Planting Portfolios - Panel Trees in the Swedish Equity Market

Sammanfattning

This study evaluates the performance of the P-Tree algorithm (Cong et al., 2025) in the Swedish equity market, a setting characterized by lower liquidity, fewer listed stocks, and limited historical data. The P-Tree algorithm identifies return patterns by splitting stocks based on firm characteristics, capturing both linear and non-linear relationships. In the Swedish context, the in-sample Sharpe ratio of the P-Tree portfolio was 0.92, exceeding the market portfolio's 0.64. Out-of-sample, the Sharpe ratio was 0.53 when training on the first half of the sample and testing on the second half, and 0.48 in the reverse scenario. The P-Tree strategy produced a statistically significant Fama-French three-factor of 9.61% in-sample, and 3.21% out-of-sample when predicting from the past to the future. In the reverse future-to-past prediction, the was 3.39%, which was not statistically significant at the 10% confidence level. The P-Tree model offers flexible structure, with the number of splits adapted to the available data. In markets with limited historical or cross-sectional information, a shallow-tree configuration can be implemented that focuses on the most informative characteristics, avoiding overfitting. In data-rich environments, a deep-tree configuration can capture complex, non-linear relationships while maintaining robustness. This adaptability ensures that P-Trees balance predictive power with data availability, delivering interpretable and reliable predictions. Such properties make them a promising tool for portfolio optimization, risk management, and factor-based investing in the finance industry.