A Novel way to Diversify Portfolio Weights

Abstract

There is broad agreement among academics and practitioners that, in general, a diversified portfolio delivers more robust out-of-sample performance than a concentrated one. This paper develops a new method to diversify portfolio weights, which does not rely on expected returns or volatilities and is therefore robust to measurement error in these variables. It uses pair-wise regressions to estimate how much variation each asset explains in terms of another asset. The optimization function finds the portfolio weights that maximize the unexplained variation of the portfolio. We benchmark our method against established methods from the literature, such as the minimum volatility and the equal risk contribution portfolios. We demonstrate that our method adheres to established diversification properties and it performs well in empirical tests.