New Hybrid Machine Learning Model Predicts Financial Market Volatility with Increased Accuracy

Researchers at Carnegie Mellon University have developed a hybrid machine learning model called GARCH-Informed Neural Network (GINN) that combines GARCH and long short-term memory deep neural networks to predict financial market volatility with increased accuracy. The model outperforms both GARCH and competing models in predicting daily close prices across seven major stock market indexes worldwide. The study highlights the potential of GINN for applications in time series modeling and prediction.

The world of finance is a dynamic and unpredictable one, where market volatility plays a significant role in investment risk and returns. Over the years, various statistical methods have been employed to forecast time series volatility, with the Autoregressive Conditional Heteroskedasticity (ARCH) model being a notable Nobel Prize-winning approach [1]. However, most ARCH-based models lack the flexibility to capture nonlinear market features, leading to limited accuracy and generalizability.

To address this challenge, researchers at Carnegie Mellon University (CMU) have developed an innovative hybrid deep learning model called GARCH-Informed Neural Network (GINN) [1]. This groundbreaking model combines the strengths of Generalized-ARCH (GARCH) with the flexibility of a long short-term memory deep neural network (LSTM) to provide more accurate and generalizable financial market volatility predictions.

The GINN model's creation was inspired by physics-informed machine learning, which directly embeds physical laws into the architecture of a deep learning model. In the context of finance, the researchers merged machine learning with stylized facts, empirical market patterns captured by the GARCH model. This approach enables GINN to learn from both the factual ground truth and the knowledge acquired by the GARCH model, thereby grasping both general market trends and finer details.

According to Zeda Xu, a CMU Ph.D. student and the lead author of the study, traditional machine learning models risk overfitting, which occurs when a model closely mimics the data it has been trained on. By building a hybrid model like GINN, researchers ensure generalizability and improved accuracy [1].

The results of the study are impressive, with GINN outperforming both the GARCH model and competing models in predicting daily close prices across seven major stock market indexes worldwide [1]. The model achieved a 5% improvement in volatility prediction accuracy compared to the GARCH model alone.

These findings are not only valuable for investors who rely on GARCH as a resource but also for other applications involving time series modeling and prediction, such as autonomous vehicles and GenAI. As the financial world continues to evolve, the need for more accurate and generalizable market volatility predictions will only grow, making the GINN model an essential tool for financial professionals and researchers alike.

References:
[1] CMU Engineering. (2024, December 13). Trend is your friend: CMU researchers develop hybrid model for more accurate financial market volatility predictions. Retrieved from https://engineering.cmu.edu/news-events/news/2024/12/13-trend-is-your-friend.html