Article

Deep Learning Forecasting of FinTech Stock Performance under Digital Economy Regime Shifts: Evidence from Clean-Tech Financing

---

Abstract

This study investigates the predictive capability of deep learning architectures in forecasting the stock performance of FinTech firms under the structural evolution of the digital economy and the dynamics of clean-tech financing. Utilizing a panel dataset of 50 publicly listed FinTech companies across major markets from 2015 to 2024, and including macro-financial controls, digital economy activity indexes, and clean-technology financing proxies (green bond issuance and clean-tech equity returns), we first identify three distinct digital economy regimes (Normal, Expansion, Stress) via a Markov-switching model. Our regime classification results indicate that the expansion regime accounted for 42% of the sample period, exhibited an average digital-economy index increase of 4.8% per quarter, and corresponded with 35% higher clean-tech financing volumes relative to the normal regime. We then train and test three deep learning architectures (LSTM, GRU, Transformer) on a rolling-window scheme. The LSTM model achieves the lowest out-of-sample RMSE of 0.0189 and MAE of 0.0131, with Directional Accuracy (DA) of 61.4%. In regime-specific analysis, the LSTM delivers an RMSE of 0.0164 and DA of 64.8% in the Normal regime, RMSE of 0.0189 and DA of 62.3% in the Expansion regime, and RMSE of 0.0217 and DA of 58.1% in the Stress regime. Robustness checks reveal that combining green bond issuance and clean-tech equity returns as proxies yields a further RMSE reduction to 0.0186, and that an input window length of 30 days represents the optimal setting. The findings substantiate the importance of integrating regime labels and clean-tech financing features into forecasting models for FinTech stock returns. The study contributes to the literature by bridging digital finance, sustainable capital markets, and advanced machine-learning forecasting, and offers actionable insights for investors and policymakers navigating evolving digital-economy and clean-tech landscapes.

Keywords: Fintech Stock Forecasting, Digital Economy Regime Shifts, Clean-Tech Financing, Deep Learning, LSTM

How to Cite:

Ramadani, N. C. & Fahreza, S., (2025) “Deep Learning Forecasting of FinTech Stock Performance under Digital Economy Regime Shifts: Evidence from Clean-Tech Financing”, FinTech Innovation Journal 1(2), 137-163. doi: https://doi.org/10.63913/ftij.v1i1.94