This paper studies monetary-policy shocks, de ned from federal funds target movements relative to daily interest-rate data. These shocks are nearly ideal measures of unexpected movements in monetary policy. Market expectations can summarize the vast amount of information used by the Fed in setting policy, and used by Fed-watchers in guessing the Fed’s actions, thus surmounting the omitted-variables problem in estimated policy rules. Interest-rate-based forecasts can adapt to changes in the Fed’s reactions to the rest of the economy—the time-varying parameter problem. If in one year the Fed worries about in ation, but in another year it places more weight on unemployment, market forecasts will adapt, but vector autoregressions (VARs) may not adapt and thus may incorrectly interpret anticipated actions to be shocks. Finally, high-frequency data surmount painlessly the pervasive orthogonalization problem. Interest rates all move together. Does this movement re ect Fed reaction to interest rates, or interest-rate reactions to the Fed? Neither recursive identi cation is plausible for monthly data. Fed of cials obviously look at interest rates just before the Federal Open Market Committee meeting, and just as obviously, interest rates react immediately to any change in the federal funds target. By contrast, the one-day correlation between a target change and interest-rate changes is obviously not a Fed reaction to intraday interest-rate news. Orthogonalization matters a lot in monthly data. If one orthogonalizes the funds rate before other interest rates, one estimates that policy shocks have a strong, “level” effect on other interest rates. If one orthogonalizes with the funds rate after other interest rates, one recovers an idiosyncratic funds-rate movement that does not affect other rates. The effect of the funds shock on long-term interest rates is entirely determined by the orthogonalization. Following this attractive intuition, Glenn Rudebusch (1998) and others have used fed funds futures data and the expectations hypothesis that the futures rate is equal to the expected future spot rate to de ne an expected fed funds target, and thus to de ne a shock. Alas, the institutional details of the funds rate and its futures market make this approach more complex than it seems. Also, the expectations hypothesis is currently most famous for the failure of the forward–spot spread to forecast interest-rate changes. Finally, this approach is limited to the sample since federal funds futures were introduced in 1988. With these thoughts in mind, we follow Piazzesi (2001) in de ning shocks from interest rates more generally, and without imposing the expectations hypothesis. We use two approaches. First, we run a regression of targetrate changes on interest rates just before the target change. Second, we de ne the shock as the change in the one-month Eurodollar rate from just before to just after the target change. Both measures use the fact that there has been a target-rate change; they omit from the shocks all dates on which the funds rate might have been expected to change but did not. Throwing out shocks need not bias responses, and we suspect that the response to unexpected target changes is different from the response to target changes, † Discussants: Mark Gertler, New York University; Narayana Kocherlakota, University of Minnesota; Helene Rey, Princeton University.