Evaluating the Volatility Forecasting Performance in the Sukuk Market

Abstract

This study investigates the predictability of sukuk market volatility and evaluates the forecasting performance of six models(ARIMA, GARCH(1,1), HAR, PROPHET, ANN, and LSTM) across short- , medium-, and long-term horizons using a panel of 31 sukuk instruments based on realized volatility. The dataset covers 1,086 daily observations for each instrument over the period from January 2019 to April 2023, including sukuk funds issued across diverse regions such as the GCC, Southeast Asia, and selected global markets. The first stage of analysis confirms that sukuk returns exhibit volatility clustering, stationarity, non-normality, and time-varying conditional variance, indicating strong potential for accurate forecasting. The second stage assesses in-sample forecast performance through a two-step evaluation process: visual comparison of predicted versus actual volatility and quantitative measurement using MAE, MSE, and RMSE. Results indicate that forecast performance varies depending on the evaluation method used, showing that model rankings are not consistent across criteria. Based on quantitative error metrics, HAR is ranked first across all forecasting horizons, confirming its strong numerical accuracy. In the short term, ARIMA and Prophet are ranked second, while LSTM and ANN show moderate accuracy. In the medium and long term, LSTM is ranked second, followed by Prophet, and then ANN. However, from a visual evaluation perspective, the ranking changes. LSTM is ranked first in the short and medium terms, while ANN is ranked first in the long term, as both models track the realized volatility more closely. In contrast, GARCH performs worst across all horizons under both evaluation methods, producing forecasts that fail to capture market dynamics. Overall, these findings demonstrate that sukuk volatility exhibits structured and predictable behaviour, which directly supports your dissertation’s central research question on market predictability. At the same time, the results show that the optimal model depends on the evaluation criterion: HAR is the most reliable model for numerical accuracy, LSTM and ANN are the most effective models for visual tracking, whereas GARCH is consistently the weakest performer and should be avoided. These findings provide practical guidance for financial institutions, regulators, and investors in managing volatility and selecting forecasting tools in Islamic finance.