[R-SIG-Finance] Correlation on Tick Data

[Eric Zivot]

Computing correlation on tick data is very tricky. You should consult the realized covariance literature (e.g. papers by Neil shephard and his co-authors) for an introduction to the issues of getting sensible correlations from tick data. Many issues are involved such as aligning data to a common time clock, how often to sample (how many ticks to use), stale quotes for some assets, how to deal with microstructure noise (bid ask bounce etc), etc. just computing an naive correlation on tick data is not a good idea.

****************************************************************
*  Eric Zivot                  			               *
*  Professor and Gary Waterman Distinguished Scholar           *
*  Department of Economics                                     *
*  Box 353330                  email:  ezivot at u.washington.edu *
*  University of Washington    phone:  206-543-6715            *
*  Seattle, WA 98195-3330                                      *                                                           *
*  www:  http://faculty.washington.edu/ezivot                  *
****************************************************************

On Tue, 22 Jul 2008, Kenneth Spriggs wrote:

> Neil,
>
> I don't quite understand your data set.
>
>> I performed this on tick data at different tick time frequencies. For
> instance I can run a whole day's tick updates(quotes) at an >interval of
> 50,000 ticks.
>
> What does this mean exactly?
>
> You're doing cor(YM, ES, method = "whatever")  I assume.  So ES and YM have
> to have the same number of points.  If you take the returns from 50,000
> ticks this could be drastically different time frames.  Do you see my
> concern?
>
>
>
> On Tue, Jul 22, 2008 at 10:38 AM, <markleeds at verizon.net> wrote:
>
>>  just to elaborate a bit more on what matt said.
>>
>> you need to make sure you time series are stationary before you correlate
>> them.  usually one does this by using a unit root test but, in your case,
>> since you are dealing with currencies, as long as you are dealing with the
>> returns streamsand not the prices themselves, there's really no need to use
>> the unit root test.  returns should be stationary ( in general ).   if
>> you're dealing with prices, then correlations don't make sense because
>> prices aren't ( in general ),
>> or atleast i've never seen prices that were.
>>
>>
>>
>>
>> On Tue, Jul 22, 2008 at 10:14 AM, Matthieu Stigler wrote:
>>
>>  Hello
>>>
>>> If ES and YM are time series, you maybe should first test for
>>> auto-correlation of the series. High auto-correlated series can lead to the
>>> phenomen called as spurious regression, and then the correlation coefficient
>>> is "too high".
>>>
>>> Hope this helps
>>>
>>> Mat
>>>
>>> Neil Gupta a écrit :
>>>
>>>> Hello R users.
>>>>
>>>> I was using R to calculate correlation of midquote returns on ES and YM.
>>>> ES
>>>> and YM are highly correlated at close to .97. However when I run the
>>>> correlation on the MQ returns the correlation is close to 0. Should I be
>>>> expecting this or am I doing something wrong? Others have told me this
>>>> should happen, but I do not understand why. If anyone can please explain
>>>> I
>>>> would really appreciate.
>>>>
>>>> Many Thanks,
>>>> Neil
>>>>
>>>>        [[alternative HTML version deleted]]
>>>>
>>>> _______________________________________________
>>>> R-SIG-Finance at stat.math.ethz.ch mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>>>> -- Subscriber-posting only.
>>>> -- If you want to post, subscribe first.
>>>>
>>>>
>>> _______________________________________________
>>> R-SIG-Finance at stat.math.ethz.ch mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>>> -- Subscriber-posting only.
>>> -- If you want to post, subscribe first.
>>>
>>
>> _______________________________________________
>> R-SIG-Finance at stat.math.ethz.ch mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>> -- Subscriber-posting only.
>> -- If you want to post, subscribe first.
>>
>
> 	[[alternative HTML version deleted]]
>
>