I have time series with outliers, and I want to get a robust estimate of the trend and its t-value. For that I use lmrob, which works fine in most cases. However, in some cases, it returns -inf for the t-value (due to an error which is 0), and I get the warning
Warning message:
In lf.cov(init, x = x) :
.vcov.avar1: negative diag(<vcov>) fixed up; consider 'cov=".vcov.w."' instead
Reproducible example:
ts_list <- c(563, 522, 535, 728, 683, 624, 595 ,589, 535, 529, 441, 565)
res <- lmrob(ts_list~1 + seq(1:12), type='MM')
summary(res)
Curious to know what is going on here. Thanks!
Regards,