Sunday 02 February 2025
A team of researchers has developed a new system that can predict stock prices more accurately than individual models by combining the strengths of multiple approaches. The system, called Numin, uses a weighted-majority ensemble (WMA) to aggregate predictions from eight different deep-learning based models.
The WMA algorithm works by assigning weights to each model based on its past performance. These weights are then used to combine the predictions made by each model to produce a single output. The researchers found that this approach outperformed individual models in terms of both accuracy and potential profitability, even when some of the individual models were not profitable.
The team tested their system using a dataset of stock prices from the past five years, with a focus on predicting the returns of 10-candle periods (about 50 minutes). They used two different metrics to evaluate the performance of the WMA algorithm: accuracy and utility. Accuracy measures how well the predictions match the actual outcomes, while utility is a proxy for potential profitability.
The results showed that the WMA algorithm outperformed individual models in both accuracy and utility. When using accuracy as the weighting metric, the WMA algorithm achieved an average accuracy of 23.6% compared to 21.4% for the best individual model. When using utility as the weighting metric, the WMA algorithm achieved an average utility of 0.078, indicating a potential profit.
The researchers found that the performance of the WMA algorithm varied depending on the window size used. A smaller window size (5-10 minutes) performed better when using utility as the weighting metric, while a larger window size (20 minutes) performed better when using accuracy as the weighting metric.
The WMA algorithm is not only useful for predicting stock prices but also has potential applications in other areas such as trading and finance. The researchers hope that their system can be used to develop more effective trading strategies and improve the overall performance of financial markets.
In a world where high-frequency trading and machine learning are increasingly dominant, this new system could be a game-changer for those looking to make a profit from the stock market. By combining the strengths of multiple models, Numin offers a level of sophistication and accuracy that individual models simply can’t match. As the researchers continue to fine-tune their system, it will be fascinating to see how it performs in real-world scenarios.
Cite this article: “Numin: A Weighted-Majority Ensemble Algorithm for Accurate Stock Price Prediction”, The Science Archive, 2025.
Stock Prices, Deep-Learning Models, Weighted-Majority Ensemble, Wma Algorithm, Accuracy, Utility, Profitability, Machine Learning, Trading, Finance
