R: monthly expanding regression using daily data - r

Question

I want to perform an expanding regression at monthly frequency using daily data. The model is:
ret = \beta_0 + \beta_1 X + \varepsilon
Sample data and my attempt:
library(zoo)
df = data.frame(date = seq(as.Date('2011-01-01'),as.Date('2011-03-31'),by = 1), ret = rnorm(90,0,1), X = rnorm(90,0,1))
roll = function(data, n = now(data) {
rollapplyr(1:n, 1:n, function(x) coef(lm(ret ~ X, data, subset =x))[[2]]
}
output = df %>%
mutate(coefficient = roll(data.frame(ret, X)))
The code above runs expanding regression by row and I could extract just the last value in each month to get the coefficients for that month (i.e., in this example, I only need coefficients estimated on Jan-31, Feb-28 and Mar-31).
However, I need to apply this code to a large dataset, and to save time I only want the regressions to run at the last day of each month in the expanding style (i.e., not run regression every day). I'd appreciate if someone can help point out a way to improve the code here.

Answer 1

Create a function coef2 that given a yearmon object computes the second regression coeffient up to dates in that year and month.
library(zoo)
coef2 <- function(ym, data) {
coef(lm(ret ~ X, data, subset = as.yearmon(date) <= ym))[[2]]
}
yearmonth <- unique(as.yearmon(df$date))
data.frame(yearmonth, slope = sapply(yearmonth, coef2, data = df))
## yearmonth slope
## 1 Jan 2011 0.2208940
## 2 Feb 2011 0.1792896
## 3 Mar 2011 0.1180308
If performance is an issue you could try t his alternative version of coef2 which avoids the use of lm :
coef2 <- function(ym, data) {
with(subset(data, as.yearmon(date) <= ym), cov(ret, X) / var(X))
}