Optimizing investment horizons: Machine learning applications in technical analysis of the WIG20 index

The profitability of combining machine learning with technical analysis is assessed using the WIG20 index and three of its constituent stocks with extensive historical data. While most previous studies have focused on short-term forecasts derived from technical indicators, the effectiveness of machine learning-enhanced technical analysis across different time frames and market conditions remains underexplored. To bridge this gap, XGBoost classification models incorporating 771 technical indicators’ variations within 121 types are applied to evaluate predictive accuracy and trading performance over forecast horizons ranging from 5 to 120 trading days between 2020 and 2023. This scale of technical indicators is much wider than the most studies in the field. The findings demonstrate stronger predictive performance at extended horizons (60–120 days) for some analyzed instruments. The research contradicts the widely accepted belief that technical analysis is primarily effective for short-term trading. Additionally, the study provides empirical evidence supporting the superiority of machine learning strategies based on technical indicators over conventional buy-and-hold approaches. It bridges the gap between theoretical profitability and practical implementation. • Machine learning-enhanced technical analysis shows stronger predictive performance at extended horizons (60– 120 days), challenging the conventional view that technical indicators primarily benefit short-term trading. • XGBoost classification models incorporating 771 technical indicators outperformed traditional buy-and-hold strategies for the WIG20 index and selected constituent stocks in terms of both returns and risk-adjusted returns. • Despite theoretical profitability in longer-term forecasts, practical implementation faces market liquidity constraints, with viable strategies limited to the 60-day investment horizon for specific instruments.