Covariance estimation using high-frequency data: Sensitivities of estimation methods

In this study we examine three widely used realized correlation estimators for natural gas, gasoil, and crude oil futures using data from IntercontinentalExchange (ICE). The objective is to illustrate sensitivities of estimation methods on the resulting realized correlation estimates. The empirical results show that the choice between the various correlation estimators is not at all trivial and depends strongly on the specific features and liquidity of the observed price processes. These findings suggest that great care must be taken when using high-frequency data in portfolio risk applications.