Estimating Financial Volatility with High-Frequency Returns

Abstract

The primary value of a time series model lies in its ability to provide reliable approximations of the modelled variable, both in-sample (where data are used to estimate model parameters) and out-of-sample (where the model is updated with new information and produces forecasts). In this paper, an overview of the various models in the GARCH family is followed by their application in estimating the daily volatility of Citigroup Inc., a major player in the US subprime mortgage crisis. Fitting these estimates to the ex-post realized volatility measure constructed from high-frequency returns provides superior goodness-of-fit than fitting them to the conventional absolute returns measure. This suggests that when modelling latent financial volatility, information revealed by high-frequency data can greatly enhance GARCH estimates' performance.