RDP 8602: Short-Term Interest Rates, Weekly Money Announcements and Rational Forecasts 1. Introduction

One of the most important issues in monetary economics is whether monetary policy has real effects. The real rate of interest provides a potential channel through which monetary policy may affect real activity. Recent empirical studies that attempt to test for monetary effects via this channel typically take one of two forms. An example of the first type is provided by Mishkin's (1983) tests for a contemporaneous effect of unanticipated monetary policy on interest rates. As Mishkin points out, however, such studies are beset by the familiar problem of simultaneous causality between interest rates and the money supply. Unless one is willing to assume that Federal Reserve policy is independent of current interest rates, the data can only be interpreted by first assuming a structural model of the relationship between interest rates and policy. Since there is no consensus on such a model, there is little likelihood of agreement over the interpretation of the empirical “evidence” obtained.

This has led to a second line of testing for the real effects of monetary policy. Towards the end of each week, the Fed announces its estimate of the narrowly defined monetary aggregate for the statement week ending some eight to ten days previously. These data circumvent the causation question. At the time of the announcement, the announced change in the money stock must be exogenous with respect to current Fed policy, interest rates and asset prices. Any correlations between this announcement and subsequent changes in interest rates or asset prices, will thus be evidence of causation from money to interest rates. They can not be interpreted as effects that run from interest rates to money.

These studies have typically found a significant, positive, effect of the announcement on interest rates. A rise in interest rates has been found to be associated with an unexpected increase in the money supply which is revealed by the announcement. Several authors have offered explanations for this positive response. There are four main contenders. By necessity, all of them share a common feature. The announcement can only affect interest rates by changing agents' expectations of variables. The effect must be through information sets, and not via other monetary transmission channels.^[1]

Cornell (1983) identifies these four explanations as the Keynesian, expected inflation, real activity and risk premium hypotheses. The Keynesian (or policy anticipation) hypothesis explains the correlation by suggesting that the announcement alters agents' expectations about future Fed policy. An unanticipated increase in the money stock will induce the Fed to tighten monetary policy in the future. In a Keynesian framework, anticipation of future monetary contraction leads to increases in real and nominal interest rates today.^[2] Under the expected inflation hypothesis, an unanticipated increase in the money supply is interpreted as an upward revision in the Fed's operating target. This produces expectations of higher inflation in the future, raising the nominal interest rate now, via the well known Fisher equation.^[3] There is no effect on real interest rates. The real activity (or signaling) hypothesis postulates that the money announcement provides a signal of fluctuations in future real activity by revealing unperceived information about money demand.^[4] Hence, an unanticipated increase in the money stock implies that real activity will be higher than previously thought. This news induces an increase in interest rates. Finally, under the risk premium hypothesis, the money announcement provides agents with information about other agents' risk preferences and their beliefs about the riskiness of assets that compete with money. ^[5] For example, an unanticipated increase in the money stock may indicate that agents' view non-monetary assets as riskier than had previously been perceived, leading to a fall in the prices of these assets and a rise in interest rates.

These hypotheses each have different implications for the causal relationship between money and real interest rates. However, they have similar predictions for the response of short-term and long-term interest rates to the weekly money announcement.^[6] This has lead to a proliferation of studies that seek to distinguish among the hypotheses on the basis of the response of prices in other asset markets.^[7] For instance, Hardouvelis (1984) looks at foreign interest rates and exchange rates. The latter are also examined by Cornell (1982 and 1983), Rngel and Frankel (1984), Gavin and Karamouzis (1984) and Husted and Kitchen (1985). The stock market response is evaluated in studies by Cornell (1983) and Pearce and Roley (1984). Frankel and Hardouvelis (1985) study the response of commodity prices.

The net result of all these regressions has been an inability to distinguish among the competing hypotheses. Loeys (1985) concludes that even a hypothesis (suggested by Hardouvelis (1984) and formalised in Hardouvelis (1985)) that combines the policy anticipation and expected inflation explanations^[8], is only partially consistent with the data.

This paper does not attempt to wade into the muddy waters of testing these competing hypotheses. Rather, it takes the analysis back a step in focussing on two key assumptions that are implicitly made in all these studies.

Firstly, it has almost always been assumed that the survey data provided by Money Market Services Inc. provide the best predictor of the announced change in the money supply.^[9] The results presented here reject the null hypothesis that the survey data provide a rational expectation of the announcement on the basis of information available just before the announcement. The gap between the announced change and the expected change as measured by the survey is not, therefore, the best measure of the news content of the Fed's announcement.

The second assumption also concerns news and is, perhaps, more important for the interpretation of these studies in terms of the wider issue of the relationship between money and real interest rates. The “news” relevant for this issue is that about the actual money stock. It is implicitly assumed in these studies that both the survey and the announcement provide rational expectations (with respect to the information sets available at the time) of the actual (as opposed to the announced) change in the money supply. The analysis in this paper shows that neither the survey nor the announcement are rational predictors of this change relative to the other information available to agents. In particular, subsequent revisions made to the announced money supply data are partly forecastable at the time of the announcement. This suggests that agents who make efficient use of information would combine these announced data with other available data when forming their expectations. The gap between the announced change and the expected change as measured by the survey is not, therefore, the best measure of the unexpected money stock change. The maintenance of both these assumptions in previous studies biases the parameter estimates and invalidates the related hypothesis tests.

The analysis presented here takes these problems into account by estimating rational expectations of the change in the money stock. These are calculated both with respect to information available just before the announcement and that available just after the announcement. The former provides an estimate of the expected change, and the gap between them yields a measure of the money surprise. Appropriate econometric techniques (recently outlined in Pagan (1984)) are then used to obtain consistent and efficient estimates of the “announcement effect” on short-term interest rates. This enables tests to be calculated to determine whether this effect, and/or the structure of the forecasts, change in response to changes in Fed policy or in its measurement of the announcement data.

The remainder of the paper is in five sections. Section 2 outlines the model that has previously been used to measure the announcement effect, and the modifications that are necessary in the light of the inadequacies of the survey and announcement data. The next section discusses the estimation technique that is appropriate for the modified model. Forecasting equations for the survey and announced changes are obtained in Section 4, and the rationality tests presented. The estimation of the announcement effect, and the extent to which it has changed over time, is dealt with in Section 5. Some concluding comments are made in the final section.

Footnotes

Although these channels will generally be necessary for the effects on expectations to lead to changes in real activity. [1]

See, for example, Urich and Wachtel (1981). [2]

Cornell (1983) and Hardouvelis (1984) discuss this explanation in more detail. [3]

Nichols, Small and Webster (1983) and Siegel (1985) provide different examples of this hypothesis. The interest rate response generally depends on the relationships between real output, interest rates and money. [4]

See Cornell (1983). [5]

Cornell (1982 and 1983), Gavin and Karamouzis (1984), Grossman (1981), Hardouvelis (1984), Loeys (1985), Roley (1983) and Urich and Wachtel (1981) all examine the response of short-term interest rates. The response of long-term rates is studied by Cornell (1983) and Gavin and Karamouzis (1984). Forward interest rates are examined by Gavin and Karamouzis (1984), Hardouvelis (1984), Loeys (1985) and Shiller et al (1983). [6]

Cornell (1983) has a good discussion of the responses predicted by each hypothesis in the different markets. [7]

The argument is that the lag structure of the economy is such that both effects work simultaneously with the first one dominating the response at the short end of the financial market and the second one dominating at the long end. [8]

A partial exception is provided by Roley's (1983) adjustment of the survey data in one of three estimation periods. [9]