Linear regression with Newey-West errors

Question

I have a data frame with 30 columns and want to run linear regressions over all columns. I did this with the lapply() function:

my_lms <- lapply(1:30, function(x) lm(ts[,x] ~ v, data = ts))

Again with the lapply() function I get the summary statistics:

lapply(my_lms, summary)

My question is how to get the same summary statistics using Newey-West errors? I tried:

lapply(my_lms, coeftest, vcov. = NeweyWest)

But this gives me the error message: Error in if (ncol(x) == 1) { : Argument has length 0

Thanks

reproducible example: ts is a time series with index prices and a dummy variable for summer/winter month (the original dataset contains 30 indices and I want to do a linear regression for every index on the dummy variable)

library(zoo)
library(lmtest)
library(sandwich)
ts <- structure(c(4.29528246942341, 6.74996509842245, 3.17792454063952, 
        -1.28018252107232, 3.49874815433011, -2.80982508373651, 3.47182334545917, 
        5.95818170603837, -0.704348332266147, 1.61214679329347, 9.13286793976806, 
        -0.43878198350602, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1),
        index = structure(c(604627200,607305600, 609724800, 612576000, 615168000,
                            617846400, 620524800, 623030400, 625795200, 628387200, 630892800, 633744000),
        tzone = "UTC", tclass = "Date"), class = c("xts", "zoo"), .indexCLASS = "Date", .indexTZ = "UTC",
        tclass = "Date", tzone = "UTC", .Dim = c(12L, 2L), .Dimnames = list(NULL, c("S.P.GSCI.Commodity.Total.Return", "dum")))
reg <- lapply(1:1, function(x) lm(ts[,x] ~ dum, data = ts))
sum <- lapply(reg, summary)
test <- lapply(reg, coeftest, vcov. = NeweyWest)

thanks for your hint! I added an example... hope this helps for answering my question — LuckyLuke
I can't reproduce your problem. It runs without error for me. Test with sandwich_2.5-0, lmtest_0.9-36, and zoo_1.8-3 — MrFlick

ak17 ak17 · Accepted Answer · 2019-04-08T18:09:55

Assuming you are using the coeftest function from the lmtest package. You either have to pass a function that calculates the covariance matrix or the covariance matrix itself as the vcov. argument.

In your question you don't specify what your input NeweyWest to the .vcov argument is.

I believe the vcovHAC function from the sandwich package would be the right choice for Newey-West.

Linear regression with Newey-West errors

1 Answers