This study addresses the problem of forecasting financial panel data in the presence of distributional shifts and cross-sectional heterogeneity, which often deteriorate the performance of traditional machine learning models. Conventional normalization methods are typically static and fail to adapt to non-stationary conditions, resulting in reduced predictive accuracy and model instability. To overcome these limitations, this paper proposes a hybrid framework that integrates Reversible Instance Normalization (RevIN) with a Multilayer Perceptron (MLP). The proposed approach applies instance-wise normalization during training and reverses the transformation at inference, thereby preserving the intrinsic statistical properties of each cross-sectional unit while improving model stability. The framework is evaluated using a balanced panel dataset of ten Iraqi banks over the period 2011-2024 to predict the Capital Adequacy Ratio (CAR). The results show that the proposed model outperforms the baseline MLP, achieving an R² of 0.824 and an RMSE of 0.555, indicating improved predictive accuracy and generalization. These findings demonstrate the effectiveness of adaptive normalization in enhancing model robustness under heterogeneous and non-stationary conditions. This study contributes by extending the application of RevIN to financial panel data and providing a practical, interpretable, and efficient tool for banking risk management and regulatory forecasting.
F. Jawadi, T. M. Pondie, and A. I. Cheffou, “New challenges for green finance and sustainable industrialization in developing countries: A panel data analysis,” Energy Economics, vol. 142, p. 108120, 2025.
D. Kotios, G. Makridis, G. Fatouros, and D. Kyriazis, “Deep learning enhancing banking services: a hybrid transaction classification and cash flow prediction approach,” Journal of Big Data, vol. 9, no. 1, p. 100, 2022.
N. Passalis, A. Tefas, J. Kanniainen, M. Gabbouj, and A. Iosifidis, “Deep Adaptive Input Normalization for Time Series Forecasting,” Sep. 2019, , [Online]. Available: http://arxiv.org/abs/1902.07892.
S. Makridakis, E. Spiliotis, and V. Assimakopoulos, “The M4 Competition: 100,000 time series and 61 forecasting methods,” International Journal of Forecasting, vol. 36, no. 1, pp. 54-74, 2020, doi: 10.1016/j.ijforecast.2019.04.014.
T. Kim, J. Kim, Y. Tae, C. Park, J.-H. Choi, and J. Choo, “Reversible instance normalization for accurate time-series forecasting against distribution shift,” in International Conference on Learning Representations, 2021.
C. Hsiao, Analysis of Panel Data, no. 64. Cambridge University Press, 2022.
J. M. Wooldridge, Introductory Econometrics: A Modern Approach, Sixth Edition. Cengage, 2015, [Online]. Available: www.cengage.com/highered.
W. H. Greene, Econometric Analysis, Eighth Edition. 2020.
J. Bai and S. Ng, “A new look at panel testing of stationarity and the PPP hypothesis,” Cambridge University Press, 2005.
A. Saly-Kaufmann, K. Wood, J. Peter-Calliess, and S. Zohren, “Deep learning for financial time series: A large-scale benchmark of risk-adjusted performance,” arXiv preprint arXiv:2603.01820, 2026.
I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT Press, 2016.
A. H. Abd-elaziem and T. H. M. Soliman, “A multi-layer perceptron (MLP) neural networks for stellar classification: A review of methods and results,” International Journal of Advanced Applied Computer Intelligence, vol. 3, no. 10, p. 54216, 2023.
R. J. Hyndman and G. Athanasopoulos, Forecasting: Principles and Practice. OTexts, 2018.
R. J. Hyndman and A. B. Koehler, “Another look at measures of forecast accuracy,” International Journal of Forecasting, vol. 22, no. 4, pp. 679-688, 2006.
