Correspondence: Roberto Ferulano, Department of Economics, Finance and Statistics, University of Perugia, via Pascoli, 20-06123, Perugia, Italy. E-mail: ferulano@stat.unipg.it

¹is a quantitative analyst at P&G SGR S.p.a., based in Rome, Italy. Mr Roberto Ferulano holds a degree in Electronic Engineering and a PhD in 'Mathematical and statistical methods for economic and social sciences' from the University of Perugia, Italy.

Received 16 October 2008; Revised 16 October 2008.

Abstract

This study presents a new methodology for forecasting volatility. It relies on a weighted mean of short and long estimates of variance, based on a Moving Average framework. The quality of the predictions obtained with the proposed formula was checked with both simulated and real data. When applied to the analysis of simulated data, the new formula provides the least reliable forecast when a Random Walk is used as Data Generating Process (DGP) and the forecast variance is a simple Moving Average. This is also the case when the DGP belongs to the ARCH model family and the associated forecast formula is used. However, compared to existing approaches, the new methodology allows for the most reliable forecast on 5-day and 20-day horizons, when it is applied to Index, Fixed Income and Foreign Exchange data series.