In time series, the covariance between a random variable and its lagged values. Since it depends on the units of measurement of the underlying variable, it is often more convenient to divide it by the variance and use the resulting autocorrelation coefficient instead.