Itraday returns/ time series creation

Hi,

I work with the below sample of (intraday) high freequency stock data:

sample:

STOCK DATE TIME PRICE VOLUME

ETE 22082011 11301514 05.21 500

ETE 22082011 11301713 06.51 1000

ETE 22082011 11311517 07.11 500

ETE 22082011 11321514 06.31 1000

I need to create new time series of stock returns (defined as Rt= ln(Pt/Pt-1)) for multiple time scales, i.e 3-min returns, 5-min returns,..., up to 180-min returns.

However the totalsize of the data is huge ( almost 9 mil observations).

Could someone suggest a time-efficient method of create these time series.

Many Thanks in advance

Panos,

PS: The column TIME indicates hour/minute/seconds/millisecs with observations being even at the same millisecond

Answer 1

[Oleg Komarov]

Before any suggestion:

• Observations with the same time-stamps are usually aggregated using the median.
• Which price do you want to take for the sampling grid? First, last, max, min, previous, next, linear interp.?

See Brownlees & Gallo 2006, Computational Statistics & Data Analysis.

I suppose you need daily grids and you're sampling in calendar time (as opposed to tick-time sampling)?

If you're trying to calculated Realised Variance measures: http://www.kevinsheppard.com/wiki/MFE_Toolbox

This toolbox takes care of most of the aspects.

[Pap]

Thanks Oleg

. I need the next price (or linear interp) for the sample.

. My intention is to create a time series with continuous observations over different days.

any suggestions?

Thanks again

[Oleg Komarov]

I am working on my own function but in the meantime you can check realized_price_filter from the toolbox link I posted.You can decide the frequency but the previous price interploation is used, a very common choice in literature.