DEHypGpOls

Paper Url: https://link.springer.com/article/10.1007/s00500-022-07571-1

This study introduces a hyperparameter optimal genetic programming-based algorithm, DEHypGpOls, for a-day-ahead prediction of stock market index trends. Utilizing a differential evolution (DE) algorithm, it optimizes hyperparameters of the genetic programming orthogonal least square (GpOls) algorithm, adapting it to the modeling dataset. The proposed approach enhances data-driven modeling performance and allows optimal autotuning of user-defined parameters. Tested on Istanbul Stock Exchange 100 (ISE100) and Borsa Istanbul 100 (BIST100) indexes, the DEHypGpOls algorithm achieves 57.87% average accuracy in buy-sell recommendations across four different time slots. Market investment simulations show a 4.8% higher average income compared to a long-term investment strategy when following the buy or sell signals of the DEHypGpOls forecast model for daily investments in ISE100 and BIST100 indexes.