Hi, I am performing a rolling regression (i.e., I am using a moving window). I have two variables, the y (called ntl) and x (called tirs20). Here is the code:

block.data = read.csv ("path/block.data.psf.csv")

eq = ntl ~ tirs20

model = roll_regress(eq, block.data, width = 5, do_downdates = TRUE, do_compute = "r.squareds")

coeff = as.data.frame(model$coefs)

The function's output is the model's coefficients (intercept and slope). I was wondering if there is a way to add the residuals of the regression as a new column to the coeff data frame.

If necessary, I can share my csv.

I encounter this error when I run roll_regres. What does this mean? There is no NA in my data.

The functionality of updating in blocks is very usefull in asset pricing, e.g., rolling regressions of stock returns on market returns. Sometimes we need to do this using the daily returns of last one month/year, and then an fatal error comes. Replication code (in the comparisions vignette):

library(rollRegres)
set.seed(96235555)
n <- 10L * 12L * 21L # x years w/ 12 months of 21 trading days
month <- (seq_len(n) - 1L) %/% 21L + 1L # group by months
set.seed(29478439)
X <- matrix(rnorm(n * 4L), ncol = 4L)
y <- rnorm(n)

randomly drop rows

keep <- seq_along(y) %in% sample.int(nrow(X), as.integer(n * .5))
X <- X [keep, ]
y <- y [keep]
month <- month[keep]

width=2 works well

pck_out <- roll_regres.fit(X, y, width = 2L, grp = month, min_obs = 10L * 2L, do_downdates = TRUE)

while width=1 incurs an fatal error

pck_out <- roll_regres.fit(X, y, width = 1L, grp = month, min_obs = 10L * 1L, do_downdates = TRUE)

Thank you!

would you consider adding support for weights like lm does?
i believe that lm uses a weighted least squares. i.e. minimizing sum(w*e^2), when given an input vector of weights

I noticed this package is removed from CRAN. I have used it a lot, is that possible to publish in CRAN again? Thanks.

Forgot to pass do_downdates here

Hello. Thank you for your contribution.

I am using this package very usefully. I have some issues.

I want to calculate the most recent residual of a dataset, during operating rolling windows.

Is there any method? I know do.compute calculate one period ahead prediction error , R2 and

sigma. Is there any way to derive residuals?

Is it possible to calculate the root mean squared residuals on a cumulative (do_downdates = F) rolling basis using this library? The only idea ive come up with so far is to compute full prior regression line for each data point, then compute the a rolling mean on the residuals, but that seems I have to re-create the cumulative rolling loops and mean function all over again

R hard-crashes when I attempt to run the following code using the latest release of roll_regress. Note that groups 11 and 12 are skipped and the width is 12.

ehbSample <- structure(list(
    group_ = c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 5L, 5L, 5L, 5L, 6L, 6L, 6L, 6L, 7L, 7L, 7L, 7L, 8L, 8L, 8L, 8L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L, 13L), 
    y = c(7, 0, -1, 1, 3, 13, 9, 4, 32, -15, 28, -23, -40, -15, 57, 7, -24, -4, 10, -6, 7, 10, 16, -2, 16, -10, 4, -6, 5, 1, -2, 2, 4, 10, -28, -11, 5, -8, 17, -9, -4),
    x = c(0, -6, 0, -1, -13, 3, 13, 9, -17, 32, -15, 28, -10, 12, -15, 57, -9, -24, -4, 10, -1, 7, 10, 16, -1, 16, -10, 4, 8, 8, 1, -2, -1, 4, 10, -28, 3, 5, -8, 17, 1)), row.names = c(NA, -41L), 
    groups = structure(list(
      group_ = c(1L, 2L, 3L, 4L, 5L, 6L, 7L, 8L, 9L, 10L, 13L), 
      .rows = structure(
        list(1:4, 5:8, 9:12, 13:16, 17:20, 21:24, 25:28, 29:32, 33:36, 37:40, 41L), ptype = integer(0), 
             class = c("vctrs_list_of", "vctrs_vctr", "list"))), 
      row.names = c(NA, -11L), class = c("tbl_df", "tbl", "data.frame"), .drop = TRUE), 
    class = c("grouped_df", "tbl_df", "tbl", "data.frame"))

fit <- roll_regres(
    y ~ x,
    ehbSample,
    width = 12,
    grp = ehbSample$group_,
    do_downdates = TRUE)

The manual pages states the following about the width argument:

Only used if do_downdates == True.

But it is the minimum number of observations in expanding window regressions.

The code crashes if it enters this if block

A simple example is

library(rollRegres)
x <- matrix(rnorm(4 * 20), ncol = 4)
y <- rnorm(20)
roll_regres.fit(x, y, width = 11L)

which crash while this does not

library(rollRegres)
x <- matrix(rnorm(4 * 20), ncol = 4)
y <- rnorm(20)
roll_regres.fit(x, y, width = 12L)

Maybe reconsider how low or high this should be

Add standard errors for the coefficients as a possible output. The Choleksy decompaction of the inner product of the covariance matrix is computed so this should be straight forward to implement.