시장 상태 적응형 Hybrid GA-ESN 기반 주가 예측 알고리즘 트레이딩 시스템에 관한 실증연구: DIA, QQQ, SPY, KOSPI200 지수 중심으로 = An Empirical Study on a Market Regime-Adaptive Hybrid GA-ESN based Algorithmic Trading System for Stock Price Prediction: Evidence from DIA, QQQ, SPY, and KOSPI200

다국어 초록 (Multilingual Abstract)

Traditional technical analysis methods rely on empirically chosen parameters, which limits their effectiveness in improving prediction accuracy and trading stability for nonlinear and non-stationary financial time series. This study proposes a market regime-adaptive Hybrid GA–ESN (Genetic Algorithm–Echo State Network)-based algorithmic trading system for stock price prediction. A genetic algorithm is employed to automatically optimize key parameters of the Golden/Dead Cross (GC), Envelope Moving Average (ENV), and Relative Strength Index (RSI), mitigating parameter-sensitivity issues. Smoothed prices and a triangle target representation for turning points are then used as input features to the ESN. In addition, the market is classified into four regimes according to trend and volatility, and a cross-validation-based ESN is adopted to enhance adaptability to changing market conditions. Empirical results for the DIA, QQQ, SPY, and KOSPI200 indices demonstrate that the proposed Hybrid GA–ESN model outperforms the Buy-and-Hold strategy and static ESN models in terms of both return and maximum drawdown.