This paper derives the Ramsey optimal fiscal policy for taxing asset income in a model where government expenditure is a function of net output or the inputs that produce it. Extending work by Kenneth L. Judd, I demonstrate that the canonical result that the optimal tax on capital income is zero in the medium to long term is a special case of a more general model. Employing a vector error correction model to estimate the relationship between government consumption and net output or the factor inputs that generate it for the United States between 1948Q1 and 2015Q4, I demonstrate that this special case is empirically implausible, and show how a cointegrating vector can be used to determine the optimal tax schedule. I simulate a version of the model using the empirical estimates to measure the welfare implications of changing the tax rate on asset income, and contrast these results with those generated in a version of the model where government consumption is purely exogenous. The shifting pattern of welfare measurements confirms the theoretical results. I calculate that the prevailing effective tax rate on net asset income in the United States between 1970 and 2014 averaged 0.449. Hence abolishing the tax completely does generate welfare improvements, though only by the equivalent of between 1.103% and 1.616% permanent increase in consumption—well under half the implied welfare benefit when the endogeneity of the government consumption is ignored. The maximum welfare improvement from shifting part of the burden of tax from capital to labor is the equivalent of a permanent increase in consumption of between only 1.491% and 1.858%, and is attained when the tax rate on asset income is lowered to between 0.148 and 0.186. Allowing the tax rate to vary over time raises the maximum welfare benefit to 1.865%. All the results are very robust to a wide range of elasticities of labor supply. (JEL E62, H21, H50)