Hybrid RNN models with quantum-inspired activation functions for financial time series forecasting in fintech

Abstract

In the context of the global technological revolution across all sectors, quantum computing is emerging as a breakthrough force in the financial field, with capabilities to handle big data, simulate complex fluctuations, and optimize computation time superior to classical computers. Globally, quantum computing has been applied by major corporations such as IBM, Google, and JPMorgan Chase to solve problems like portfolio optimization and derivative pricing, reducing the time from hours to seconds thanks to the principles of superposition and entanglement in quantum physics. This research focuses on a model that combines classical mathematics (genetic algorithm for feature selection) and quantum computing, leveraging the advantages of superposition states. The proposed model is validated on currency exchange rate datasets (from AUD/JPY). The architecture proposes replacing standard activation functions with a Quantum-Inspired Activation Function (QIAF) implemented as a direct mathematical function, combined with norm distribution to enhance nonlinearity and standardize normal distribution with bias to stabilize gradients in noisy data. The results show a Sharpe ratio of 2.85 and a maximum drawdown of -8. 2%, demonstrating the effectiveness of the proposed model. This research highlights the need for investment in quantum infrastructure so that developing countries are not left behind in the global fintech race, promoting sustainable digital transformation in finance is inevitable.