Portfolio allocation across variance risk premia

Abstract

PurposeIn asset management, what if clients want to purchase protection from risk factors, under the form of variance risk premia. This paper aims to address this topic by developing a portfolio optimization framework based on the criterion of the minimum variance risk premium (VRP) for any investor selecting stocks with an expected target return while minimizing the risk aversion associated to the portfolio according to “good” and “bad” times.Design/methodology/approachTo accomplish this portfolio selection problem, the authors compute variance risk-premium as the difference from high-frequencies' realized volatility and options' implied volatility stemming from 19 stock markets, estimate a 2-state Markov-switching model on the variance risk-premia and optimize variance risk-premia portfolios across non-overlapping regions. The period goes from March 16, 2011, to March 28, 2018.FindingsThe authors find that optimized portfolios based on variance-covariance matrices stemming from VRP do not consistently outperform the benchmark based on daily returns. Several robustness checks are investigated by minimizing historical, realized or implicit variances, with/without regime switching. In a boundary case, accounting for the realized variance risk factor in portfolio decisions can be seen as a promising alternative from a portfolio performance perspective.Practical implicationsAs a new management “style”, the realized volatility approach can, therefore, bring incremental value to construct the conditional covariance matrix estimates.Originality/valueThe authors assess the portfolio performance determined by the variance-covariance matrices that are derived by four models: “naive” (Markowitz returns benchmark), non-switching VRP, maximum likelihood regime-switching VRP and Bayesian regime switching VRP. The authors examine the best return-risk combination through the calculation of the Sharpe ratio. They also assess another different portfolio strategy: the risk parity approach.