Supervised Similarity for Firm Linkages

We introduce characteristic vector linkages (CVLs), a new way to measure how closely firms are related by comparing rich sets of stock characteristics rather than traditional classifications. Using CVLs, we construct equity–equity distances with a simple Euclidean metric and with a supervised similarity model based on quantum cognition machine learning (QCML). Both distance measures support momentum spillover strategies that observe lead–lag effects between economically linked stocks, and QCML-based similarity generates signals that, in our empirical tests, appear stronger and more persistent. These results illustrate how machine-learned linkages may help enhance equity portfolio construction. The findings are provided solely for institutional investors and for informational purposes and are not indicative of future performance.