PROC EXPAND can currently create rolling sums, and rolling sums-of-squares (e.g. rolling sum of X**2 and Y**2)
If it could also generate rolling sums-of-cross-products (i.e. rolling sum of X*Y), then one could generate rolling sums-of-squares-and-cross-products, which could then be fed into PROC REG for rolling-window regressions.