Forecasting regular and extreme gold price volatility: The roles of asymmetry, extreme event, and jump

Abstract

Gold as a vital hedging asset plays increasing critical roles in risk management during turmoil macroeconomic environments. For the massive and indistinct impactors of gold price volatility, this paper tries to investigate whether the short‐ and long‐term asymmetry, extreme observations, and jump components in past gold volatility help to obtain higher forecasting accuracy in future volatility from both in‐sample and out‐of‐sample perspectives. A variety of evaluation methods are utilized to compare the performances of GARCH‐MIDAS models incorporating these volatility components and the standard ones without them. The results of in‐sample estimation show first that all the short‐term and long‐term asymmetry, extreme observations, and jump components have significantly impact on gold volatility. The evaluation results of out‐of‐sample forecasts suggest that the forecasting accuracy of gold volatility can be significantly improved by most of the extended GARCH‐MIDAS models including asymmetry, extreme observations, and jump components. The model including short‐term jump intensity and the model with both long‐term asymmetry and long‐term leverage effects have better forecasting performances than other models for gold volatility, especially for regular volatility. Moreover, GARCH‐MIDAS models incorporating long‐term leverage and long‐term jump have better performances in forecasting accuracy of extreme gold volatility.

Gold Prices Time-Series Forecasting: Comparison of Statistical Techniques

“…The heart of this methodology lies in the training of neural networks, where the algorithms are fine-tuned to recognize intricate patterns within the historical data and grasp the nuanced interplay between various global economic factors. This process forms a critical foundation for the subsequent predictive analyses [10].…”

“…This data is processed and cleaned to remove noise. Subsequently, the data is used to train neural networks that will be utilized in deep learning algorithms to predict future gold prices [10]. These predictions are based on patterns identified in historical data and global economic factors used during the neural network training.…”