Dynamic Trading Strategies for Volatile Assets: A Hybrid GM-LSTM Model with Finite State Machine Optimization

Haoran Sun, Bohan Song, Meng Cong, Zeran Wang, Zexian Liu, Xueru Wang, Yujie Han, Xinqi Cui
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:632-643, 2025.

Abstract

This study addresses the challenge of predicting investments in highly volatile assets, such as gold and ancient coins, by proposing a hybrid forecasting strategy that integrates the Grey Model (GM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). A finite automaton-based trading decision model is developed to enable rapid decision-making in dynamic market environments. Traditional methods often struggle to adapt to the unpredictability of such assets, prompting the need for advanced predictive frameworks. The methodology encompasses data preprocessing, model training, and validation, with a focus on optimizing short- and long-term forecasts. Experimental results demonstrate that the hybrid GM-LSTM strategy significantly enhances prediction accuracy: GM excels in short-term forecasting (first 200 days) due to its efficiency with limited data, while LSTM outperforms in long-term scenarios by capturing complex temporal dependencies. A dynamic weight adjustment mechanism, incorporating profit (PI) and risk indices (RI), balances returns and risks. Sensitivity analysis reveals the model’s robustness under varying transaction costs (0.1%–10%), maintaining profitability even at higher cost levels. Key performance metrics—annualized return, Sharpe ratio, and maximum drawdown—validate the strategy’s superiority over benchmarks like buy-and-hold. The state machine-driven trading model, evaluated through Value at Risk (VaR) and sliding window protocols, ensures adaptability across market conditions. This work provides traders with a data-driven decision-making tool, optimizing investment strategies while mitigating risks in volatile markets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-sun25b, title = {Dynamic Trading Strategies for Volatile Assets: A Hybrid GM-LSTM Model with Finite State Machine Optimization}, author = {Sun, Haoran and Song, Bohan and Cong, Meng and Wang, Zeran and Liu, Zexian and Wang, Xueru and Han, Yujie and Cui, Xinqi}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {632--643}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/sun25b/sun25b.pdf}, url = {https://proceedings.mlr.press/v278/sun25b.html}, abstract = {This study addresses the challenge of predicting investments in highly volatile assets, such as gold and ancient coins, by proposing a hybrid forecasting strategy that integrates the Grey Model (GM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). A finite automaton-based trading decision model is developed to enable rapid decision-making in dynamic market environments. Traditional methods often struggle to adapt to the unpredictability of such assets, prompting the need for advanced predictive frameworks. The methodology encompasses data preprocessing, model training, and validation, with a focus on optimizing short- and long-term forecasts. Experimental results demonstrate that the hybrid GM-LSTM strategy significantly enhances prediction accuracy: GM excels in short-term forecasting (first 200 days) due to its efficiency with limited data, while LSTM outperforms in long-term scenarios by capturing complex temporal dependencies. A dynamic weight adjustment mechanism, incorporating profit (PI) and risk indices (RI), balances returns and risks. Sensitivity analysis reveals the model’s robustness under varying transaction costs (0.1%–10%), maintaining profitability even at higher cost levels. Key performance metrics—annualized return, Sharpe ratio, and maximum drawdown—validate the strategy’s superiority over benchmarks like buy-and-hold. The state machine-driven trading model, evaluated through Value at Risk (VaR) and sliding window protocols, ensures adaptability across market conditions. This work provides traders with a data-driven decision-making tool, optimizing investment strategies while mitigating risks in volatile markets.} }
Endnote
%0 Conference Paper %T Dynamic Trading Strategies for Volatile Assets: A Hybrid GM-LSTM Model with Finite State Machine Optimization %A Haoran Sun %A Bohan Song %A Meng Cong %A Zeran Wang %A Zexian Liu %A Xueru Wang %A Yujie Han %A Xinqi Cui %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-sun25b %I PMLR %P 632--643 %U https://proceedings.mlr.press/v278/sun25b.html %V 278 %X This study addresses the challenge of predicting investments in highly volatile assets, such as gold and ancient coins, by proposing a hybrid forecasting strategy that integrates the Grey Model (GM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). A finite automaton-based trading decision model is developed to enable rapid decision-making in dynamic market environments. Traditional methods often struggle to adapt to the unpredictability of such assets, prompting the need for advanced predictive frameworks. The methodology encompasses data preprocessing, model training, and validation, with a focus on optimizing short- and long-term forecasts. Experimental results demonstrate that the hybrid GM-LSTM strategy significantly enhances prediction accuracy: GM excels in short-term forecasting (first 200 days) due to its efficiency with limited data, while LSTM outperforms in long-term scenarios by capturing complex temporal dependencies. A dynamic weight adjustment mechanism, incorporating profit (PI) and risk indices (RI), balances returns and risks. Sensitivity analysis reveals the model’s robustness under varying transaction costs (0.1%–10%), maintaining profitability even at higher cost levels. Key performance metrics—annualized return, Sharpe ratio, and maximum drawdown—validate the strategy’s superiority over benchmarks like buy-and-hold. The state machine-driven trading model, evaluated through Value at Risk (VaR) and sliding window protocols, ensures adaptability across market conditions. This work provides traders with a data-driven decision-making tool, optimizing investment strategies while mitigating risks in volatile markets.
APA
Sun, H., Song, B., Cong, M., Wang, Z., Liu, Z., Wang, X., Han, Y. & Cui, X.. (2025). Dynamic Trading Strategies for Volatile Assets: A Hybrid GM-LSTM Model with Finite State Machine Optimization. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:632-643 Available from https://proceedings.mlr.press/v278/sun25b.html.

Related Material