I have daily (financial) time series. As every month does not end in the 30th or 31st, it has created problems for myself using apply functions. Using monthly time series I can just specify every n'th month. However, how can I specify every n'th month with daily series.
Let's say I have this date data set:
dates <- seq(as.Date("2000-01-01"), length = 1000, by = "days")
data <- cbind(rnorm(1000, mean = 0.01, sd = 0.001),
rnorm(1000, mean = 0.01, sd = 0.001),
rnorm(1000, mean = 0.01, sd = 0.001))
Particularly, I want to run covariance matrices for every n'th month similarly to:
cov.fixed <- lapply(as.data.frame(t(rollapply(1:nrow(data), 90, c))),
function(i) cov(data[i,]))
But instead of 90 is it possible to write as every 3rd month such that it takes into account the calendar days?
Many thanks
If you want the covariances every month, you can do something like the following. The key is to split the matrix by month.
library(zoo)
month <- as.yearmon(dates)
cov.month <- lapply(split(as.data.frame(data), month), cov)
names(cov.month) <- sub(" ", "_", names(cov.month))
cov.month$Jan_2000
# V1 V2 V3
#V1 7.825062e-07 7.063689e-08 9.561721e-08
#V2 7.063689e-08 8.989207e-07 1.293351e-07
#V3 9.561721e-08 1.293351e-07 1.175318e-06
cov.month[[1]] # The same
As for quarters, the code is similar, just substitute as.yearqtr for as.yearmon.
quarter <- as.yearqtr(dates)
cov.quarter <- lapply(split(as.data.frame(data), quarter), cov)
Then make better names, without spaces, with sub like above.
Data.
Unlike in the question, I have set the RNG seed.
set.seed(3658) # Make the results reproducible
data <- cbind(rnorm(1000, mean = 0.01, sd = 0.001),
rnorm(1000, mean = 0.01, sd = 0.001),
rnorm(1000, mean = 0.01, sd = 0.001))