The paper "Ensemble of temporal Transformers for financial time series" by Kenniy Olorunnimbe and Herna Viktor introduces an ensemble approach to enhance the accuracy of price forecasts in financial markets. The authors compare traditional models like ARIMA with state-of-the-art deep learning techniques, highlighting that temporal Transformers with similarity embedding outperform these methods for multi-horizon forecasts in financial time series due to their ability to account for conditional heteroscedasticity. To further improve accuracy, they propose an ensemble method that combines multiple temporal Transformer models trained within sliding windows, optimizing data utilization. They also introduce a stacking meta-learner that uses a quantile estimator to determine optimal weights for combining base models. Experiments using 20 companies from the Dow Jones Industrial Average show over 40% and 60% improvements in predictive performance compared to baseline temporal Transformers, particularly in volatile and non-volatile extrapolation periods. The study emphasizes the importance of ensemble techniques in financial deep learning, especially for handling the non-constant variance in financial time series data.The paper "Ensemble of temporal Transformers for financial time series" by Kenniy Olorunnimbe and Herna Viktor introduces an ensemble approach to enhance the accuracy of price forecasts in financial markets. The authors compare traditional models like ARIMA with state-of-the-art deep learning techniques, highlighting that temporal Transformers with similarity embedding outperform these methods for multi-horizon forecasts in financial time series due to their ability to account for conditional heteroscedasticity. To further improve accuracy, they propose an ensemble method that combines multiple temporal Transformer models trained within sliding windows, optimizing data utilization. They also introduce a stacking meta-learner that uses a quantile estimator to determine optimal weights for combining base models. Experiments using 20 companies from the Dow Jones Industrial Average show over 40% and 60% improvements in predictive performance compared to baseline temporal Transformers, particularly in volatile and non-volatile extrapolation periods. The study emphasizes the importance of ensemble techniques in financial deep learning, especially for handling the non-constant variance in financial time series data.