A Comprehensive Review of Stock Market Forecasting Methods: From Statistical Models to Deep Learning

Rutgers University, School of Arts and Sciences, NJ 08854 New Brunswick, USA

^* Corresponding author: tc1146@scarletmail.rutgers.edu

Abstract

Stock forecasting is a key area of financial research, attracting widespread attention due to its impact on capital allocation, risk management, and investment decisions. This paper systematically reviews the main methods for stock forecasting in recent years, covering traditional statistical models, machine learning methods, deep learning models, and hybrid and extended methods. This paper first introduces the advantages and limitations of statistical models in time series modeling; then analyzes the contributions of machine learning in handling nonlinear relationships and feature selection, as well as deep learning’s benefits for identifying long- term dependency and automatic feature extraction. Furthermore, this paper explores the potential of hybrid models and multi-source data fusion to improve forecast accuracy and robustness. Finally, this paper discusses the challenges faced by existing methods in terms of data quality, model evaluation, computational resources, and interpretability. By comparing the applicable scenarios and performance of different methods, this paper aims to provide a reference for subsequent research and practical applications.