roll_lm checks nrow(x) > width rather than nrow(x) > min_obs

[rrichmond]

Right now roll_lm throws the error when you try to run a rolling regression on data that is shorter than width:

Error in eval(substitute(expr), envir, enclos) : 
  value of 'width' must be between one and number of rows in 'x' and 'y'

It seems more appropriate to check the nrow(x) and nrow(y) are just greater than min_obs. Here's an example:

library(roll)
n_vars <- 10
n_obs <- 50
x <- matrix(rnorm(n_obs * n_vars), nrow = n_obs, ncol = n_vars)
y <- matrix(rnorm(n_obs), nrow = n_obs, ncol = 1)
# Rolling regressions
result <- roll_lm(x, y, width=100, min_obs = 10)

[jasonjfoster]

Thanks for the feedback. Fyi, I typically use the min_obs argument when there are many NA’s in a data set, yet I’m comfortable with a certain threshold of missing values. For example, rolling regressions with a 252-day window and min_obs = 63 days, i.e. a quarter. My sense is that most users don’t change the default min_obs = width argument though.

Anyway, in your example, I could just use width = min(100, nrow(x)) instead of width = 100 to satisfy the condition; however, I understand the latter is easier to write and, if a user has determined a reasonable min_obs threshold, then he/she should understand the concept of an expanding window and wants to get the desired result.

My concern is when users, who don’t change min_obs, see an error message about min_obs. Something like the following message could be confusing if the default arguments remain untouched: value of 'min_obs' must be between one and number of rows in 'x' and 'y'

In general, maybe a better solution is dependent on whether the min_obs is changed from the default or not. That is, if a user does not specify min_obs, then the error message stays the same as it is currently, otherwise a min_obs specific error message is returned like above. And then you can get the result, like in your example, if satisfied. What do you think?

[jasonjfoster]

I have simplified checks for the width and min_obs arguments in the development version:

# install.packages("devtools")
devtools::install_github("jjf234/roll")

Specifically, if the min_obs argument is different than the width argument, then there is a check whether the value of min_obs is between one and the minimum of either the number of rows in data or width.