HAC function

HAC Covariance Matrix Estimation HAC computes the central quantity (the meat) in the HAC covariance matrix estimator, also called sandwich estimator. HAC is the abbreviation for "heteroskedasticity and autocorrelation consistent".

Source

Heberle, J. and Sattarhoff, C. (2017) doi:10.3390/econometrics5010009 "A Fast Algorithm for the Computation of HAC Covariance Matrix Estimators"

HAC(mcond, method = "Bartlett", bw)

Arguments

• mcond: a q-dimensional multivariate time series. In the case of OLS regression with q regressors mcond contains the series of the form regressor*residual (see example below).
• method: kernel function, choose between "Truncated", "Bartlett", "Parzen", "Tukey-Hanning", "Quadratic Spectral".
• bw: bandwidth parameter, controls the number of lags considered in the estimation.

Returns

mat a (q,q)-matrix

Examples

data(MUSKRAT)
y <- ts(log10(MUSKRAT))
n <- length(y)
t <- c(1:n)
t2 <- t^2
out2 <- lm(y ~ t +t2)
mat_xu <- matrix(c(out2$residuals,t*out2$residuals, t2*out2$residuals),nrow=62,ncol=3)
hac <- HAC(mat_xu, method="Bartlett", 4)

mat_regr<- matrix(c(rep(1,62),t,t2),nrow=62,ncol=3)
mat_q <- t(mat_regr)%*%mat_regr/62
vcov_HAC <- solve(mat_q)%*%hac%*%solve(mat_q)/62
# vcov_HAC is the HAC covariance matrix estimation for the OLS coefficients.