Enhancing Interval Forecasting Accuracy of Iraqi Stock Market Prices based on v-support vector regression

Abstract

Stock price forecasting poses significant challenges due to non-stationarity, nonlinearity, and noise in financial markets, particularly for the Iraqi stock exchange. This study proposes an enhanced interval-valued forecasting model for daily prices of the national chemical and plastic industry (WSKB) company (2020–2025) using v-support vector regression (VSVR) with hyperparameters optimized via the coati optimization algorithm (COA). Interval time series are constructed from lower and upper price bounds, modeling center and radius components to capture uncertainty more effectively than point forecasts. The COA approach tunes key VSVR parameters through population-based exploration and exploitation phases inspired by coati hunting behaviors, outperforming grid search (GS-VSVR) and cross-validation (CV-VSVR). On training data (637 days), COA-VSVR achieves superior metrics for center (MAE=0.158, RMSE=0.253, DA=0.665, R²=0.965) and radius (MAE=0.169, RMSE=0.264, DA=0.642, R²=0.957) compared to baselines; testing results (308 days) confirm robustness. Further, Diebold-Mariano tests validate center-based COA-VSVR superiority over radius-based at 95% confidence (p<0.05). Visualizations and error reductions demonstrate the model's practical value for risk-aware investment in volatile emerging markets.