Aggregate time series from weeks to month

Question

Assume a timeseries ts with frequency(ts) = 52:

Time Series:
Start = c(2010, 34) 
End = c(2013, 25) 
Frequency = 52 
...

I want to aggregate ts such that frequency(ts) = 12.

If the new frequency divides the old one, one can use

aggregate(ts, nfrequency = k, FUN = sum)

But if old frequency mod new frequency > 0 it does not work:

> ts <- aggregate(ts, nfrequency = 12, FUN = sum)
Error in aggregate.ts(ts, nfrequency = 12, FUN = sum) : 
cannot change frequency from 52 to 12

I can a imagine to loop through the time series and decide for each value if to add it completely or a fraction of it, but I am surely not the first one with that requirement. So is there any predefined function doin that?

Answer 1

Yes there are predefined function doing aggregation by month or more generally by any period of time.You can use xts package is very handy for time series manipulations.

For example , you can use apply.monthly :

library(xts)
ts.month <- apply.monthly(as.xts(ts),FUN=sum)

Answer 2

As mentioend, if you do not mind loading another package, I recommend using the aggregate function in zoo or similar functions in xts . You will probably need these packages anyway, especially if you run into ragged timeseries data. Here are examples of aggregating to a monthly timeseries using zoo followed by xts :

zoo:

monthlyTS <- aggregate(dailyTS, as.yearmon, sum)

xts:

monthlyTS <- apply.monthly(xts(dailyTS), mean)

Pure syntactic sugar! What's really nice about xts is that you can aggregate by week. However, it's trivial to add new aggregation functions to the zoo function too.

Cheers,

Adam