2

I was wondering if someone could guide me through the rollapply function in R. I want to conduct a 1 year rolling regression with fama-french factors. The dataset is prepared in Excel and contains weekly data from 2011-2017. Therefore the time window is set to 52 weeks. I want to compute 1 year rolling beta coefficients for the time period 2012-2017, so the time window will move from [1:52] to [2:53] starting from the first week in 2011. I have several pricing factors, this means that I have to run a multiple linear regression.

This is what I have tried so far:

Rollingregression <- read_excel("C:/Users/neri_/Desktop/Fama French Regression.xlsx", col_names = TRUE, sheet = 1)
Rollingregression <- xts(Rollingregression[,2:7],order.by = Rollingregression[,1])
coeffs <- rollapply(Rollingregression, width=52, FUN = function(Z) coef(lm(Rollingregression$`Excess Return` ~ Rollingregression$OBX + Rollingregression$SMB + Rollingregression$HML + Rollingregression$PR1YR + Rollingregression$LIQ, data = as.data.frame(Z))), by.column=FALSE, align="right")

However, I dont get the 5 years weekly rolling betas, but instead the regression over the full period: 

  (Intercept)           Rollingregression$OBX Rollingregression$SMB
  [1,] 0.0001511391             0.7529416          -0.007408397
  [2,] 0.0001511391             0.7529416          -0.007408397
  [3,] 0.0001511391             0.7529416          -0.007408397
  [4,] 0.0001511391             0.7529416          -0.007408397
  [5,] 0.0001511391             0.7529416          -0.007408397

The output also gives me the same betas for the other fama-french factors throughout the period. 

2011-01-04                                        
2011-01-11                                        
2011-01-18                                        
2011-01-25                                        
2011-02-01

I thought this xts code would make sure that R understood that the dataset is a time-series.

I also tried this: 

coeffs <- (rollapply(zoo(Rollingregression), width=52, FUN = function(Z){t = lm(Rollingregression$`Excess Return` ~ Rollingregression$OBX + Rollingregression$SMB + Rollingregression$HML + Rollingregression$PR1YR + Rollingregression$LIQ, data = as.data.frame(Z), model = TRUE); return(t$coef)}, by.column = FALSE, align = "right"))

Still gives me the same output.

I am fairly new with R, so any help is much appreciated.

Thanks!

micstr
  • 5,080
  • 8
  • 48
  • 76
Neri Kim
  • 135
  • 1
  • 2
  • 9

1 Answers1

1

try replacing function(Z) with function(Rollingregression)...

coeffs <- rollapply(Rollingregression, width=52, FUN = function(Rollingregression) coef(lm(Rollingregression$Excess Return ~ Rollingregression$OBX + Rollingregression$SMB + Rollingregression$HML + Rollingregression$PR1YR + Rollingregression$LIQ, data = as.data.frame(Z))), by.column=FALSE, align="right")

(ref Eric Zivot)

Rod Morley
  • 11
  • 2