The stochastic, non-linear, and dynamic nature of financial markets significantly diminishes the effectiveness of traditional trading stratsgies relying on fixed parameters over extended periods. While the Efficient Market Hypothesis (EMH) suggests that asset prices reflect all available information, rendering systematic profit generation impossible, the field of algorithmic trading operates on the premise that temporary market inefficiencies and behavioral anomalies can be exploited. This study presents a comprehensive Genetic Algorithm (GA) framework designed to develop and optimize an adaptive trading strategy for multi-asset portfolios consisting of high-liquidity technology stocks (Apple, Microsoft, Google). Unlike traditional optimization methods that focus solely on parameter tuning for a single indicator, the proposed system introduces a novel "genetic switch" mechanism. This mechanism allows the algorithm to simultaneously optimize the structural components of the strategy determining which combination of indicators (EMA, MACD, RSI, Momentum) yields the best performance and their respective parameters. The model’s fitness function prioritizes risk-adjusted returns by utilizing a Calmar-like ratio, explicitly penalizing excessive drawdowns. To ensure robustness and mitigate the prevalent risk of overfitting (data snooping bias), a rigorous Walk-Forward Optimization (WFO) technique was applied to daily data spanning the 2020-2024 period. The findings demonstrate that the proposed GA framework generates a robust trading system that statistically outperforms the passive "buy-and-hold" strategy, achieving a higher Sortino Ratio (1.98 vs 1.21) and significantly lower maximum drawdown (-18.5% vs -35.1%). The outperformance over the buy-and-hold benchmark is statistically validated across all walk-forward windows, indicating robustness rather than data snooping effects.
Kör et al. (2026) studied this question.