The effect of government spending on income and employment is a central unresolved question in macroeconomics. This paper employs a novel identification strategy to isolate exogenous and unexpected variation in state government spending. State governments manage large defined-benefit pension plans for which they bear the investment risk. Using a newly-collected dataset on the returns and portfolios of these plans, I show that the idiosyncratic component of their returns is a strong predictor of subsequent state government spending. Instrumenting with this ‘windfall’ component of returns, I find that state government spending has a large positive effect on income and employment. Baseline estimates indicate that each dollar of spending raises in-state income by 2.12, and that 35,000 of spending generates one additional job. These effects are not due to in-state investment bias, are concentrated in the non-traded sector, and are larger during times of labor force ‘slack.’ Finally, I consider how these results compare with the predictions of a standard macroeconomic model and outline which features in the model are consistent with the empirical findings.