The financial market, characterized by non-linearity and non-stationarity, poses a significant challenge for traditional forecasting models. This study proposes a Hybrid Genetic Algorithm-Echo State Network (GA-ESN) Model to overcome the limitations o...
The financial market, characterized by non-linearity and non-stationarity, poses a significant challenge for traditional forecasting models. This study proposes a Hybrid Genetic Algorithm-Echo State Network (GA-ESN) Model to overcome the limitations of conventional Technical Analysis (TA) and enhance the accuracy of stock price prediction, specifically focusing on market turning points. The methodology integrates three core elements: Technical Analysis (GC, ENV, RSI), Genetic Algorithms (GA), and the Echo State Network (ESN). GA is utilized as an intelligent optimization tool to automatically determine the optimal parameters for the TA indicators and ESN Hyper-parameters, compensating for their inherent weaknesses such as noise and parameter sensitivity. The optimized TA signals are then fed as input features into the ESN, a type of Recurrent Neural Network based on Reservoir Computing, known for its efficiency and ability to learn complex non-linear temporal patterns in financial time series. A crucial aspect of this research is the Market Regime Classification based on Stability and Volatility. The market is segmented into four distinct regimes—Stable–Non-Volatile, Stable–Volatile, Trend–Non-Volatile, and Trend–Volatile—to empirically verify the model’s performance and adaptability. The empirical analysis is performed on major global index and their tracking ETFs (DIA, QQQ, SPY, and KOSPI200) using daily closing price data from January 2000 to September 2025. Furthermore, the study compares the performance of a Static ESN with a Cross-Validation ESN (CV-ESN) to assess the effectiveness of an adaptive learning structure. The empirical results demonstrate that the proposed hybrid CV-ESN approach significantly improves market adaptability and stability of returns compared to B&H(Buy and Hold) strategy and the Static ESN. Specifically, This study successfully validates that integrating GA-optimized TA signals with a market-adaptive ESN learning structure can enhance prediction accuracy and risk management efficiency, providing a robust Intelligent Trading Framework for stable profit generation even in volatile market regimes.
Keyword : Algorithmic Trading, Stock Price Prediction, Hybrid Model, Genetic Algorithm (GA), Echo State Network (ESN), Market Regime Classification.