Multi-Horizon Equity Returns Predictability via Machine Learning

We investigate the predictability of global expected stock returns across various forecasting horizons using machine learning techniques. We find that the predictability of returns decreases with longer forecasting horizons both in the U.S. and internationally. Despite this, we provide evidence that using firm-specific characteristics can remain profitable even after accounting for transaction costs, especially when we consider longer forecasting horizons. Studying the profitability of long-short portfolios, we highlight a trade-off between higher transaction costs connected to frequent rebalancing and greater returns on shorter horizons. Increasing the forecasting horizon while matching the rebalancing period increases risk-adjusted returns after transaction costs for the U.S. We combine predictions of expected returns at multiple horizons using double-sorting and a turnover reducing strategy, buy/hold spread. Double sorting on different horizons significantly increases profitability in the U.S. market, while buy/hold spread portfolios exhibit better risk-adjusted profitability.