Realised variance forecasting under Box-Cox transformations

Abstract

This paper assesses the benefits of modeling Box-Cox transformed realised variance data. In particular, it examines the quality of realised variance forecasts with and without this transformation applied in an out-of-sample forecasting competition. Using various realised variance measures, data transformations, volatility models and assessment methods, and controlling for data mining issues, the results indicate that data transformations can be economically and statistically significant. Moreover, the quartic root transformation appears to be the most effective in this regard. The conditions under which the use of transformed data is effective are identified.