From Market Noise to Signal: Machine Learning and Quantitative Alpha in Financial Markets

Abstract

The article examines the application of machine learning methods to the extraction of persistent alpha signals from classical technical indicators in the BTC/USD market under conditions of elevated volatility and shifting market regimes. The relevance of the study stems from the gap between the widespread use of indicators with historically established parameters and the requirements of the cryptocurrency market, which is characterized by a continuous trading cycle, a distinct microstructure, and a high level of noise. The aim of the work is to test whether a genetic algorithm can identify reproducible trading regularities that retain predictive power outside the training sample. The scientific novelty of the article is demonstrated by empirical substantiation of the thesis that classical indicators retain value under deep parametric recalibration and strict overfitting control through multistage back testing, sample splitting, and Monte Carlo stress tests. The principal findings show that alpha persists within a multidimensional parameter space. A portfolio of 11 strategies
demonstrated an annual return of 26% with a maximum drawdown of 15%, and most standard indicator settings required revision. The article will be useful for researchers in quantitative finance, algorithmic traders, trading system developers, and investors studying methods for generating persistent alpha in cryptocurrency markets.