Is there an ideal in-sample length for forecasting volatility?

There is limited research carried out to date in the academic literature addressing the issue of the ideal in-sample size when forecasting volatility. This paper therefore considers how much data is required in order to produce accurate forecasts. Broadly speaking, two views exist between practitioners/investors who typically prefer a small in-sample to minimise data holding requirements and researchers/academics who typically chose large in-sample periods. Using a process of expanding window regressions where the in-sample start period expands (backward recursion) we conduct forecasts over twenty-three international markets, including both developed and emerging. Our findings, which demonstrate a degree of homogeneity, show that for the majority of the markets large in-sample periods are not necessary in order to produce the most accurate forecasts supporting the practitioners'/investors' view.