AbstractThis study employs four (4) Generalized Autoregressive Conditional Heteroscedasticity (GARCH) variants namely GARCH (1, 1), Glosten–Jagannathan–Runkle (GJR), Auto Regressive Integrated Moving Average (ARIMA)-GARCH and ARIMA-GJR as benchmark models to assess the performance of a proposed novel Gene Expression Programming (GEP) based univariate time series modeling approach used to conduct ex ante oil price volatility forecasts. The report illustrates that the GEP model is more superior to any of the traditional models on issues relating to both loss functions applied. The GEP model is of a greater volatility forecasting precision at different forecast horizons, therefore. There is also the existence of evidence that GJR and ARIMA-GJR differ in their loss functions, the performance is nevertheless better than GARCH (1, 1) and ARIMA-GARCH. This study conducted herein achieves importance in literature by broadening the application of gene algorithms in finance and forecasting. It also solves the problem of high error associated with the use of GARCH related models in oil price volatility forecasting.
Optimizing Neural Architecture Search using Limited GPU Time in a Dynamic Search Space: A Gene Expression Programming Approach
