Hi,When I want to compare the intraday volatility for a certain asset between days, I must choose/set a time-interval because the different days haven't got the same (ticker-)length. But the consequence is that I miss a lot of data between the taken points. I was thinking of taking a High-Low-(Open)-Close volatility for the small intervals. P.e. cut the average 3000 ticks into 255 intervals (every two minutes). But then I have relatively less information over the busy part of the day and more during lunchtime. What is a good way to handle the problem?Greetings,Folkert

You need the book High-Frequency Finance by the guys from Olsen (Dacorogna et al). The entire book is about your problem!P

engle wrote a paper called "the econometrics of ultra high frequency data", which may be of use. paper no. 114 at:http://www.econ.ucsd.edu/~mbacci/engle/ ... ecent.html

> P.e. cut the average 3000 ticks into 255 intervals (every two minutes). But then I have > relatively less information over the busy part of the day and more during lunchtime. > What is a good way to handle the problem?A simple (although not very precise) solution would be to use standing bid and ask quotes every two minutes and take the standard deviation of the mid price returns. This gets around the problem of variable trading activity during the day.You can pull these from Bloomberg in variable intervals (starting 1 minute wide) without having to filter through the time stamps yourself.Good luck,Hari

http://dept.econ.yorku.ca/jasj/classes.html has the early verion of "Financial Econometrics: Problems, Models, and Methods" (Nov 01) may be very useful to you. It has sections on high-frequency time-series analysis.

I have no idea what kind of asset you intend to measure. The Olsen Research Group (G. Zumbach et al) together with Zurich Re (M. Dacorongna et al) is at the very top in currencies strategies research. In reference to your question, using tick data to measure volatility tick-by-tick will product a lot of noise. Even so, you gonna use filters the volatility will be quite meaningless. Thus, in effort to eliminate much noise as possible the filters tend to eliminate real signals as well. Therefore, if you will measure the intra-day volatility on very specific time frame, then high frequent data might help. For example, measuring the volatility in 30 min or 60 min intervals may product much reliable actual daily volatility then using just simple daily volatility. David

I agree with the previous answers if you are using the estimated volatility as a signal. But if you actually want to trade on the basis of it, or value something based on theoretical trading, you have to adjust the data to executable prices. It would be a mistake to use a general theoretic technique instead of something tailored to your precise execution potential.To take a simple example, suppose the true underlying price follows a Guassian random walk, but the reported trades randomly add or subtract a fixed bid-ask spread. Most of your short-term volatility will be the difference between the bid and ask prices. Your actual executable price will depend on your trading access to the market, but in no case will it equal your time series. On the other hand, if you are designing a program trading system, you need to know the reporting, computation and execution delays to know what prices are available to you.

On the other hand, if you are designing a program trading system, you need to know the reporting, computation and execution delays to know what prices are available to you. >>I'm in complete agreement with that. I once worked on a project where we were trying to predict stock prices over a 1 to 2 minute horizon using a large time series model, to help option traders "lean" and adjust their deltas. The idea was, the traders would have to buy and sell stock anyway, so we could tell them whether to "wait" or "trade".We couldn't figure out why there were so many jumps in the data until we realized that (e.g.) Cincinnati stock prices were superimposed on NYSE prices and the time stamps were about a minute off. Asynchonous data can be a real bother in intraday analysis.

Thanks for the comments. I'll try to get the suggested literature. The reason I dislike setting my interval too big or middle prices is due to the fear to miss peaks. In my eyes they are the most important for the result of trading. Despite the uncertainty of having been able to trade anyway on that price, I still want it to have influence on the (historical) volatility.