On the Forecasting of Financial Volatility Using Ultra-High Frequency Data

The measurement of the volatility is key in financial markets. This is true not only because decisions are made in an environment of uncertainty, but because sometimes the volatility element overpowers all the remaining aspects in the decision process. Huge movements in the prices of the assets (volatility) can lead to huge losses and also huge gains. There are models to establish the fair prices for certain kind of assets, in that the only parameter that is not directly observable is the parameter characterizing the volatility. However, it is well established in the literature that the evolution of the volatility can be forecasted. Several parametric models have been proposed for modeling the volatility evolution, for example, the Autoregressive Conditional Heteroscedastic (ARCH) and the Stochastic Volatility model (SV). Nowadays, we live in a “Big Data” world, and even for non-professionals of financial markets, it is possible to record data obtained at every second. Recently, measures of volatility have been developed using intraday data, for example, the measure of realized volatility. One of the main aspects to consider is that intraday data and measures of realized volatility are associated with unequal time-spaced observations. In this paper, we compare the forecasts of the volatility evolution using intradaily observations and daily observations, and by trying to conciliate both kind of forecasts, for the data obtained from US and European stock markets, we find out that the use of measures of realized volatility represent an important improvement in volatility forecasting, that can be added to the more well established models that are used in this context, ARCH and SV models.