Evaluating the Directional-Weighted Mean Absolute Error in Long Short-Term Memory Models for Stock Price Prediction

Shuaiting Li

2023

Abstract

In the intricate landscape of financial forecasting, accurate prediction of stock prices remains a pivotal challenge, demanding continual innovation in modeling techniques. This paper introduces the Directional-Weighted Mean Absolute Error (D-MAE) as a potential loss function to refine the predictive capabilities of Long Short-Term Memory (LSTM) models. Leveraging a comprehensive dataset of leading technology firms, namely Apple Inc., Alphabet Inc., Microsoft Corporation, and Amazon.com, Inc., spanning from January 1, 2012, to September 1, 2023, the research contrasts the performance of D-MAE against conventional loss functions. D-MAE’s uniqueness stems from its ability to weigh prediction errors differentially based on the accuracy of directional stock price movements, striving for an equilibrium between numerical prediction precision and the discernment of price trends. Preliminary assessments, utilizing metrics such as accuracy, precision, recall, and F1-score, offer insights into D-MAE’s potential benefits in the realm of stock price forecasting. This exploration underlines the evolving nature of financial analytics and the pressing need to integrate innovative methodologies that can capture the nuanced dynamics of global stock markets.

Paper Citation

in Harvard Style

Li S. (2023). Evaluating the Directional-Weighted Mean Absolute Error in Long Short-Term Memory Models for Stock Price Prediction. In Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-705-4, SciTePress, pages 171-177. DOI: 10.5220/0012814700003885

in Bibtex Style

@conference{daml23,
author={Shuaiting Li},
title={Evaluating the Directional-Weighted Mean Absolute Error in Long Short-Term Memory Models for Stock Price Prediction},
booktitle={Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2023},
pages={171-177},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012814700003885},
isbn={978-989-758-705-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - Evaluating the Directional-Weighted Mean Absolute Error in Long Short-Term Memory Models for Stock Price Prediction
SN - 978-989-758-705-4
AU - Li S.
PY - 2023
SP - 171
EP - 177
DO - 10.5220/0012814700003885
PB - SciTePress