Create Your First Project
Start adding your projects to your portfolio. Click on "Manage Projects" to get started
Characteristics and Non-Parametric Optimal Portfolio Policies
Subject
Portfolio Allocation
Date
2023
Motivation
Portfolio optimization is forward-looking by nature: it consists of creating positions today for a return achieved in the next period. Predicting future returns is notoriously tricky; however, variables such as characteristics are observable today and likely convey information about future returns. In this paper, I link characteristics to portfolio weights using a non-parametric (and hence general) relationship. It leads to an optimal portfolio strategy that conveniently summarizes the information contained in the characteristics. The strategy is then back-tested using data.
Abstract: We model portfolio weights as non-parametric functions of the asset characteristics. The resulting return decomposition allows the mean-variance optimization to be solved in closed-form. Instead of optimizing over the usual stock portfolio weights, the optimization is done over characteristic-order portfolio weights. The output consists of non-parametric optimal portfolio policies that can be conveniently represented graphically. We test their out-of-sample performance on a large sample of more than 70 characteristics and 21'000 US stocks. The results show that the optimal strategy significantly improves the risk-return trade-off with respect to the value-weighted stock portfolio benchmark. Moreover, the linear specification systematically under-performs with respect to the non-linear ones, suggesting that the latter should not be ignored in a portfolio choice context.