From Price to Returns: The Performance Evolution of Machine Learning Models in Stock Prediction Tasks

Zirui Zeng

doi:10.54097/prgt6724

Authors

Zirui Zeng School of Management, Shenzhen University, Shenzhen, China

DOI:

https://doi.org/10.54097/prgt6724

Keywords:

Stock Price Prediction, Machine Learning, Time Series Forecasting, Data Stationarity.

Abstract

More and more machine learning models have been used in financial forecasting. However, its application in real life has encountered many difficulties, mainly due to the non-stationary nature of the stock price data. This paper systematically examines the important effect that problem formulation has on predictive quality by looking at two different modeling approaches. The first paradigm makes direct end-to-end prediction on raw non-stationary price sequences. On the other hand, the second is theoretically informed, and focused on forecasting stationary log returns of carefully engineered features. A suite of models (LSTM, Random Forest, Ridge) were compared to a Naive Forecast in both frameworks. Empirical tests resulted in very poor performance from models of the first paradigm: the model performed abysmally and the predictions were a lot worse than the baselines. On the other hand, they applied to stationary returns are extremely predictive. The paper’s main idea is that making data stationary is more than something people do to improve the model, it is essential to obtain any form of significant financial forecast. The selection of a well founded methodology over the selection of a particular complex algorithm is much more significant.

Downloads

Download data is not yet available.

References

[1] Fischer T, & Krauss C. Deep learning with long short-term memory networks for financial market predictions. European journal of operational research, 270(2), 654-669. 2018.

[2] Gu S, Kelly B, & Xiu, D. Empirical asset pricing via machine learning. The Review of Financial Studies, 33(5), 2223-2273. 2020.

[3] Sezer O B, Gudelek M U, & Ozbayoglu A M. Financial time series forecasting with deep learning: A systematic literature review: 2005–2019. Applied soft computing, 90, 106181. 2020.

[4] Itri B, Mohamed Y, Omar B, Latifa E M, Lahcen M, & Adil O. Hybrid machine learning for stock price prediction in the Moroccan banking sector. International Journal of Electrical & Computer Engineering (2088-8708), 14(3). 2024.

[5] Cao H, Lin T, Li Y, & Zhang H. Stock price pattern prediction based on complex network and machine learning. Complexity, 2019(1), 4132485.

[6] Behera R K, Das S, Rath S K, Misra S & Damasevicius R. Comparative Study of Real Time Machine Learning Models for Stock Prediction through Streaming Data. J. Univers. Comput. Sci., 26(9), 1128-1147. 2020.

[7] Vuong P H, Dat T T, Mai T K, & Uyen P H. Stock-price forecasting based on XGBoost and LSTM. Computer Systems Science & Engineering, 40(1). 2022.

[8] Kamalov F, Gurrib I, & Rajab K. Financial Forecasting with Machine Learning: Price Vs Return. Journal of Computer Science, 17(3), 251-264. 2021.

[9] Guennioui O, Chiadmi D, & Amghar M. Machine learning-driven stock price prediction for enhanced investment strategy. International Journal of Electrical & Computer Engineering (2088-8708), 14(5). 2024.

[10] Zhu Z, Chen H, Qu Q, & Chung V. FinCast: A Foundation Model for Financial Time-Series Forecasting. arXiv preprint arXiv:2508.19609. 2025.

[11] Chaudhary T, & Minirani S. Machine Learning and Time Series Approaches for Stock Price Prediction: A Case Study of Indian Banks. Procedia Computer Science, 260, 626-634. 2025.

[12] Li C, Gan Z, Yang Z, Yang J, Li L, Wang L, & Gao J. Multimodal foundation models: From specialists to general-purpose assistants. Foundations and Trends® in Computer Graphics and Vision, 16(1-2), 1-214. 2024.