A1420
Title: A comparison of sequential and integrated estimation of realized volatility models used in portfolio optimization
Authors: Erik-Jan Senn - University of St. Gallen (Switzerland) [presenting]
Abstract: Portfolio optimization relies on precise volatility forecasts. However, in many cases, volatility models are estimated independently of the task-specific problem, leading to forecasts that may not be optimal for the intended application. The purpose is to discuss the simple example of a mean-variance utility investor seeking to maximize expected utility. Two approaches to estimating HAR-type realized volatility models are compared: i) the standard sequential procedure, which first estimates the realized volatility model based on a statistical loss function such as mean squared error, and then applies the plug-in principle to compute optimal portfolio weights and realized utilities; ii) the integrated procedure, which uses the same HAR-type model but estimates the parameters using the realized utility of the mean-variance investor as the loss function. The simulation-based and empirical comparison can guide the choice of loss functions and evaluation metrics for forecast evaluation in this setting.
