A1970
Title: The application of various architectures of the LSTM model in algorithmic investment strategies on BTC and S\&P500 Index
Authors: Robert Slepaczuk - University of Warsaw (Poland) [presenting]
Abstract: The use of various architectures of the LSTM model in algorithmic investment strategies is investigated. LSTM models are used to generate buy/sell signals, with previous levels of Bitcoin price and the S\&P500 Index value as inputs. Four approaches are tested: two are regression problems (price level prediction) and the other two are classification problems (prediction of price direction). All approaches are applied to daily, hourly, and 15-minute data and use a walk-forward optimization procedure with numerous IS and OOS periods. The out-of-sample period for the S\&P 500 Index is from February 6, 2014 to August 26, 2022, and for Bitcoin it is from February 1, 2014 to August 26, 2022. We discover that classification techniques beat regression methods on average and that intraday models perform much better in the classification approach, while daily ones produce outperforming results in the case of regression methods. The research covers s3 types of ensemble models: through frequencies, assets, and the combination of both of them. We conclude that the ensembling of models positively affects their performance only on the condition of specific characteristics of the component parts. Finally, a sensitivity analysis is performed to determine how changes in the main hyperparameters of the LSTM model affect strategy performance. It reveals that we can distinguish the specific hyperparameters, which can increase the performance of LSTM model.
