Rasmus Lönn


We use a combination of financial risk factors and sparse hedging portfolios to allocate a large number of assets into a minimum variance portfolio. Regularized hedging portfolios can be formed using the graphical lasso but relies on statistical sparsity assumptions. To motivate these assumptions we make use of common financial risk factors. The estimated hedging portfolios are formed to complement the risk factors by allowing for deviations from the strict factor structure.

Empirically we find that our method reduces portfolio volatility more than competing methods, especially when the number of assets far exceed the amount of historical returns. The strong performance stems from three related sources: (1) the power of parsimonious factor models to explain common risk; (2) the ability of the graphical lasso to perform effective model selection; (3) the benefits of avoiding high-dimensional matrix inversion in the portfolio allocation.