[R-SIG-Finance] Correlation between two asynchronous time series?

[Patrick Burns]

 From an offline message, it appears that a bit of explanation
might be in order for those that don't understand what we
are talking about.

Suppose we want to know the correlation between
Ford and GM stock prices (an example that may well
become obsolete).  We just get daily returns for each
of the stocks and compute their correlation.  No problem.

Now suppose we have daily data for the S&P500 and
the Nikkei.  We can compute a correlation from this
data -- no problem.  Except that the estimate we get
doesn't have the meaning we want it to have (it will tend
to be too close to zero).  In this example lining up the
dates does not line up the time of the data.

There are three possible strategies (as far as I can see):

* Move to lower frequency -- weekly seems to get rid
of the majority of the asynchrony effect while not being
excessively low frequency.

* Stay with daily data but model it.

* Move to intraday data.  This works for Europe versus
US data because there is overlap in the times that the
markets are open.  It does not work for Japan with a
lot of markets because there is no overlap.


Pat


Patrick Burns wrote:
> When I was playing with that, it looked like
> 5 days (i.e., weekly) was about right.  There
> seemed to be some asynchrony visible (in
> equities) at 3 days.
>
> Pat
>
>
> Nicolas Mougeot wrote:
>> one simple method used in practice (and the basis for correlation 
>> swap for example) is to use 3 series of correlation using 3-day 
>> returns (or even 5 or 20)
>>
>>
>>
>>
>> Adrian Trapletti <a.trapletti at swissonline.ch> Sent by: 
>> r-sig-finance-bounces at stat.math.ethz.ch
>> 11/14/2008 03:07 PM
>>
>> To
>> R-Finance <r-sig-finance at stat.math.ethz.ch>
>> cc
>>
>> Subject
>> Re: [R-SIG-Finance] Correlation between two asynchronous time   series?
>>
>>
>>
>>
>>
>>
>>  
>>> Message: 4
>>> Date: Thu, 13 Nov 2008 21:53:14 -0800
>>> From: Michael <comtech.usa at gmail.com>
>>> Subject: [R-SIG-Finance] correlation between two asynchronous time
>>>                series?
>>> To: r-sig-finance at stat.math.ethz.ch
>>> Message-ID:
>>> <b1f16d9d0811132153qd277378ka3d29b9d30a5d753 at mail.gmail.com>
>>> Content-Type: text/plain; charset=ISO-8859-1
>>>
>>> Hi all,
>>>
>>> If I want to find out the correlation between two time series, one is
>>> the Hong Kong stock index, the other is the S&P 500. The two markets
>>> open at two different time.
>>>
>>> What impact it might have on my estimate of the correlation of the two
>>> series?     
>> Big impact.
>>  
>>> How do I address this asynchronous time series problem? Any
>>> good models?
>>>
>>>     
>> Either you use something along the lines of 
>> http://papers.ssrn.com/sol3/papers.cfm?abstract_id=332901or better 
>> you use high frequency data (e.g., the S&P 500 futures is open almost 
>> 24 hours and you will get overlapping hours with the Asian markets).
>>  
>>> Thanks a lot!
>>>
>>>
>>>     
>> Best regards
>> Adrian
>>
>>   
>