Cross-market attention for futures forecasting and ETF performance enhancement

This study investigates whether U.S. investor attention can predict Taiwan's stock index futures returns and enhance exchange-traded fund (ETF) investment performance. Investor attention measures are constructed from 52-week high and low prices, and forecasting models are developed by integrating the elastic net (ENet) with factor-based dimensionality reduction techniques to improve out-of-sample predictive accuracy. The empirical results show that the U.S. investor attention variables individually exhibit significant predictive power for Taiwan's stock index futures returns. Moreover, ENet combined with factor-based dimensionality reduction models provide the most robust forecasting gains, outperforming not only the buy-and-hold and historical average benchmarks, but also models based solely on individual predictors and conventional dimensionality reduction approaches. Finally, the findings show that investors holding Taiwan-focused ETFs can enhance portfolio performance by dynamically adjusting their index futures positions in response to model-generated signals.