Russia Develops AI Model to Predict Stock Market Crises

Researchers at the Higher School of Economics have created a neural network capable of predicting short-term stock market downturns a day in advance with over 83 percent accuracy.
Russian scientists from the Higher School of Economics have developed a unique neural network model that can warn of an impending short-term stock market crisis within 24 hours. The system boasts an impressive 83 percent accuracy rate.
The team built a hybrid model combining three advanced machine learning techniques: attention mechanisms, temporal convolutional networks, and the LSTM method with short-term memory elements.
To train the AI, researchers analyzed data from 2014 to 2024, including the MOEX Russia Index, macroeconomic indicators, and investor sentiment. Initially, the model achieved 78.70 percent accuracy for same-day crisis predictions and 78.85 percent for the next day.
After a month of retraining and the use of adaptive time windows, the accuracy increased to 83.87 percent. The most significant factors in forecasting were market indicators, issuer company capitalization, and currency exchange rates.
According to the university’s press service, the development could become a valuable tool for investors, financial analysts, and regulators. It may also prove useful for other countries, especially those experiencing significant macroeconomic volatility, potentially enhancing national financial stability.