Abstract <p>Accurate forecasting of gold prices is essential for financial decision-making, given their volatility and their role in global markets. This study develops CEEMDAN-based hybrid forecasting frameworks for world gold prices by integrating ARIMA and ANN models. Building on existing CEEMDAN-based hybrid approaches, the proposed framework differs in its reconstruction strategy and IMF classification, where intrinsic mode functions are systematically separated into stochastic and deterministic components using an autocorrelation-based criterion, and stochastic components are modeled individually rather than combined, allowing more targeted modeling of individual dynamics. The proposed hybrid model works in four steps: First, a set of IMFs and a trend component are extracted from the given gold price data. In the second step, the deterministic and stochastic components are obtained from the IMFs and the trend function based on autocorrelation and average mutual information (AMI). Third, ARIMA and ANN techniques are employed to predict each IMF and the reconstructed components, which are stochastic and deterministic. In the last step, all results are combined to produce the final output. The model’s performance is evaluated using MAE, RMSE, MAPE, MSE, and R², while comparative forecast accuracy is further validated using the Diebold–Mariano (DM) statistical test. The empirical findings show that the CEEMDAN-ISD-ARIMA framework consistently outperforms the benchmark models, indicating its effectiveness for gold price forecasting and policy-oriented decision support.</p> Graphical abstract <p></p>