Analysing High Frequency Data Using ARCH and GARCH Methods

Abstract

High frequency data is a recent entrant to the world of statistics as they relate to the markets. With tick by tick data we get to see the microstructure of the markets and often are better able to see how they vary from the traditional portrayal. Traditional tools used to look at daily and weekly volatilities are not often very useful in timescales of seconds and minutes. In this paper we try to look at two of the most highly traded stocks in the Indian stock market. The large and small errors tend to cluster together, and thus autoregressive conditional heteroscedasticity models are introduced. First we look at ARCH models on tick by tick data of SBI. Then we look at the GARCH models – with two stocks SBI and TATA – and its variants such as PGARCH and EGARCH to try to see if we can predict the conditional variance. We also glance at the DCC GARCH model to see if a bivariate view gives us any new insights. Finally we try to sum up the various techniques by evaluating them according to their utility in estimating high frequency data.