Method evolution and comparison in asset price prediction: a review from traditional statistical models to deep learning

Abstract

Asset price prediction is one of the most challenging research topics in finance. The inherent nonlinearity, low signal-to-noise ratio, and non-stationarity of Financial Time Series (FTS) make it difficult for traditional time series models to perform adequately. This paper reviews the evolution from traditional statistical models to Machine Learning (ML) and Deep Learning (DL), with a focus on comparing their performance in predictive accuracy, interpretability and robustness. Traditional models offer good interpretability and computational efficiency but are constrained by linear assumptions. ML models do not significantly outperform traditional methods, but exhibit conditional advantages that depend on asset characteristics and market conditions. DL models can capture complex temporal dependencies, but face challenges such as poor interpretability, overfitting and deployment difficulties. Based on current research, this review concludes that no universally optimal model exists, but only relatively suitable models exist for specific tasks, asset characteristics, and market environments. The future direction lies not in more complex architectures, but in reshaping prediction tools into interpretable decision support systems, which requires simultaneous advances in model architecture innovation, interpretability methods, and standardized evaluation frameworks.