Deep learning in financial time series forecasting

Deep learning based financial time series forecasting is considered during the work. Moreover, the problem of selecting a loss function for predicting data with speculative interest is also discussed. During the research, a number of neural network architectures (LSTM, Transformer, Tsmuxer) are tested, then the best model is integrated into a trading algorithm, which is further optimized according to its hyperparameters. At the last stage, the resulting trading algorithm is tested.