Barro's (1979) tax smoothing hypothesis (TSH) assumes that the government is always subject to an "optimal" degree of discretion in budget policy, i.e., optimal in the sense that the welfare costs from taxation are minimized. This paper proposes a generalization of the TSH that relaxes this crucial assumption. Postwar evidence for Sweden indicates that in contrast to the TSH, the generalized model provides close to a perfect fit: Tax smoothing behavior in combination with more discretion in budget policy relative to what is optimal, can explain all shifts in the central government's budget balance, including the dramatic shifts during the period 1970-96.