What does this research mean for the field?

Machine learning models fail to outperform trivial benchmarks or buy-and-hold strategies in forecasting ETF returns across both developed and emerging markets, demonstrating that statistical accuracy does not necessarily translate into economic value. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To evaluate whether machine learning models can extract genuine predictive signals and generate economic value from ETF returns across markets with differing efficiency levels.

September 15, 2026MathematicsOpen Access

Statistical Accuracy, Economic Value and Model Instability in ETF Return Forecasting: A Comparison Across Developed and Emerging Markets

View Full Paper

Ask AI

Bookmark

Authors

EBEdson Vinícius Pontes Bastos RFRoberto Ivo da Rocha Lima Filho LMLino Guimarães Marujo

Discussion

Member takes

Overview

Comparative analysis evaluates machine learning algorithms for ETF return forecasting across developed and emerging markets, revealing poor economic value and failure to beat simple baselines.

Key Points

To evaluate whether machine learning models can extract genuine predictive signals and generate economic value from ETF returns across markets with differing efficiency levels.
Evaluated historical data from January 2010 to July 2026 (trained through December 2022, tested thereafter) on iShares MSCI Brazil (EWZ) and iShares Core S&P 500 (IVV).
Tested Random Forest, XGBoost (random search and Bayesian optimization), LSTM, GRU, and an LSTM + XGBoost ensemble across 1-day, 5-day, and 21-day horizons using 10 technical indicators.
Benchmarked models against historical mean, random walk, AR(1), and majority-class classifiers using Diebold–Mariano tests with false-discovery control and bootstrap intervals.
No machine learning model outperformed the trivial majority-class classifier or the historical mean benchmark under Diebold–Mariano testing across any horizon.
Performance failed via distinct market mechanisms: predictions collapsed onto the majority class in the developed market, while generating dispersed but unprofitable signals in the emerging market.
No trading strategy surpassed Buy-and-Hold, no Sharpe ratio was distinguishable from zero in the emerging market, and isolated directional significance at the monthly horizon yielded no economic value.

Cite This Study

Bastos et al. (2026) studied this question.

synapsesocial.com/papers/6aa913609013453be30a1393 https://doi.org/10.3390/math14183318

View Full Paper

Ask AI

Bookmark

September 15, 2026MathematicsOpen Access

Statistical Accuracy, Economic Value and Model Instability in ETF Return Forecasting: A Comparison Across Developed and Emerging Markets

View Full Paper

Ask AI

Bookmark

Authors

EBEdson Vinícius Pontes Bastos RFRoberto Ivo da Rocha Lima Filho LMLino Guimarães Marujo

Discussion

Member takes

Overview

Comparative analysis evaluates machine learning algorithms for ETF return forecasting across developed and emerging markets, revealing poor economic value and failure to beat simple baselines.

Key Points

To evaluate whether machine learning models can extract genuine predictive signals and generate economic value from ETF returns across markets with differing efficiency levels.
Evaluated historical data from January 2010 to July 2026 (trained through December 2022, tested thereafter) on iShares MSCI Brazil (EWZ) and iShares Core S&P 500 (IVV).
Tested Random Forest, XGBoost (random search and Bayesian optimization), LSTM, GRU, and an LSTM + XGBoost ensemble across 1-day, 5-day, and 21-day horizons using 10 technical indicators.
Benchmarked models against historical mean, random walk, AR(1), and majority-class classifiers using Diebold–Mariano tests with false-discovery control and bootstrap intervals.
No machine learning model outperformed the trivial majority-class classifier or the historical mean benchmark under Diebold–Mariano testing across any horizon.
Performance failed via distinct market mechanisms: predictions collapsed onto the majority class in the developed market, while generating dispersed but unprofitable signals in the emerging market.
No trading strategy surpassed Buy-and-Hold, no Sharpe ratio was distinguishable from zero in the emerging market, and isolated directional significance at the monthly horizon yielded no economic value.

Cite This Study

Bastos et al. (2026) studied this question.

synapsesocial.com/papers/6aa913609013453be30a1393 https://doi.org/10.3390/math14183318

View Full Paper

Ask AI

Bookmark