Neural Nets to Forecast Switches of Market Conditions: Empiric Research on Crude Oil Market

The article studies the goal of automatic forecasting switches of different market conditions on oil market with the help of neural net models. For the analysis the authors used data on oil (Brent) prices in 2000-2025. Classification of market periods was done on the basis of fluctuating statistics – volatility, asymmetry and excess, which provides an opportunity to identify automatically stable, volatile and crisis phases. The efficiency of three types of competitive neural nets is compared: Simple RNN, LSTM and GRU. It was found that more complicated architecture (LSTM, GRU) surpasses the basic one (RNN) in accuracy and full identification of events of condition switch. The authors highlighted the importance of rolling-sign engineering and showed that this approach provides sustainability and adaptability of models to market changes. Research findings demonstrate promising nature of deep neural nets for monitoring and early warning of market events. Finally, restrictions of the approach were discussed, as well as trends of further investigations, including integration of external data and development of methods of explainable AI.