<![CDATA[Accurate forecasting of financial time series remains challenging due to non-stationarity, complex data patterns, and difficulties in parameter optimization within traditional models. Although ARIMA is widely used, its performance is often limited by static parameter estimation and sensitivity to evolving data structures. Existing metaheuristic-based approaches have attempted to address these issues; however, many lack adaptive mechanisms that account for varying data complexity. This study proposes a Continuous Hybrid ARIMA–Metaheuristic (GA–PSO) framework with adaptive parameter tuning guided by Model Complexity Assessment (MCA). The framework enables continuous optimization of ARIMA parameters, allowing the model to dynamically adapt to changing time-series characteristics. Empirical results demonstrate consistent improvements in forecasting performance compared to the baseline ARIMA model. For instance, in the Gold dataset (300 observations), the model achieved RMSE = 56.96, MAE = 41.86, and MAPE = 1.12%, indicating stable and accurate predictions. Statistical validation using the Diebold–Mariano test further confirms the significance of these improvements. The main contribution lies in the integration of adaptive GA–PSO optimization with complexity-aware tuning, which enhances both forecasting stability and responsiveness. However, the findings also indicate the presence of heteroscedasticity in several cases, suggesting that volatility dynamics are not fully captured by the current framework. This limitation highlights the need for incorporating volatility-aware models, such as ARIMA–GARCH, to better represent time-varying variance and improve forecasting robustness in future research.]]>
