R help - Oct 2012 - ccf(x,y) vs. cor() of x and lagged values of y

Hi

 

I'm computing the correlation between two time-series x_t and y_t-1
(time-series lagged using the lag(y,-1) function) using the cor() function
and the returned value is different from the value of ccf() function at the
same lag.  Any ideas why this is so?

 

Thanks in advance for any hints.

Mihnea

 

 


	[[alternative HTML version deleted]]

On Fri, Oct 12, 2012 at 12:28 AM, Mihnea Constantinescu
<mihnea.const at gmail.com> wrote:
> Hi
>
>
>
> I'm computing the correlation between two time-series x_t and y_t-1
> (time-series lagged using the lag(y,-1) function) using the cor() function
> and the returned value is different from the value of ccf() function at the
> same lag.  Any ideas why this is so?
I'd bet that you'd also find that

cor(x,y) = cor(x, lag(y,-1))

? That's because the effect of "lagging" seems to be discarded by
cor() which only looks at the data itself and doesn't align the
observations. The ccf() result is the one you want.

Cheers,
Michael

Hi Mihnea,

I'm afraid I don't have much time to look at this right now -- I'm
sort of swamped for the next 5 days -- so I'm forwarding back to the
list so more eyeballs pass over it. In general, you should always cc
the list so the archives remain complete and so that you get a quicker
response.

Michael

On Sun, Oct 14, 2012 at 2:02 PM, Mihnea Constantinescu
<mihnea.const at gmail.com> wrote:
> Hi Michael
>
> Thank you for the reply. I forgot to mention that I cbind() after applying
> the lag() function and then ran the cor() on the new data object. I had
> noticed too that lag() passed inside of the cor() is useless.
>
> I'm getting somehow closer to troubleshooting the observed
difference...
>
> Have a nice day
> Mihnea
>
> -----Original Message-----
> From: R. Michael Weylandt [mailto:michael.weylandt at gmail.com]
> Sent: Saturday, October 13, 2012 12:31 PM
> To: Mihnea Constantinescu
> Cc: r-help at r-project.org
> Subject: Re: [R] ccf(x,y) vs. cor() of x and lagged values of y
>
> On Fri, Oct 12, 2012 at 12:28 AM, Mihnea Constantinescu
> <mihnea.const at gmail.com> wrote:
>> Hi
>>
>>
>>
>> I'm computing the correlation between two time-series x_t and y_t-1
>> (time-series lagged using the lag(y,-1) function) using the cor()
>> function and the returned value is different from the value of ccf()
>> function at the same lag.  Any ideas why this is so?
>
> I'd bet that you'd also find that
>
> cor(x,y) = cor(x, lag(y,-1))
>
> ? That's because the effect of "lagging" seems to be
discarded by
> cor() which only looks at the data itself and doesn't align the
> observations. The ccf() result is the one you want.
>
> Cheers,
> Michael
>