentum signals to volatile crypto markets rather than rely only on fixed heuristics.

The account reports that machine learning reduced false signals and improved simulated trading outcomes relative to traditional rules, with K-nearest neighbors performing best on average and especially well for Bitcoin and Ethereum. It cites a Bitcoin average trade return comparison as an example. However, the supplied text gives no dataset dates, validation design, transaction costs, or detailed risk-adjusted results, so the reported backtest performance cannot establish live profitability or generalization.

Key ideas

• The method labels each day as positive momentum, negative momentum, or normal.
• It uses technical indicators from historical prices to predict the following day’s momentum state and direction.
• Several classification models are compared, and K-nearest neighbors is reported as the strongest on average.
• The reported backtests favor machine learning over heuristic rules, but the provided description omits important validation and trading-cost details.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.