Answers are grounded in this paper and cite their sources.
This study extends the Markowitz-Informed Neural Network (MINN) of Smyth, W. S., Ernst, P. A., and Miao, Y. (2025. Markowitz-Informed Neural Networks (MINNs): An interpretable deep learning approach to portfolio optimization. SSRN. https://doi.org/10.2139/ssrn.5336779) to a macro-financial setting for 20 emerging economies (January 2012–June 2025), allocating across 42 assets conditioned on 81 factors under a strict real-time information constraint and monthly rebalancing. Out-of-sample, the robust minimum covariance determinant estimator attains the highest risk-adjusted performance (net Sharpe 1.131), whereas Ledoit-Wolf shrinkage deteriorates monotonically in intensity (0.733–0.353); the two corrections diverge. The neural specifications generate higher terminal wealth than every optimiser-based benchmark but do not surpass naive diversification, with gains concentrated in the 2021–2022 inflation surge. No pairwise Sharpe difference is statistically significant over 66 months.
No takes yet. Share an insight, caveat, or question.
Eyyüp Ensari Şahin (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: