Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
173 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

An Evaluation of Deep Learning Models for Stock Market Trend Prediction (2408.12408v1)

Published 22 Aug 2024 in cs.LG

Abstract: The stock market is a fundamental component of financial systems, reflecting economic health, providing investment opportunities, and influencing global dynamics. Accurate stock market predictions can lead to significant gains and promote better investment decisions. However, predicting stock market trends is challenging due to their non-linear and stochastic nature. This study investigates the efficacy of advanced deep learning models for short-term trend forecasting using daily and hourly closing prices from the S&P 500 index and the Brazilian ETF EWZ. The models explored include Temporal Convolutional Networks (TCN), Neural Basis Expansion Analysis for Time Series Forecasting (N-BEATS), Temporal Fusion Transformers (TFT), Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS), and Time-series Dense Encoder (TiDE). Furthermore, we introduce the Extended Long Short-Term Memory for Time Series (xLSTM-TS) model, an xLSTM adaptation optimised for time series prediction. Wavelet denoising techniques were applied to smooth the signal and reduce minor fluctuations, providing cleaner data as input for all approaches. Denoising significantly improved performance in predicting stock price direction. Among the models tested, xLSTM-TS consistently outperformed others. For example, it achieved a test accuracy of 72.82% and an F1 score of 73.16% on the EWZ daily dataset. By leveraging advanced deep learning models and effective data preprocessing techniques, this research provides valuable insights into the application of machine learning for market movement forecasting, highlighting both the potential and the challenges involved.

Summary

  • The paper introduces the xLSTM-TS model that employs exponential gating and novel memory structures to overcome traditional LSTM limitations.
  • The study benchmarks model performance using precise metrics like MAE, RMSE, and F1 scores to demonstrate superior prediction accuracy.
  • The paper highlights the critical role of wavelet denoising in preprocessing to enhance deep learning models for financial forecasting.

An Evaluation of Deep Learning Models for Stock Market Trend Prediction

Authors: Gonzalo López Gil, Paul Duhamel-Sebline, Andrew McCarren

In their paper, "An Evaluation of Deep Learning Models for Stock Market Trend Prediction," Gonzalo López Gil, Paul Duhamel-Sebline, and Andrew McCarren provide a rigorous examination of various deep learning models for the task of stock market trend prediction. The authors leverage historical data from the S&P 500 index and the Brazilian ETF EWZ to assess the efficacy of different models, including their proposed Extended Long Short-Term Memory for Time Series (xLSTM-TS).

Methodology and Models

The investigation encompasses several advanced deep learning architectures: Temporal Convolutional Networks (TCN), Neural Basis Expansion Analysis for Time Series Forecasting (N-BEATS), Temporal Fusion Transformers (TFT), Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS), and the Time-series Dense Encoder (TiDE). Among these, the xLSTM-TS stands out as the focal point of the paper. This model introduces exponential gating and novel memory structures to overcome common LSTM limitations such as gradient vanishing and inadequate handling of long-term dependencies.

Data preprocessing, particularly wavelet denoising, is highlighted as a crucial step to enhance model performance. Wavelet transforms decompose non-stationary signals into approximation and detail coefficients, which are used to remove noise and provide a cleaner dataset. This preprocessing significantly improves the models' ability to predict stock price directions.

Results and Performance Metrics

The paper's findings demonstrate the superiority of the xLSTM-TS model across multiple datasets. The model consistently outperforms others, achieving a test accuracy of 72.82% and an F1 score of 73.16% on the EWZ daily dataset. For the S&P 500 daily data, the xLSTM-TS model attained an F1 score of 73.00%, indicating robust performance in predicting stock market trends.

By employing metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Root Mean Squared Scaled Error (RMSSE), and Mean Absolute Scaled Error (MASE), the authors provide a comprehensive evaluation of the models' performance. Notably, the xLSTM-TS model achieves the lowest MAE and RMSE across the evaluated datasets, underscoring its predictive accuracy. Furthermore, the implementation of wavelet denoising plays a crucial role in enhancing model performance.

Discussion

The paper underscores the importance of selecting appropriate model architectures and effective preprocessing techniques in financial time series forecasting. While the xLSTM-TS model demonstrates significant improvements over state-of-the-art models like TCN and TiDE, it is evident that the choice of preprocessing methods, such as wavelet denoising, critically impacts model accuracy.

The results also reveal variability in model performance across different datasets and timeframes. Predicting daily closing prices generally results in higher accuracy compared to hourly prices, attributable to the higher volatility and smaller price variations in the latter. This highlights the inherent challenges of high-frequency data prediction and the necessity for robust model architectures to handle such complexities effectively.

Implications and Future Work

The practical implications of this research are manifold. Accurate stock market trend predictions can significantly impact investment strategies and financial decision-making. The demonstrated success of the xLSTM-TS model offers promising avenues for its application in real-world trading environments. Future work should focus on generalizing these findings across a broader range of financial instruments, including commodities and foreign exchange. Additionally, integrating more sophisticated preprocessing techniques and exploring novel deep learning architectures could further enhance predictive accuracy.

Furthermore, real-world implementation of these models in automated trading systems could yield valuable insights and drive advancements in financial forecasting.

In conclusion, this paper provides a comprehensive evaluation of deep learning models for stock market trend prediction, with the xLSTM-TS model demonstrating robust performance across various datasets. The paper highlights the critical role of data preprocessing and model selection in enhancing predictive accuracy, offering valuable insights for future research and practical applications in the field of financial forecasting.