With an accuracy of 83%: A neural network predicting crises in the stock market was created at the Higher School of Economics

The forecast is given 24 hours before the event

Scientists from the Higher School of Economics have created a neural network that can warn of an impending short-term stock market crisis with an accuracy of over 83% 24 hours before the event.

The work has high practical significance for the national financial sector: it offers effective tools for the timely detection of market shocks, which is especially important for an unstable macroeconomic environment.
Tamara Teplova, Professor of the Faculty of Economic Sciences at the Higher School of Economics

A hybrid model was created to obtain such a forecast. It combines three machine learning architectures — an attention mechanism, temporal convolutional networks, and LSTM, which allows AI to have short-term memory. This innovation in the application of complex structures to Russian exchange data has not previously been used by economists.

To train the model, researchers analyzed data from 2014 to 2024, including market and macroeconomic indicators, such as the Moscow Exchange IMOEX index and investor sentiment indicators. Scientists had to take into account the rarity of crises and the influence of subjective sentiments on investor behavior.

To solve these problems, composite indices of internal and external investment sentiment were developed using the principal component method. These indices complement traditional macroeconomic and market variables, allowing us to capture the hidden emotional signals of traders over longer time horizons. Thanks to this, the model has achieved high forecasting accuracy, which makes it possible to predict a short-term crisis in the stock market one day before it begins.

The model effectively processes uneven data and achieves an accuracy of 78.70% when predicting crisis events on the day of observation and 78.85% on the next trading day. The use of monthly retraining and adaptive time windows made it possible to increase the accuracy to 83.87%. The key factors influencing predictions were exchange indicators, the capitalization of companies - stock issuers and market exchange rates.
Tamara Teplova, Professor of the Faculty of Economic Sciences at the Higher School of Economics

Earlier www1.ru reported that Smart Engines created a new bipolar morphological neural network.

Read materials on the topic:

Scientists at NSU have created a powerful AI that works on the principle of the human brain

Organs on order: an innovative bioreactor will be created at Sechenov University

Neural activity of the brain will make the drone fly: the video shows the operation of the NeuroPlay-6C headset.

Sources
TASS

Now on home