One of the downsides of the autocorrelation function is that it does not remove the effect of lags 1 up to k-1 on the series when calculating the correlation of the series with the k lag. The partial autocorrelation function (PACF), the sister function of the ACF, provides a solution for this by computing the conditional correlation of the series with the k lag given the relationship of the 1, 2, ..., and k-1 lags with the series. In other words, the PACF provides an estimation for the direct correlation of the series with the k lag after removing the correlation of the k lag with the previous lags. The pacf function from the stats package provides an estimation for the PACF values for a given input. Let's review the PACF output for the first 60 lags of the USgas dataset:
pacf(USgas, lag.max = 60)
We will get the following plot:
Both...
