R help - Aug 2006 - Moving Window regressions with corrections for Heteroscedasticity and Autocorrelations(HAC)

# Using Moving/Rolling Windows, here we do an OLS Regression with corrections
for #Heteroscedasticity and Autocorrelations (HAC) using Newey West Method. This
code is a #extension of Ajay Shah?s code for moving windows simple OLS
regression.
 
# The easiest way to adjust for Autocorrelations and Heteroscedasticity in the
OLS residuals is to #use the coeftest function that is included in the ?lmtest?
packages.
# The ?lmtest? package includes the coeftest (model, NeweyWest) command which
returns # the ?HAC? consistent estimators along with its standard error, t-stat
and pvalue.
 
# To implement the corrections for Heteroscedasticity and Autocorrelations for a
moving/rolling
# window OLS regression, here is the actual function:
#---------------------------------------------------------------------------------------------------------
>movingWindowRegression <- function(data, T, width, model, K) {
  results = matrix(nrow=T, ncol=4*K)
  for (i in width:T) {
        details <-coeftest(lm(model,data[(i-width+1):i,]),NeweyWest)
            n=1;
    for (j in 1:K) {
      results[i,n:(n+3)]   = details[j, 1:4]
      n=n+4
    }
  
  }
  return(results)
}
 
#------------------------------------------------------------------------------------------------------
#STEP1: Install ?lmtest? package using library (lmtest).
 
#STEP 2: Copy and paste the function defined above into the console and submit
it.
 
# The function returns a matrix with T rows (You will find ?NA ?values on all
rows less than width.)
#The columns are organized as :  beta1 se1 tstat1 pvalue1 beta2 se2 tstat2
pvalue2
# i.e. we have each coefficient followed by its s,e., t-stat and p-value. Note
that K is the no of #variables in the regression.
 
#STEP3: Get the results using 
 
> r= movingWindowRegression(CAPM, 240, 60,Y~X1+X2,3)
 
# Here CAPM is the data set, T=240 monthly data, K =60 as we want to find the 5
year #(5*12=60) rolling  regression with adjustment to Autocorrelations and
Heteroscedasticity
 #and the model is Y~X1+X2.
 
# Note that if you want only the ?Heteroscedasticity-Consistent (HC) standard
errors? in the #moving/rolling window framework then install ?sandwich? package
and change the ?NeweyWest? # to ?sandwich? in the ?details? of the function.
 
# Please send your comments/suggestions to
<alok@igidr.ac.in>alok@igidr.ac.in.
# www.igidr.ac.in


	[[alternative HTML version deleted]]