This link has been bookmarked by 1 people . It was first bookmarked on 06 Aug 2008, by Jeremy Gollehon.
Is High Frequency Data the Answer?
One route has been to use longer samples and higher frequency sampling. If returns are normal and parameters do not change across time, then the number of data points used for estimation and the frequency of those data points are not very important. Reasonable standards, like the common practice of using 36 monthly observations for performance evaluation and 60 for risk matrices, will generally produce reasonable estimates. Departures from normality and changing parameters, by contrast, can significantly increase the amount of data required for reliable inference.
High-frequency data can help to flesh out the dynamic structure of volatility. For example, there may be little change in prices over a day on a close-to-close basis when, in fact, significant movements took place intra-day. Interestingly, it is often not the case that going all the way to tick-by-tick data provides the best information: successive quotes frequently bounce between bid and ask prices, and sampling these as proxies for the mid-price gives an upward bias to measured volatility. In some very interesting recent work, Diebold, Brandt, and Alizadeh2 demonstrate that the daily range (high minus low price) captures a great deal of the information in high-frequency price movements useful for modeling volatility, without much bias, and at a significantly lower cost than estimates based on tick data.
Would you like to comment?
Join Diigo for a free account, or sign in if you are already a member.