This paper argues that the following ingredients are necessary in order to draw causal inference from historical correlations: explication of the real-world mechanism that is supposed to have produced a proposed structural relation; the existence of a natural experiment through which the data were generated; and extensive statistical corroboration of the econometric model. This research strategy is employed in an effort to identify the sources of disturbances to the supply and demand for Federal Reserve deposits. The effect of a shock in the supply of reserves on the federal funds rate depends on the marginal benefits banks perceive to holding excess reserves and the marginal costs of discount window borrowing. We find empirically that a temporary shock to the supply of reserves will only induce banks to borrow at the discount window if it occurs on settlement Wednesday or the last day of the quarter. The structural estimates presented here suggest that a $1 billion loss in reserves will raise the federal funds rate by 6.6 basis points if it occurs on settlement Wednesday or the last day of a quarter and by 2.6 basis points if it occurs on other days. A number of alternative sources of evidence are used to corroborate these structural estimates and to provide independent statistically significant confirmation that a reduction in the supply of reserves clearly raises the overnight interest rate.