The simplest method to model seasonal patterns is to use dummy variables to represent the seasons. With such a short time series, a weekly periodicity is about the longest effect you can detect (i.e., you would not be able to see a monthly effect with only 4 months). A regresion model with the seasonal dummy variables as your inputs will serve you, but if there is any residual autocorrelation remaining, your MSE may seriously underestimate the variance of the error terms, so your p-values could be way off.
I'd use proc autoreg with the dummy variables and ask for the Durbin-Watson statistic to check for autocorrelation. Proc autoreg can handle autoregressive errors, if present.