Purpose - This study aims to investigate the performance of traditional, modern, and emerging forecasting models, focusing on their effectiveness in predicting stock indices and prices in the Vietnamese stock market. Addressing a critical gap in the literature, the research explores the suitability of models such as Autoregressive Integrated Moving Average (ARIMA), Kolmogorov-Arnold Networks (KAN), and ensemble methods in an emerging market context characterized by high volatility and non-linear dynamics.

Design/Methodology/Approach - The study employs a quantitative approach, utiliz- ing historical stock data from the Vietnamese stock market, including VN-Index and individual stocks. Data were analyzed across different forecast horizons (h = 1, 3, 5, 20) using models like ARIMA, Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), KAN, hybrid models between ARIMA and neural networks and ensemble methods.

Findings - The results reveal that ARIMA excels in short-term forecasting (h = 1), while the ensemble methods outperform on longer horizons (h = 3, 5, 20), effectively integrating strengths between models. KAN, despite its basic implementation, shows significant promise as a robust tool for non-linear dynamics, while MLP consistently outperformed LSTM, highlight- ing its resilience to noise. These findings confirm the complementary strengths of traditional and emerging methods, offering new insight into the challenges and opportunities of stock market forecasting in emerging markets.

Originality/Value - This research offers a novel perspective by assessing the performance of KAN, a relatively new model, in the context of an emerging market. It provides valuable insights into the effectiveness of hybrid and ensemble approaches, contributing to the devel- opment of more adaptive forecasting models. The findings highlight practical implications for traders, investors, and policymakers in emerging markets, while paving the way for future research into refining KAN and tailoring hybrid models for complex market dynamics.