ABSTRACT:

Financial markets and modern trading systems utilize thousands of different indicators to cap- ture all potential market signals across multiple dimensions. However, high-dimensionality in- troduces substantial redundancy and noise, necessitating robust dimensionality reduction (DR) techniques. Traditional linear dimensionality reduction methods struggle to capture nonlin- ear behaviors, such as superposition and entanglement, leading to poorly represented data and loss of information. Indicators are dependent on one another, as well as follow unstable fac- tors like impulse, momentum, and sentiment interactions. This paper introduces a quantum- inspired nonlinear approach to compress these trading indicators while taking into account their complex interdependence. This work will make two key contributions to the emerging bridge between machine learning and quantitative finance. We establish performance baselines for quantum-inspired DR techniques in financial forecasting, demonstrating their viability rel- ative to classical approaches. And we present a framework to reduce overfitting, making it applicable to generalized dimensionality reduction research. Our empirical results demonstrate modest but statistically significant predictive performance across all methods. We document the limitations of current methods, particularly their inability to adapt to structural breaks, and we identify clear directions for future extensions, including Bayesian structural break detec- tion, ensemble-based DR pipelines, and interpretable tools (SHAP, LIME) for inexperienced users.