Multistage stochastic linear programs (MSLP) can be approximated by applying linear decision rules (LDR) on the recourse decisions. This reduces MSLP (its dual) into a static problem which provides an upper (lower) bound on the optimal value. We introduce two-stage LDR whose application reduces MSLP (or its dual) into a two-stage stochastic linear program. In addition to yielding better policies and bounds, this approach requires many fewer assumptions than are required to get an explicit reformulation when using the static LDR approach. On a capacity expansion model, we find that the two-stage LDR policy has expected cost 20-34% lower than the static LDR policy, and yields lower bounds that are 0.1-3.3% better.