Related Experiment Videos
Enhancing Deep Learning Forecasts with Wavelet Decomposition: Evidence from the Ghana Stock Exchange
Osei K Tweneboah1, Maria C Mariani2
1Ramapo Data Science Program, Ramapo College of New Jersey, Mahwah, NJ 07430, USA.
Entropy (Basel, Switzerland)
|July 28, 2026
Summary
Discrete wavelet transformation improves stock market return forecasting in emerging economies. Wavelet preprocessing enhanced deep learning models, with Wavelet-LSTM showing the best predictive performance for the Ghana Stock Exchange Composite Index.
Area of Science:
- * Computational Finance
- * Econometrics
- * Machine Learning
Background:
- * Emerging economies present unique challenges for stock market forecasting, including high volatility and data limitations.
- * Deep learning models show promise but require optimization for complex financial time series.
- * Discrete wavelet transformation is a signal processing technique that can decompose data into different frequency components.
Purpose of the Study:
- * To investigate the effectiveness of discrete wavelet transformation in enhancing deep learning model performance for stock market return forecasting in emerging markets.
- * To evaluate standard deep learning models against their wavelet-preprocessed counterparts.
- * To identify the optimal hybrid model for forecasting the Ghana Stock Exchange Composite Index.
Main Methods:
- * Daily returns data for the Ghana Stock Exchange Composite Index (GSE-CI) from 2011 to 2022 were utilized.
- * Three deep learning architectures—Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN)—were employed.
- * Models were evaluated in their standard form and after preprocessing with Daubechies-4 (db4) discrete wavelet transform.
Main Results:
- * Wavelet preprocessing consistently reduced forecasting errors across all tested deep learning models.
- * Wavelet-enhanced models demonstrated superior performance compared to their baseline versions.
- * The Wavelet-LSTM model achieved the lowest forecasting error, indicating its strong predictive capability.
Conclusions:
- * Discrete wavelet transformation is an effective technique for multiscale feature extraction and noise reduction in financial time series from emerging markets.
- * Wavelet augmentation enhances the predictive accuracy of deep learning models, offering a richer representation of nonlinear temporal dynamics.
- * This study provides valuable insights for forecasting emerging financial markets, particularly in data-constrained environments.