R型线性回归中roll_lm与lm的差异

Question

我想做一个滚动线性回归，我找到了函数roll_lm，但它提供了与函数lm不同的结果。它们不应该产生相同的结果吗？

# creating dataset
set.seed(123)
ABC <- sample(seq(from = 20, to = 50, by = 5), size = 50, replace = TRUE)
DCE <- sample(seq(from = 20, to = 50, by = 5), size = 50, replace = TRUE)

# Calculating rolling correlation and using last coeff
library(roll)
Rolling.Correl <- roll_lm(ABC , DCE, 50)
last(Rolling.Correl$coefficients[,2])
# [1] -0.233245

# Calculating basic regression using lm
Trad.Rolling.Correl <- lm(ABC ~ DCE)
Trad.Rolling.Correl

# Call:
# lm(formula = ABC ~ DCE)
#
# Coefficients:
# (Intercept)          DCE  
#     41.9204      -0.2112  

在这个特定的例子中，我一只手得-0.233245，另一只手得-0.2112。

Answer 1

?roll_lm说是roll_lm(x, y,...)，所以你需要将它与lm(DCE ~ ABC)进行比较

lm(formula = DCE ~ ABC)

# Coefficients:
# (Intercept)          ABC  
#     43.6236      -0.2332