Forecasting VIX: the illusion of forecast evaluation criteria

The paper uses daily realised volatility measures in order to gain forecast accuracy over stocks' market implied volatility, as proxied by VIX Index, for forecast horizon of 1, 5, 10 and 22 days ahead. We evaluate forecast accuracy by incorporating a traditional statistical loss function, along with an objective-based evaluation criterion, that is the cumulative returns earned from the different HAR-type volatility models, through a simple yet effective trading exercise on VIX futures. Findings illustrate how illusive the choice between the two metrics may be, as it ends in two contradicting results.