Why Exponential Almon Lag Works Well in Econometrics: An Invariance-Based Explanation