On the Limits of Low-Frequency OHLCV Signals in Machine Learning-Driven Portfolio Optimization A Cost-Aware, Leak-Free Benchmark on Large-Cap US Equities
A 2026 paper by Pardo on portfolio optimization in equities, using a transformer on a 2010–2025 sample: the abstract mentions a Sharpe ratio, alpha, drawdowns, an out-of-sample test and transaction costs and leaves out a benchmark, multiple testing and the size of the universe.
“We report a negative result: in the out-of-sample period (2017-2025) the best model (MLP, Sharpe 0.96) only marginally edges a no-ML Black-Litterman "Prior-Only" baseline (Sharpe 0.94) that shares the same covariance cleaning, while incurring more than three times the cumulative transaction cost (15.13% vs. 4.67%).”
Also stated: Daily, 2010–2025, Sharpe ratio, Alpha, Drawdown, Long-short, Price and volume, out-of-sample test, costs
Missing from abstract: benchmark, multiple testing, universe
- Audit of the paper claims accounting for costs, luck, crises, originality, geography, etc.
- Reproduction code integratable into Bloomberg, IBKR, etc.
- Bias-free datasets for backtesting included
- Delivered in 2 working days
| Similar papers | citation | same method | same asset | same authors | |
|---|---|---|---|---|---|
| An Automated Black-Litterman Framework for Global Fixed Income by Carver et al. | 2026 | ||||
| Comparative Evaluation of Modern Deep Learning Methodologies for Portfolio Optimization by Ozechi et al. | 2026 | ||||
| A Sustainable Portfolio Construction Model Based on ESG and Deep Learning Algorithms: Evidence from the U.S. Market by Sajadi et al. | 2025 | ||||
| Earnings Forecasting and Mean-Variance Efficient Portfolios in the United States by Guerard et al. | 2024 | ||||
| The Development and Evolution of Mean-Variance Efficient Portfolios in Japan: 30 Years After by Guerard and Beheshti | 2023 |