Deep Learning Outperforms Traditional Models in Korean Stock Price Prediction

On April 2, 2025, Sungwoo Kang published Stock Price Prediction Using Triple Barrier Labeling and Raw OHLCV Data: Evidence from Korean Markets, revealing that LSTM networks using raw OHLCV data achieve comparable results to traditional models with technical indicators in predicting stock prices.

This paper demonstrates that LSTM networks trained on raw OHLCV data achieve comparable performance to traditional models using technical indicators for stock price prediction in Korean markets. Using a dataset from 2006 to 2024, the study optimizes triple barrier labeling with a 29-day window and 9% barriers. The best-performing LSTM configuration uses a window size of 100 and hidden size of 8. Full OHLCV data provides better predictive accuracy than close price or close price with volume alone. These findings suggest that simpler approaches focusing on raw data may outperform complex feature engineering strategies in financial forecasting.

Feature engineering has proven to be a critical component in enhancing the accuracy of ML-based stock price predictions. By carefully selecting and transforming variables such as technical indicators (e.g., moving averages, RSI), macroeconomic factors, and even sentiment data from social media, researchers have been able to significantly improve model performance. For instance, studies have shown that combining multiple datasets can yield more robust predictions than relying on a single source of information.

While traditional statistical models like ARIMA (AutoRegressive Integrated Moving Average) and linear regression have long been staples in financial forecasting, they often fall short when dealing with the non-linear relationships inherent in stock markets. LSTM networks, by contrast, are better equipped to handle such complexities due to their ability to learn from long-term dependencies in data. This has led to a growing preference for ML models among practitioners seeking more accurate and reliable predictions.

As machine learning continues to evolve, so too does its potential application in financial markets. Innovations such as attention mechanisms and hybrid models that combine the strengths of different approaches are opening new avenues for research. However, challenges remain, particularly in ensuring data quality, avoiding overfitting, and maintaining model interpretability. Despite these hurdles, the future looks promising for ML-driven stock price prediction, with ongoing advancements poised to revolutionize how financial markets operate.

Integrating machine learning into stock price prediction represents a significant leap forward in financial forecasting. By leveraging advanced models like LSTM networks and carefully engineered features, researchers and practitioners are unlocking new possibilities for understanding market dynamics. While there is still much work to be done, the potential for ML to transform the financial landscape is undeniable. As this technology continues to mature, it will undoubtedly play an increasingly central role in shaping how we approach investment decisions in the years to come.