Adaptive Performance-WeightedSignal Ensemble : A ComparativeStudy of Portfolio Performance

Sammanfattning

The thesis investigates whether an adaptive ensemble of trading signals can outperform a traditional equal-weighted portfolio in terms of return and risk- adjusted performance. Financial markets are characterized by uncertainty, non- stationarity and varying market regimes. This makes it difficult for any single strategy to deliver strong performance all the time. To address this challenge, the study proposes an Adaptive Performance-Weighted(APW) framework that dynamically assigns weights to trading signals based on their recent predictive accuracy. The proposed model is evaluated using a backtesting framework applied to a diversified universe of 65 exchange-traded funds (ETFs) across multiple asset classes, including equities, fixed income commodities and currencies. The analysis covers the period from 2017 to 2025 and compares the APW model with a baseline Equal-Weighted (EW) portfolio. Performance is assessed using both return-based and risk-adjusted metrics, including annual returns, volatility, Sharpe ratio, Sortino ratio and maximum drawdown. The empirical results indicate that the APW model is capable of improving return generation in several periods in trending market environments. The adaptive weighting mechanism allows the model to allocate more capital to signals that have recently performed well, leading to higher returns in certain years compared to the equal-weighted benchmark. However, the results also show that this improvement is not consistent across all the market conditions. During periods of market instability, both portfolios experience similar levels of drawdown and volatility. In terms or risk-adjusted performance, both strategies show limitations, as reflected in the presence of negative Sharpe and Sortino ratios in several years. This suggests that although the APW model can enhance return potential, it does not eliminate the trade-off between return and risk. The analysis further reveals that the adaptive model introduces greater variability in returns. A robustness analysis shows that model performance is sensitive to parameter choices. The choice of lookback window, prediction horizon and weighting scheme. No single configuration is found to be universally optimal. This highlights the importance of balancing adaptability and robustness in portfolio design.