"See especially Chapter 13 of their book-an item that every macroeconomist should read.

4t is possible that similar studies of open economies could provide stronger evidence concerning the importance of monetary forces. For example, shifts in monetary policy to combat high rates of inflation in small, highly open economies appear to be associated with

does not adjust the money supply fully in response to these disturbances, the IS-LM-AS model predicts that they will lead to a negative relationship between money and output; a positive money demand shock, for example, will increase the money stock but increase the interest rate and reduce output. And even if the Federal Reserve accommodates the shifts, the fact that they are so large may cause a few observations to have a disproportionate effect on the results.

As a result of the money demand shifts, the estimated relationship between money and output is sensitive to such matters as the sample period and the measure of money. For example, if equation (5.40) is estimated using M2 in place of Ml, or if it is estimated over a sample period that is slightly longer, the results change considerably.

Because of these difhculties, regressions like (5.40), or more sophisticated statistical analyses of the association between monetary and real variables, cannot be used to provide strong evidence concerning the relative merits of monetary and real theories of fluctuations.

A very different approach to testing whether monetary shocks have real effects stems from the work of Friedman and Schwartz (1963). Friedman and Schwartz undertake a careful historical analysis of the sources of movements in the money stock in the United States from the end of the Civil War to 1960. On the basis of this analysis, they argue that many of the movements in money, especially the largest ones, were mainly the result of developments in the monetary sector of the economy rather than the response of the money stock to real developments. Friedman and Schwartz demonstrate that these monetary movements were followed by output movements in the same direction. Thus, Friedman and Schwartz conclude, unless the money-output relationship in these episodes is an extraordinary fluke, it must reflect causation running from money to output rather than in the opposite dlrection.2

C. Romer and D. Romer (1989) provide more recent evidence along the same lines. They search the records of the Federal Reserve for the postwar period for evidence of policy shifts designed to lower inflation that were not motivated by developments on the real side of the economy. They identify six such shifts, and find that all of them were followed by recessions. For example, in October 1979, shortly after Paul Volcker became chairman of the Federal Reserve Board, the Federal Reserve tightened monetary policy ch-amatically. The change appears to have been motivated by a desire to reduce inflation, and not by the presence of other forces that would have caused output to decline in any event. Yet it was followed by one of the largest recessions in postwar U.S. history.

large changes m real exchange rates, real Interest rates, and real output. Whether the evidence from such episodes m fact provides strong support for monetary nonneutrality has not been investigated systemahcally. The issue is complicated by the fact that the policy shifts are often accompanied by fiscal reforms and by large changes in uncertainty (see, for example, Sargent, 1982, and Dornbusch and Fischer, 1986).

effect, natural experiments provide potential instrumental variables for the St. Louis equation. The way to address the problem that there may be correlation between money growth and other factors that affect real output is to find variables that are correlated with money growth but uncorrelated with the other factors. One can then estimate the money-output regression by instrumental variables (or two-stage least squares). That is, one can examine how output growth is related to the component of money growth that is correlated with the instruments, and that is therefore uncorrelated with the omitted factors. Or, if one is interested simply m whether monetary movements affect real output but not in the precise \ alues of the coefficients, one can estimate the reduced form of the model-that is, one can regress output growth directly on the instruments. In effect, Friedman and Schwartz and Romer and Romer are using historical evidence about the source of monetary developments to try to find such instruments, and then examining the reduced-form relationship between output movements and their proposed instruments.

What Friedman and Schwartz and Romer and Romer are doing is searching for natural experiments to determine the effects of monetary shocks. If economies were laboratories, economists could randomly perturb the money supply and examine the subsequent output movements. Since the monetary disturbances would be chosen at random, the possibility that they were caused by output movements, or that there were other factors systematically causing the changes in both money and output, could be ruled out.

Unfortunately for economic science (though fortunately for other reasons), economies are not laboratories. The closest we can come to a laboratory experiment is to look for times when historical developments bring about monetary changes that are not caused by the behavior of output. For example, Friedman and Schwartz argue that the death in 1928 of Benjamin Strong, the president of the Federal Reserve Bank of New York, provides an example of such an independent monetary disturbance. Strongs death, Friedman and Schwartz argue, left a power vacuum in the Federal Reserve System and therefore caused monetary policy to be conducted very differently over the next several years than it otherwise would have been.

Natural experiments such as Strongs death are unlikely to be as ideal as genuine randomized experiments for determining the effects of monetary changes. There is room for disagreement concerning whether any episodes are sufficiently clear-cut to be viewed as independent monetary disturbances, and if so, what set of episodes should be considered. But since randomized experiments are not possible, the evidence provided by natural experiments may be the best we can obtain.

A related approach is to use the evidence provided by specific monetary interventtons to investigate the impact of monetary changes on relative prices. For example, as described in Section 9.3, Cook and Hahn (1989) confirm formally the common observation that Federal Reserve open-market

Problems

5.1. Consider the IS-LM model presented in Section 5.2. In this model, what are di IdM and dY/dM for a given value of P?

Barro (1989) presents a model where monetary expansions lower expected inflation. The model requires that prices jump instantaneously in response to the expansions, however.

operations are associated with changes in nominal interest rates. Given the discrete nature of the open-market operations and the specifics of how their timing is determined, it is not plausible that they occur endogenously at times when interest rates would have moved in any event. And although the issue has not been investigated formally, the fact that monetary expansions lower nominal rates strongly suggests that the changes in nominal rates represent changes in real rates as well. For example, monetary expansions lower nominal interest rates for terms as short as a day; it seems unlikely that they reduce expected inflation over such horizons.** Since real and Keynesian theories agree that changes in real rates affect real behavior, this evidence suggests that monetary changes have real effects.

Similarly, the exchange-rate regime appears to affect the behavior of real exchange rates. Under a fixed exchange rate, the central bank adjusts the money supply to keep the nominal exchange rate constant; under a floating exchange rate, it does not. There is strong evidence that not just nominal but also real exchange rates are much less volatile under fixed than floating exchange rates. In addition, when a central bank switches from pegging the nominal exchange rate against one currency to pegging the nominal exchange rate against another, the volatility of the two associated real exchange rates seems to change sharply as well. (See, for example, Genberg, 1978; Stockman, 1983; Mussa, 1986; and Baxter and Stockman, 1989). Since shifts between exchange-rate regimes are usually discrete, explaining this behavior of real exchange rates without appealing to real effects of monetary forces appears to require positing sudden large changes in the real shocks affecting economies. And again, both real and Keynesian theories predict that the behavior of real exchange rates has real effects.

The most significant limitation of this evidence is that the importance of these apparent effects of monetary changes on real interest rates and real exchange rates for quantities has not been determined. Baxter and Stockman (1989), for example, do not find any clear difference in the behavior of economic aggregates under floating and fixed exchange rates. Since real-business-cycle theories attribute fairly large changes in quantities to relatively modest movements in relative prices, however, a finding that the price changes were not important would be puzzling from the perspective of both real and Keynesian theories.