A1540
Title: Tensor portfolio
Authors: Tianyan Tu - University of California, Riverside (United States) [presenting]
Tae-hwy Lee - UC Riverside (United States)
Abstract: Motivated by the multidimensional nature of financial data, Tensor Portfolios is a framework exploiting the intrinsic multiway structure of stock returns to reduce the number of free parameters required for portfolio construction. Three distinct methods tailored to specific structural assumptions are developed. Tensor and vector Portfolios are systematically compared through Monte Carlo simulations and empirical studies. Simulation results show Tensor Portfolios yield significantly higher out-of-sample Sharpe ratios whenever the data exhibits a Tensor structure. Empirical analysis further corroborates the effectiveness of Tensor Portfolios; their general outperformance over vector Portfolios in real-world markets highlights the practical significance of exploiting multiway information.
