A1655
Title: The portfolio regression approach to mean-variance analysis via the elastic net
Authors: Red Laviste - University of Basel (Switzerland) [presenting]
Dietmar Maringer - University of Basel (Switzerland)
Abstract: The portfolio regression approach encompasses models that estimate mean-variance efficient portfolio weights using multiple linear regression. Being the regression coefficients, estimates of the portfolio weights obtained from the regression approach are equivalent to the more famous portfolio optimization approach, but can be assessed with standard errors and other well-known statistical measures and tests. Since the squared residual loss in least squares regression has a tendency to overfit, regularized regression models such as ridge and lasso are often employed to induce stability and sparsity in portfolio weight estimates. We unify existing portfolio regression models into an elastic net framework that nests the global minimum variance, tangency and frontier portfolios as special cases, and study the optimal calibration of the regularization factor $\lambda$.
