[R-SIG-Finance] Rolling through fixed-length time windows

[Gabor Grothendieck]

On Mon, Nov 7, 2011 at 8:50 AM, Matthew Clegg <matthewcleggphd at gmail.com> wrote:
>
>
> On Fri, Nov 4, 2011 at 9:24 AM, Gabor Grothendieck <ggrothendieck at gmail.com>
> wrote:
>>
>> On Fri, Nov 4, 2011 at 9:09 AM, Matthew Clegg <matthewcleggphd at gmail.com>
>> wrote:
>> > Hello R-Sig-Finance members:
>> >
>> > I was wondering if anyone has contributed functions that are similar
>> > to the zoo roll* functions but which operate on fixed-length time
>> > windows?  For example, suppose I have a zoo-based object consisting
>> > of the daily closing prices of a stock, and I wish to know for each
>> > date, what was the volatility over the succeeding 30 calendar days?
>> > Probably many people would settle for something like:
>> >  rollapply (log(lag(P))-log(P), 21, sd, align="left") * sqrt(252)
>> > (where P is the price series).  However, this is an approximation.
>> > Not all periods of 30 calendar days include precisely 21 trading days.
>> >
>> > This seems like an obvious enough question that I would think that it
>> > has been asked (and answered) many times before, but I could not find
>> > a reference to the recommended solution.
>> >
>> > If no one has tackled this problem before, I might try to put together
>> > a small library of functions that are like roll* but which operate
>> > on fixed time windows.  I am including an example of one such function
>> > below.
>> >
>> > Matthew Clegg
>> >
>> > [snip]
>> >
>>
>> Here is a one liner (two if you count making the result into a zoo
>> object):
>>
>> > z <- zoo(1:25)
>> > zz <- sapply(seq_along(z), function(i) sum(z[time(z) <= time(z)[i] &
>> > time(z) > time(z)[i] - 3]))
>> > zoo(zz, time(z))
>>  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
>>  1  3  6  9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69 72
>>
>> --
>> Statistics & Software Consulting
>> GKX Group, GKX Associates Inc.
>> tel: 1-877-GKX-GROUP
>> email: ggrothendieck at gmail.com
>
>
> Aha!  That's an elegant solution and another great illustration of the
> power of vector processing in R.
>
> I found that after tweaking my code, I could achieve a significant
> improvement in running time over this sapply()-based one liner.  The
> following table compares the running times for various lengths of the
> underlying zoo vector:
>

The rollapply slowdown was reported and fixed in the development
version of zoo already. It only affected recent versions of zoo since
rollapply was rewritten to add certain features. See:

http://r.789695.n4.nabble.com/zoo-performance-regression-noticed-1-6-5-is-faster-tt3990753.html#a3993387

Certainly zoo indexing can be expensive and in those cases that do
involve indexing in an inner loop, replacing zoo object z with zc <-
coredata(z) and tt <- time(z) speeds things up.  Typically that covers
fewer computations than you might think because most R code takes the
whole object approach.

-- 
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com