A0508
Title: The constrained Dantzig-type estimator: An application to selection of high-dimensional portfolios
Authors: Dechuan Zhu - Nanyang Technology University (Singapore) [presenting]
Chi Seng Pun - Nanyang Technological University (Singapore)
Abstract: Asset selection problems, especially sparse portfolio construction, have been getting uprising attention in recent years due to the vast increase in available assets, paired with comparably limited yet noisy information in the market. Moreover, investment activities are often restricted by various constraints, e.g. budget constraints. A constrained Dantzig-type estimator (CDE) is developed for sparse learning problems with equality constraints, such as sparse portfolio construction. It is shown that CDE is able to produce an estimate of the oracle solution and counter the curse of dimensionality. At the same time, its non-asymptotic statistical error bounds under l1 and l2 norms are derived. Compared to the constrained Lasso, CDE has a significant computational advantage as CDE can be obtained via the solution to a linear problem. Moreover, CDE is extremely versatile and widely applicable. Extensive simulations and empirical studies show that sparse portfolios constructed using CDE have superior out-of-sample performance compared to various benchmark portfolios, including the equally weighted portfolio.
