By definition, an exponential smoother does not use a finite number of previous values. Instead, it uses a smoothing parameter that specifies how much weight to give to previous values. If th eparameter is alpha, then the weight of the k+th previous observation (k>1) is proportional to (1-alpha)**k. Smoothing parameters close to 1 discount earlier observations, so to get an exponential smoother that mostly relies on the most recent (4) observations, use alpha > 0.5, such as alpha near 0.66 or 0.75.
You can use PROC EXPAND to compute exponential moving averages. See the example in the article "Compute a moving average in SAS." I also wrote an article that describes the mathematical details for rolling statistics, along with manual computations of common rolling statistics.
... View more
