Federal Reserve Information and the Behavior of Interest …dromer_aer2000.pdf · Federal Reserve...

Federal Reserve Information and the Behavior of Interest Rates

By CHRISTINA D. ROMER AND DAVID H. ROMER*

This paper tests for the existence of asymmetric information between the FederalReserve and the public by examining Federal Reserve and commercial inflationforecasts. It demonstrates that the Federal Reserve has considerable informationabout inflation beyond what is known to commercial forecasters. It also shows thatmonetary-policy actions provide signals of the Federal Reserve’s information andthat commercial forecasters modify their forecasts in response to those signals.These findings may explain why long-term interest rates typically rise in response toshifts to tighter monetary policy.(JEL E52, E43, D82)

Asymmetric information between the Fed-eral Reserve and the public is a phenomenonthat is often posited but rarely tested. Numer-ous models of central-bank behavior, for ex-ample, show that the existence of asymmetricinformation has important implications forthe effectiveness of policy and the conse-quences of dynamic inconsistency.1 Yet, thereis little evidence concerning whether the Fed-eral Reserve does indeed possess informationabout the state of the economy that is notknown by the public.

Asymmetric information between the Fed-eral Reserve and the public is also often men-tioned as a possible explanation for a puzzlingempirical phenomenon: the response of long-term interest rates to monetary-policy actions.Standard theories of the effects of monetarypolicy imply that an exogenous shift to tighterpolicy raises short-term interest rates tempo-rarily by raising real rates, but lowers them inthe long run by reducing inflation. When

these theories are coupled with the expecta-tions theory of the term structure, they predictthat a shift to tighter policy lowers interestrates on bonds of sufficiently long maturities.In fact, however, when the Federal Reserveundertakes contractionary open-market oper-ations, interest rates for securities of all ma-turities typically rise (Timothy Cook andThomas Hahn, 1989a). These increases occuron the day of the action and occur even whenthe actions are planned in advance; thus theymust represent responses to the actions them-selves. A common explanation of the in-creases is that when the Federal Reservetightens policy, market participants infer thatit has unfavorable information about thelikely behavior of inflation, and they thereforerevise their expectations of inflation upward.It is this upward revision in inflation expec-tations caused by the revelation of FederalReserve information that causes long-term in-terest rates to rise.

In this paper we use Federal Reserve andcommercial forecasts to test whether the centralbank actually does possess additional informa-tion about the current and future states of theeconomy. The key idea is that information theFederal Reserve has about the economy that isnot known to market participants is likely to bereflected in the Federal Reserve’s internal fore-casts. Because the Federal Reserve makes itsforecasts public only after five years, the fore-casts can contain information that is not knowncontemporaneously by market participants. Inthis analysis we look primarily at the FederalReserve’s knowledge about inflation, because

* Department of Economics, University of California,Berkeley, CA 94720. We are grateful to Normand Bernard,Dean Croushore, Carol Low, Stephen McNees, RandellMoore, and David Wyss for providing forecast data andinformation, and to Jeffrey Fuhrer, Alan Krueger, MichaelPrell, Vincent Reinhart, Glenn Rudebusch, Paul Samuelson,and the referees for helpful comments and suggestions. Weare also grateful to Matthew Jones, Patrick McCabe, andClara Wang for research assistance and to the NationalScience Foundation and the John Simon Guggenheim Me-morial Foundation for financial support.

1 See, for example, Thomas J. Sargent and Neil Wallace(1975), Robert J. Barro (1976), Barro and David B. Gordon(1983), Matthew B. Canzoneri (1985), and Alex Cukiermanand Allan H. Meltzer (1986).

429

we then use the results to test the asymmetric-information explanation of the response of in-terest rates to monetary actions. However, tocheck the robustness of our results, we also lookfor asymmetric information about the path ofreal output.

This analysis of asymmetric information andits implications for the behavior of interest ratesproceeds in several steps. Section I describesthe forecast data that we use. It also presentspreliminary diagnostic tests of the rationality ofthe various forecasts.

Section II then investigates whether the Fed-eral Reserve has information about inflation be-yond what is known by market participants.Specifically, we ask whether, given commercialforecasts of inflation, the Federal Reserve fore-casts are useful in predicting inflation. To ana-lyze this question, we examine regressions ofinflation on commercial and Federal Reserveforecasts. We find that the Federal Reserve pos-sesses statistically significant and quantitativelyimportant additional information. In a typicalregression, the coefficient on the commercialforecast is small and insignificant while that onthe Federal Reserve forecast is substantial andhighly significant. This suggests that the opti-mal forecasting strategy of someone with accessto both forecasts would be to put essentially noweight on the commercial forecast. These find-ings are robust across forecasting horizons,commercial forecasters, and sample periods.We also find that the Federal Reserve possessesequally important additional information aboutthe path of future output. We argue that theFederal Reserve’s information advantage stemsnot from early access to government statistics orinside information about monetary policy, butrather from the vast resources it devotes toforecasting.

Section III turns to the link between FederalReserve information and the behavior of inter-est rates. For the asymmetric-information hy-pothesis to explain why long-term rates risefollowing a monetary contraction, it is notenough that the Federal Reserve possesses use-ful information about future inflation. It is alsonecessary that monetary actions provide signalsof this information and that market participantsrespond to these signals. And these effects mustbe large enough to explain the anomalousmovements in interest rates that we observe.

To address the signaling issue, we askwhether it is rational for market participants tomake inferences about the Federal Reserve’sinflation forecasts from its policy actions. Spe-cifically, we regress the Federal Reserve fore-cast on the contemporaneous commercialforecast and an indicator of Federal Reserveactions. The results of these tests, although notas strong as the results concerning the existenceof asymmetric information, support the hypoth-esis that the Federal Reserve’s actions signal itsinformation.

To address the response issue, we examinewhether Federal Reserve actions actually affectmarket participants’ forecasts of inflation. Spe-cifically, we regress the commercial forecasters’next forecast of inflation on an indicator ofmonetary actions and their current forecast,controlling for the arrival of other informationabout inflation between the two forecast dates.The results of these regressions are broadlysimilar to those concerning the information con-tent of the Federal Reserve’s actions. The esti-mates suggest that commercial forecasters raisetheir expectations of inflation in response tocontractionary Federal Reserve actions, but thatthey do so by slightly less than one wouldexpect given the earlier results.

We then use the quantitative estimates fromthese tests to see if the effects are large enoughto explain the observed response of interestrates at different horizons to monetary-policyactions. We find that between a fifth and a halfof the rise in short-term rates following a con-tractionary action can be accounted for bychanges in expected inflation caused by therevelation of Federal Reserve information. Sim-ple simulations suggest that the effects of infor-mation revelation may be even more importantat longer horizons. We find that between halfand all of the rise in long-term rates in responseto monetary contractions may be due to therevelation of Federal Reserve information.

I. Data

We use inflation forecasts from both the Fed-eral Reserve and commercial forecasters. Weview the commercial forecasts as being the ex-pectations of market participants, or at least akey input into their expectation formation. Thisview is consistent with the fact that some of the

430 THE AMERICAN ECONOMIC REVIEW JUNE 2000

commercial forecasts we consider are createdby firms managing large portfolios; thus, theyare the forecasts of market participants. It isalso consistent with the fact that market partic-ipants pay for the commercial forecasts, sug-gesting that they view information processingas difficult and commercial forecasts as valu-able. Given this, it is plausible that marketparticipants will just adopt the commercial fore-casts as their own or use them as a key startingpoint in their analysis.2

An alternative view that is also consistent withour focus on commercial forecasts is that com-mercial forecasts are representative of market par-ticipants. In this view, the commercial forecastsare merely a well-documented example of theexpectation-formation process of market partici-pants. In either view, one can use an analysis ofthe relationship between commercial forecasts andthe Federal Reserve forecasts to see if the FederalReserve has information about future inflation thatis not known to market participants.

A. Forecasts

We consider forecasts from the Federal Re-serve and three commercial forecasters. Theparticular inflation forecasts we analyze arethose for the GNP deflator.3 We also consider

forecasts for real GNP in a robustness check onthe inflation results. This section therefore de-scribes the sources of the Federal Reserve andcommercial forecasts. It also discusses issues ofconsistency and timing related to these data.

The Federal Reserve forecasts are containedin the “Green Book” prepared by the staff of theBoard of Governors before each meeting of theFederal Open Market Committee (FOMC).These forecasts are available for the period1965:11–1991:11.4 The Green Book typicallyforecasts inflation and real GNP growth for fiveor six quarters into the future, though the hori-zon of the forecast varies over time and with thedate of the FOMC meeting.

Because the Federal Reserve forecasts are tiedto FOMC meetings, there are no forecasts inmonths when the FOMC does not meet. In the late1960’s and 1970’s, there are forecasts almost ev-ery month; in the 1980’s, there are typically eightforecasts per year. The time of the month whenthe forecast is made also varies, because the dateof the FOMC meeting varies. FOMC meetingsmore often occur during the first half of the month,but the pattern is not regular.5

The first set of commercial forecasts is fromBlue Chip Economic Indicators.6 Around thefifth of each month, Blue Chip surveys eco-nomic forecasters at approximately 50 banks,corporations, and consulting firms. It then pro-duces a consensus forecast (which is the median

2 As David S. Scharfstein and Jeremy C. Stein (1990),Owen Lamont (1995), Tilman Ehrbeck and Robert Wald-mann (1996), and others point out, there may be agencyproblems between commercial forecasters and their clientsthat cause forecasters not to report their true expectations ofinflation. This is unlikely to be a problem for our investi-gation, however. To begin with, simple models of agencyproblems imply that forecasters are concerned about theaccuracy of their forecasts and about their forecasts relativeto others’ forecasts. As a result, the models imply thatforecasters’ predictions are centered around their true ex-pectations, and thus that median forecasts, which are whatwe mainly consider, reflect forecasters’ true expectations(Lamont, 1995). More importantly, the hypothesis that theFederal Reserve’s apparent additional information is in factknown to market participants requires that the market par-ticipants pay for forecasts that they know to be biased,despite the fact that they possess enough information toproduce forecasts incorporating all of the information con-tained in the forecasts of a large organization (the FederalReserve) that devotes vast resources to forecasting. Finally,Ehrbeck and Waldmann (1996) find that agency models’predictions are rejected in the data.

3 The obvious alternative is the Consumer Price Index(CPI). We use the GNP deflator for two reasons. First,

forecasts for the GNP deflator are available for a muchlonger sample period. Second, interest rates were includedin the CPI until 1983. This greatly complicates the analysisof the link between inflation forecasts and monetary policy.

4 The end date is determined primarily by the FederalReserve’s policy of releasing information with a five-yearlag. However, we choose to stop the sample in 1991:11 toavoid the awkwardness of the switch from GNP to GDP inthe government statistics. Dean Croushore of the FederalReserve Bank of Philadelphia provided a machine-readableversion of the Green Book forecasts for the GNP deflator.We updated and revised his series using a hard copy pro-vided by the Board of Governors. The real GNP forecastswere obtained from the same documents provided by theBoard of Governors.

5 Occasionally, there are two or more Federal Reserveforecasts in a single month. This is especially common inthe late 1960’s and 1970’s. In our analysis, we use either thefirst or last forecast in a given month, depending on whetherthe particular application calls for a forecast that is early orlate in the month.

6 The historical Blue Chip Economic Indicators werepurchased from Capitol Publications, Inc.

431VOL. 90 NO. 3 ROMER AND ROMER: FEDERAL RESERVE INFORMATION

of the individual forecasts) for the percentagechange in the GNP deflator and real GNP overeach of the next six or seven quarters. The BlueChip forecasts are available starting in 1980:1.

The second set of commercial forecasts isprepared by Data Resources, Inc. (DRI).7 DRIproduces three forecasts each quarter; one early,one late, and one in the middle of the quarter.For comparability with monthly forecasts fromother sources, we assign the early forecast to thefirst month in the quarter, the middle forecast tothe second month, and the late forecast to thethird month. The early and late forecasts areavailable starting in the third quarter of 1970, sothe monthly start date is 1970:7. Because themiddle forecast is not available until the firstquarter of 1980, there are many missing obser-vations for the first decade. Each forecast ismade relatively late in the month. The forecasthorizon is typically seven quarters.

The third set of commercial forecasts is fromthe Survey of Professional Forecasters (SPF),currently conducted by the Federal ReserveBank of Philadelphia. This survey continues theAmerican Statistical Association/National Bu-reau of Economic Research Economic OutlookSurvey. Like the Blue Chip Economic Indica-tors, the Survey of Professional Forecasters isbased on many commercial forecasts. We againuse the median forecast. The SPF is conductednear the end of the second month of each quar-ter. For comparison with our other forecasts,which are monthly, we treat the Survey of Pro-

fessional Forecasters as a monthly series avail-able only in February, May, August, andNovember. Since the SPF forecasts for the GNPdeflator begin in the fourth quarter of 1968, thefirst observation on a monthly basis is 1968:11.Likewise, since the SPF forecasts for real GNPgrowth begin in the third quarter of 1981, thefirst monthly observation is 1981:8.8 The fore-cast horizon for both inflation and real growth isfour quarters.

Figure 1 summarizes the timing of the vari-ous forecasts for a typical quarter. It shows thatthe Blue Chip surveys occur early in eachmonth, the DRI forecasts occur late in eachmonth, and the Survey of Professional Forecast-ers occurs at the end of the middle month of thequarter. We have placed the timing of the Fed-eral Reserve’s forecast slightly before the mid-dle of the month to reflect the average date ofthese forecasts; however, the actual date of theforecasts varies from month to month. Like-wise, we have shown a Federal Reserve forecastin the first and third months of the quarter toreflect the fact that Federal Reserve forecastsare made roughly two months out of three.Again, the actual months in which forecasts aremade vary from quarter to quarter.

The time line also helps clarify the time-series nature of our data. We have monthlyobservations of forecasts of inflation variousnumbers of quarters in the future. For example,we have monthly predictions of inflation twoquarters ahead for each forecaster. Our subse-

7 The DRI forecasts were collated and provided by Ste-phen K. McNees of the Federal Reserve Bank of Boston.They are used with permission from DRI. The forecasts arefor the level of the GNP deflator and real GNP. Forecasts forinflation and real growth are calculated as quarterly percent-age changes at an annual rate.

8 We use a version of the forecasts compiled by DeanCroushore of the Federal Reserve Bank of Philadelphia.Like DRI, the SPF forecasts the level of the GNP deflatorand real GNP. Forecasts for inflation and real growth areagain calculated as quarterly percentage changes at an an-nual rate.

FIGURE 1. TIMING OF FORECASTS IN A TYPICAL QUARTER

Note: BC is the abbreviation for Blue Chip Economic Indicators, F is for the Federal Reserve, DRI is for Data Resources,Inc., and SPF is for the Survey of Professional Forecasters.


quent regressions will analyze the behavior ofthe forecasts for a given horizon made in acertain month by each forecaster.

B. Data on Actual Inflation and Output

One data issue involves the appropriate ac-tual series to use for comparison with the vari-ous forecasts. Because the U.S. CommerceDepartment data on the GNP deflator and realGNP are continually revised, a choice has to bemade about which revision to use. GNP statis-tics for a quarter are first released toward theend of the first month following the quarter.Because some component series are not avail-able, these initial estimates are subject to asubstantial margin of error. They are revised atthe end of the second month following the quar-ter, and again at the end of the third month asmore data become available. There is a compre-hensive annual revision each July and a rebench-marking and conceptual reworking roughlyevery five years.

We use the second revision (done at the endof the subsequent quarter) in our analysis. Thedata are collected from the June, September,December, and March issues of theSurvey ofCurrent Business.To ensure consistency in thecalculation of growth rates, the current and pre-vious quarter data are always taken from thesame issue of theSurvey.

We feel that the second revision is the appro-priate series to use because it is based on rela-tively complete data, but is still roughlycontemporaneous with the forecasts we are an-alyzing. This series does not include therebenchmarking and definitional changes thatoccur in the annual and quinquennial revisions.As a result, it should be conceptually similar tothe series being forecast. At the same time, itdoes not have the errors associated with theincomplete initial estimates.9

C. Serial Correlation

In regressions comparing forecasts and actualdata, there is inevitably the problem that fore-cast errors are serially correlated and that theserial correlation increases as the horizon for theforecasts becomes longer. To see this, considerthe implication of an unexpected rise in inflationin the fourth quarter of 1990 for forecasts ofinflation three quarters ahead. This wouldclearly cause positive forecast errors for theJanuary, February, and March 1990 forecasts,since in all three cases forecasters are predictinginflation in the fourth quarter of 1990. But thefact that inflation is serially correlated alsomeans that the forecast errors would tend to bepositive for the forecasts of inflation three quar-ters ahead made in April through December1990—that is, until forecasters could incorpo-rate the rise in inflation in the fourth quarter of1990 into their forecasts.

To deal with this potential problem, we cal-culate robust standard errors for all of ourregressions. Specifically, when we considerforecasts for inflationh quarters ahead, the stan-dard errors are computed correcting for het-eroskedasticity and for serial correlation overh 1 1 quarters [that is, over 3(h 1 1) months].For example, when we consider the Blue Chipforecast (which is available every month) forthree quarters ahead, the standard errors allowfor heteroskedasticity and for 12th-order serialcorrelation. We follow Lars P. Hansen and Rob-ert J. Hodrick (1980) in putting full weight onthe serial correlation over allh 1 1 quarters,rather than using the Bartlett window approachof Whitney K. Newey and Kenneth D. West(1987).10

D. Forecast Rationality

Before testing for asymmetric information, itis useful to examine the rationality of the fore-casts. The view that market participants take acommercial forecast as their baseline or as a keyinput into their expectation-formation processmakes sense only if the forecasts are not grossly

9 Redoing our tests using the most recently available datahas little effect on the results. It is perhaps interesting tonote that the award given by the Blue Chip EconomicIndicators to the forecaster with the best record is based ona comparison of forecasts over the past four years with themost revised data available. Thus, at least for this one highlypublicized award, forecasters are judged on their ability topredict the government’s best estimate of GNP (and threeother series) rather than the initial estimates.

10 The Hansen-Hodrick procedure occasionally yieldsnegative variances and, thus, undefined standard errors. Insuch cases, we report Newey-West standard errors instead.These cases are noted in the tables.


irrational. Therefore, we present a simple test ofthe rationality of the forecasts.

Let pht denote actual inflation in the quarterh quarters after montht. For example, ift 5January 1990 andh 5 3, then pht is actualinflation in the fourth quarter of 1990 (that is,the percentage change in the price level at anannual rate from the third to the fourth quarterof 1990). Similarly, letpht denote a forecast ofpht that is made in montht. To test for forecastrationality, we estimate regressions of the form:

(1) pht 5 a 1 bpht 1 eht ,

and test the implication of rationality thata 5 0andb 5 1. The standard errors are corrected forserial correlation and heteroskedasticity as de-scribed above. For completeness, we analyzethe rationality of each of the commercial fore-casts and of the Federal Reserve forecasts.

Table 1 reports the results. The null hypoth-esis of rationality is almost never rejected at

TABLE 1—RATIONALITY TESTS FORINFLATION FORECASTS

pht 5 a 1 bpht 1 eht

Forecast horizon(Quarters) a b p-value R2 N

Blue Chip0 20.41 (0.36) 1.02 (0.08) 0.082 0.76 1431 20.67 (0.42) 1.02 (0.09) 0.004 0.69 1432 20.71 (0.75) 0.98 (0.17) 0.001 0.62 1433 20.52 (1.07) 0.90 (0.23) 0.001 0.53 1434 0.58 (0.76) 0.63 (0.13) 0.000 0.31 1385 1.05 (1.10) 0.48 (0.19) 0.000 0.22 1026 1.46 (0.92) 0.33 (0.13) 0.000 0.19 66

DRI0 0.26 (0.27) 0.97 (0.06) 0.559 0.76 2191 0.91 (0.37) 0.87 (0.07) 0.052 0.56 2192 0.80 (0.46) 0.88 (0.09) 0.228 0.47 2193 1.27 (0.89) 0.76 (0.17) 0.342 0.35 2194 1.88 (1.25) 0.63 (0.23) 0.263 0.23 2195 2.43 (1.49) 0.52 (0.27) 0.202 0.15 2196 3.16 (1.87) 0.37 (0.32) 0.144 0.07 2197 3.53 (1.99) 0.28 (0.34) 0.089 0.04 217

SPF0 20.12 (0.41) 1.05 (0.08) 0.569 0.71 931 0.42 (0.50) 0.97 (0.10) 0.275 0.50 932 0.88 (0.83) 0.89 (0.16) 0.442 0.33 933 1.76 (1.06) 0.71 (0.19) 0.253 0.20 934 2.08 (1.19) 0.65 (0.22) 0.217 0.16 88

Federal Reserve0 0.03 (0.33) 1.03 (0.07) 0.479 0.78 2511 0.34 (0.47) 1.00 (0.11) 0.280 0.60 2422 0.74 (0.58) 0.95 (0.12) 0.275 0.44 2243 0.34 (0.72) 1.03 (0.13) 0.534 0.43 2074 0.12 (0.99) 1.05 (0.17) 0.656 0.38 1775 20.16 (1.15) 1.06 (0.22) 0.922 0.34 1186 20.80 (1.14) 1.09 (0.28) 0.312 0.47 617 21.19 (1.43) 1.03 (0.36) 0.000 0.53 38

Notes:p denotes inflation, andp denotes the inflation forecast;h andt index the horizon anddate of the forecast. The sample periods are 1980:1–1991:11 for Blue Chip; 1970:7–1991:11for DRI; 1968:11–1991:11 for SPF; and 1965:11–1991:11 for the Federal Reserve. Numbersin parentheses are robust standard errors. Thep-value is for the test of the null hypothesisa 50, b 5 1.


conventional significance levels for the DRI,Survey of Professional Forecasters, or FederalReserve forecasts. It is, however, consistentlyrejected for the Blue Chip forecasts.

Further investigation shows that these rejec-tions are due to the large weight of the Volckerdisinflation in the Blue Chip sample, whichdoes not begin until 1980. When the other fore-casts are restricted to the same start date, theytoo fail the rationality test. And when the fore-casts are restricted to the period after the disin-flation, none consistently fail the test. Thefailure of the forecasts to satisfy the usual cri-teria for rationality during a large regime shift isnot surprising.

Table 1 also shows that essentially all of theforecasts contain important information aboutinflation. The estimates ofb, although often lessthan one, are almost all above one-half andsignificantly greater than zero. This further sug-gests that starting with such forecasts as a base-line is a sensible strategy.11

II. Does the Federal Reserve HaveAdditional Information?

This section compares commercial forecastsof inflation with those of the Federal Reserve.Our method of comparison reflects the questionwe are asking. Our main interest is in whetherthe Federal Reserve’s forecasts contain infor-mation that would be useful to market partici-pants. Therefore, we focus on the issue ofwhether individuals who know the commercialforecasts could make better forecasts if theyalso knew the Federal Reserve’s.

A. Specification

As before, letpht denote actual inflationhquarters after montht. Let pht

C andphtF denote

the commercial and Federal Reserve forecastsof pht in month t. Suppose that market par-ticipants are using the commercial forecast astheir baseline forecast or forming expecta-tions by making linear projections of inflationon the forecast information they have. If theFederal Reserve forecast becomes available,market participants could use this informationby making a linear projection of actual infla-tion on the commercial and Federal Reserveforecasts. That is, their forecasts would be thefitted values of

(2) pht 5 d 1 gCphtC 1 gFpht

F 1 nht .

In this regression, the Federal Reserve forecastis useful in predicting inflation if and only ifgFdiffers from zero. Thus, testing whether theFederal Reserve forecast contains valuable in-formation requires estimating regressions like(2) and testing whethergF differs from zero.

In our basic regressions, we consider fore-casts for each quarter separately. An alternativeis to examine forecasts of average inflation overvarious horizons. That is, one can estimateequations of the form:

(3) p#ht 5 d 1 gC p# htC 1 gF p# ht

F 1 n# ht ,

wherepht is average inflation up to horizonhand p ht

C and p htF are the commercial and Fed-

eral Reserve forecasts ofpht.12 The regressions

using averages provide useful summaries of theoverall relationship between inflation and theforecasts. They also provide a check that therelationship is systematic rather than the resultof quarter-to-quarter noise.

B. Basic Results

The results of estimating equation (2) foreach commercial forecaster and each forecasthorizon are presented in Table 2. Our maininterest is ingF, the coefficient on the FederalReserve forecast. The estimates indicate over-whelmingly that the Federal Reserve possessesvaluable information not contained in the

11 We have also estimated versions of equation (1) thatinclude lagged inflation and the forecaster’s previous fore-cast error as right-hand-side variables. We choose the tim-ing of these variables so as to ensure that they wereavailable at the time of the forecasts. The results againsupport the rationality of the forecasts. Lagged inflation andthe lagged forecast error have no consistent predictivepower for inflation given the forecasts, and including thesevariables has little impact on thep-values for the test of thehypothesis thatb 5 1 and that the other coefficients are allzero.

12 For example,p4t is the average ofp0t, p1t, p2t, p3t,andp4t; p 4t

C is the average ofp0tC , p1t

C , p2tC , p3t

C , andp4tC .


commercial forecasts. For horizons fartherahead than the current quarter, the point esti-mates ofgF are typically between 1.0 and 1.5.For the current quarter, they are smaller, but stillover 0.5. Most of the estimates are highly sig-nificant, and all but two are significant at the5-percent level.

In addition, the estimates ofgC, the coeffi-cient on the commercial forecast, are typicallysmall. In fact, most of the point estimates arenegative, though not statistically significant atconventional levels. In only two cases is theestimate significantly larger than zero at eventhe 10-percent level. Thus, an individual with

access to both forecasts would want to put littleweight on the commercial forecast. And sincethe estimates ofgF are generally close to one,simply using the Federal Reserve forecast isclose to the optimal way of combining the twoforecasts.

Table 2 also reports the results of using theaverage forecasts up to four quarters ahead (thelongest horizon for which all of the commercialforecasts are available); that is, it reports esti-mates of equation (3) withh 5 4. The estimatesof gF, the optimal weight on the Federal Re-serve forecast, are large and highly significant.Thus, it does not appear that quarter-to-quarter

TABLE 2—TESTS OFFEDERAL RESERVE ADDITIONAL INFORMATION FOR INFLATION

pht 5 d 1 gCphtC 1 gFpht

F 1 nht

Forecast horizon(Quarters) d gC gF R2 N

Blue Chip0 20.06 (0.40) 0.35 (0.23) 0.64 (0.18) 0.83 971 0.49 (0.52) 20.35 (0.27) 1.21 (0.20) 0.81 972 0.56 (0.45) 20.30 (0.25) 1.12 (0.22) 0.70 973 0.22 (0.60) 20.34 (0.32) 1.23 (0.25) 0.71 974 0.18 (0.68) 20.31 (0.32) 1.19 (0.37) 0.54 935 0.64 (1.17) 20.23 (0.41) 0.93 (0.49) 0.37 696 1.30 (0.77) 0.55 (0.18) 20.20 (0.18) 0.27 38

m (0–4) 0.50 (0.36) 20.28 (0.21) 1.11 (0.21) 0.91 93

DRI0 20.17 (0.34) 0.39 (0.16) 0.66 (0.18) 0.80 1701 0.10 (0.43) 20.03 (0.21) 1.04 (0.23) 0.62 1702 0.27 (0.50) 20.19 (0.20) 1.18 (0.18) 0.49 1683 20.16 (0.57) 20.24 (0.30) 1.32 (0.29) 0.48 1614 20.51 (0.65) 20.65 (0.38) 1.80 (0.41) 0.46 1465 20.67 (0.85) 20.72 (0.49) 1.87 (0.53) 0.41 1056 20.81 (1.05) 20.33 (0.43) 1.45 (0.55) 0.45 607 21.51 (1.49) 20.30 (0.38) 1.42 (0.66) 0.54 38

m (0–4) 20.15 (0.41) 20.53 (0.36) 1.57 (0.38) 0.74 146

SPF0 20.00 (0.38) 0.15 (0.19) 0.88 (0.18) 0.76 791 0.46 (0.47) 20.47 (0.21) 1.45 (0.21) 0.64 792 1.55 (0.77) 20.78 (0.44) 1.57 (0.38) 0.49 783 1.27 (0.83) 20.83 (0.33) 1.70 (0.32) 0.46 734 0.72 (0.81) 20.93 (0.36) 1.89 (0.34) 0.48 64

m (0–4) 1.09 (0.53) 21.08 (0.38) 1.93 (0.35) 0.75 64

Notes:p denotes inflation, andpC andpF denote commercial and Federal Reserve inflationforecasts;h and t index the horizon and date of the forecasts. The sample periods are1980:1–1991:11 for Blue Chip; 1970:7–1991:11 for DRI; and 1968:11–1991:11 for SPF.Numbers in parentheses are robust standard errors. The forecast horizonm (0–4) refers to theaverage of 0 to 4 quarters ahead.


noise is driving our finding of substantial Fed-eral Reserve additional information.13

C. Sources of the Federal Reserve’sInformation Advantage

Given our findings, it is natural to considerthe source of the Federal Reserve’s informationadvantage. In this regard, it is easier to identifyfactors that are not important than those that are.First, the Federal Reserve’s additional informa-tion is probably not due to inside informationabout monetary policy. Monetary policy ap-pears to have little impact on output and theprice level for at least three to four quarters (see,for example, Robert J. Gordon, 1993; and Ro-mer and Romer, 1994). Yet, the Federal Re-serve forecast is valuable in predicting inflationjust one or two quarters ahead. The fact that theFederal Reserve forecasts continue to havevalue at fairly distant horizons could indicatethat staff members have inside informationabout the FOMC’s commitment to a given pol-icy. However, evidence presented in the nextsection contradicts this interpretation.

The advantage is also almost surely not dueto the Federal Reserve gaining access to officialdata earlier than commercial forecasters. TheChairman of the Federal Reserve receives dataon economic variables such as unemploymentand inflation only the night before they arereleased to the public, and access to these ad-vance data is tightly restricted within the Fed-eral Reserve. Even if the advance data wereavailable to Federal Reserve forecasters, a day’slead time would not give them a net advantageover the Survey of Professional Forecasters andDRI, since the Federal Reserve forecast is typ-ically made well before these commercial fore-casts. Furthermore, the Federal Reserve forecastis valuable in predicting inflation many quartersahead. One would expect a data advantage to beof most use at very short horizons.

It is possible that the Federal Reserve receivesunofficial information from business leaders andbankers. Whether such potentially unrepresenta-tive reports could consistently improve its fore-casts is highly questionable. Furthermore, even if

such reports were the source of the Federal Re-serve’s advantage, it does not follow that thisinformation is not available to commercial fore-casters. We suspect that the Federal Reserve re-ceives such reports for the most part not becauseof its official status, but simply because it has anenormous network of regional employees.

More generally, we believe that the mostlikely explanation for the Federal Reserve’s ad-ditional information is that the Federal Reservecommits far more resources to forecasting thaneven the largest commercial forecasters. As aresult, it is able to produce superior forecastsfrom publicly available information. Under thisinterpretation, the Federal Reserve has no inher-ent forecasting advantage. It has the same“technology” as commercial forecasters forconverting labor and data into forecasts. It sim-ply chooses to use more of these inputs than anycommercial forecasters find profitable, and soobtains forecasts that have value beyond theinformation contained in commercial forecasts.

D. Robustness

Outliers.—One way that we check the robust-ness of our results is to examine the contributionof outliers. Figure 2 is a scatterplot for a typicalregression of the component of inflation orthogo-nal to the commercial forecast against the compo-nent of the Federal Reserve forecast orthogonal tothe commercial forecast; it is this partial associa-tion that underlies the estimate ofgF in equation(2). The particular commercial forecast series andhorizon represented are the three-quarter-aheadforecast from DRI. However, the plots for otherforecasters and other horizons are similar.

The scatterplot makes it clear that the explan-atory power of the Federal Reserve forecast forinflation is not the result of outliers: there is aconsistent positive relationship between the twoseries. Times when the Federal Reserve forecastis unusually high given the commercial forecastare generally times when inflation is unusuallyhigh given the commercial forecast, and thereverse pattern holds in times when the FederalReserve forecast is unusually low given thecommercial forecast.

Timing Disadvantage.—We next test the ro-bustness of the results to a different specifica-tion of the relative timing of the Federal

13 Looking at the average forecasts over other horizonsyields similar results.


Reserve and commercial forecasts. In the basicspecification, we use forecasts made in the samemonth. Because the Blue Chip surveys are doneat the beginning of the month while the FederalReserve forecasts are done throughout themonth, this specification gives the Federal Re-serve a potential advantage simply because ithas more data. And if some Blue Chip partici-pants report forecasts made a week or two be-fore the date of the survey, the advantage iseven greater.14 For the DRI and SPF forecasts,which are done late in the month, our specifi-cation puts the Federal Reserve at a disadvan-tage, except to the extent that some participantsin the SPF report out-of-date forecasts.

To ensure that any possible advantage thatour basic specification gives the Federal Re-serve does not account for our results, we do theexperiment of putting the Federal Reserve at adeliberate timing disadvantage. We reestimateequation (2) replacingpht

F , the Federal Re-serve’s forecast in montht of inflation h quar-ters later, with its forecast in montht 2 1 ofinflation h quarters after montht. This specifi-cation puts the Federal Reserve at a clear dis-advantage relative to all three commercial

forecasters. Table 3 shows that when we do this,the Federal Reserve forecast remains a powerfulpredictor of inflation. Indeed, neither the sizesof the coefficients nor thet-statistics are sub-stantially reduced by this change.

Multiple Forecasts.—In a third test of therobustness of the results, we examine whetherthe Federal Reserve’s inflation forecast containsuseful information beyond that contained in twoor more commercial forecasts. Since many mar-ket participants presumably do not have accessto multiple commercial forecasts, this test islikely to understate the value of the FederalReserve’s information.

We have multiple commercial forecasts forthe same month starting only in 1980. This istrue because Blue Chip forecasts do not beginuntil 1980 and DRI forecasts are not available inthe middle month of the quarter (which is whenthe SPF forecast is available) before 1980. Ouranalysis of multiple forecasts is therefore lim-ited to the period 1980–1991. For this periodwe consider two combinations of commercialforecasts: Blue Chip and DRI forecasts (forwhich we have observations every month), andBlue Chip, DRI, and SPF forecasts (for whichwe are limited to one observation per quarter).We regress actual inflation on the Federal Re-serve forecast and either the Blue Chip and DRIforecasts or all three commercial forecasts.

The results of this exercise are reported in Table4. They are only slightly weaker than when indi-vidual commercial forecasts are considered. Forthe current quarter in both specifications and forthe six-quarter horizon using Blue Chip and DRI,the Federal Reserve forecast is of little value inpredicting inflation. But in every other case, theestimated weight on the Federal Reserve forecastis close to one, usually with at-statistic over three.

Overall Forecast Accuracy.—Our results sug-gest that someone with access to both the FederalReserve and commercial forecasts should not justput positive weight on the Federal Reserve fore-cast, but put little weight on the commercial one.This suggests that the Federal Reserve may beforecasting inflation more accurately than thecommercial forecasters are. Specifically, one canshow that if the Federal Reserve and commercialforecasts are unbiased (or equally biased) andforecasters’ errors are uncorrelated with their fore-

14 If, however, one thinks of the survey as an input intomarket participants’ expectations formation rather than as aproxy for their expectations, the relevant date is when thesurvey is published, not when the forecasts are made.

FIGURE 2. PARTIAL ASSOCIATION OFINFLATION AND

FEDERAL RESERVE FORECAST

Note: The figure shows the components of actual inflationand of the Federal Reserve forecast that are orthogonal tothe DRI forecast. The forecast horizon is three quarters.


casts, the Federal Reserve’s mean squared error(MSE) is less than the commercial forecaster’s ifand only if gF, the coefficient on the FederalReserve forecast in (2), exceedsgC, the coefficienton the commercial forecast. If, however, the fore-casts are biased by different amounts or forecasterrors are correlated with forecasts, there is nonecessary connection between the relative sizes ofthe coefficients on the two forecasts in our earlierregressions and the forecasts’ relative accuracy.

Since the results thus far are suggestive aboutthe relative accuracy of the forecasts, here webriefly present some direct evidence on this issue.Table 5 compares the MSEs of the Federal Re-

serve and commercial forecasts at each horizon.Each comparison is done using the observationsfor which both forecasts are available; as a result,the MSEs reported for the Federal Reserve for agiven horizon vary according to the commercialforecaster with which the Federal Reserve is beingcompared. The fourth column of the table reportsthep-value for the test of the null hypothesis thatthe Federal Reserve and commercial MSEs areequal.15

15 To calculatep-values, we estimate: (pht 2 phtC )2 2

(pht 2 phtF )2 5 c 1 uht. Since the dependent variable

TABLE 3—TESTS OFFEDERAL RESERVE ADDITIONAL INFORMATION FOR INFLATION

WITH FEDERAL RESERVE AT A TIMING DISADVANTAGE

pht 5 d 1 gCphtC 1 gFph,t 2 1

F 1 nht


Blue Chip0 0.04 (0.46) 0.28 (0.26) 0.67 (0.18) 0.81 971 0.23 (0.47) 20.19 (0.27) 1.09 (0.21) 0.78 972 0.18 (0.45) 20.25 (0.26) 1.13 (0.22) 0.72 973 0.03 (0.62) 20.20 (0.28) 1.10 (0.21) 0.67 974 0.59 (0.67) 20.31 (0.32) 1.06 (0.34) 0.48 875 0.79 (0.93) 20.06 (0.33) 0.69 (0.44) 0.34 616 1.24 (0.75) 0.59 (0.21) 20.26 (0.25) 0.24 37

m (0–4) 0.39 (0.40) 20.25 (0.25) 1.09 (0.24) 0.90 87

DRI0 0.05 (0.32) 0.56 (0.13) 0.45 (0.15) 0.76 1701 0.17 (0.41) 0.12 (0.21) 0.89 (0.21) 0.58 1692 0.21 (0.49) 0.19 (0.25) 0.82 (0.20) 0.49 1683 20.07 (0.64) 20.13 (0.23) 1.19 (0.21) 0.45 1604 20.08 (0.71) 20.41 (0.29) 1.46 (0.31) 0.39 1345 20.45 (0.80) 20.45 (0.40) 1.55 (0.44) 0.35 896 20.88 (1.27) 0.10 (0.27) 0.90 (0.48) 0.56 517 20.04 (1.27) 0.14 (0.09) 0.51 (0.32) 0.43 24

m (0–4) 20.02 (0.39) 20.19 (0.28) 1.20 (0.29) 0.71 134

SPF0 20.11 (0.44) 0.51 (0.18) 0.57 (0.17) 0.75 651 0.79 (0.57) 0.39 (0.42) 0.57 (0.38) 0.50 652 1.48 (0.63) 20.48 (0.33) 1.33 (0.29) 0.44 643 1.21 (0.74) 20.65 (0.31) 1.55 (0.29) 0.45 564 1.83 (1.20) 20.72 (0.36) 1.53 (0.32) 0.32 44

m (0–4) 1.57 (0.73) 21.30 (0.59) 2.12 (0.53) 0.66 44

Notes:p denotes inflation, andpC and pF denote commercial and Federal Reserve inflation forecasts;h and t index thehorizon and date of the forecasts. To make the commercial and Federal Reserve forecasts comparable, the Federal Reserveforecast int 2 1 is of inflationh quarters aftert, not aftert 2 1. The sample periods are 1980:1–1991:11 for Blue Chip;1970:7–1991:11 for DRI; and 1968:11–1991:11 for SPF. Numbers in parentheses are robust standard errors. The forecasthorizonm (0–4) refers to the average of 0 to 4 quarters ahead.


The results show that the Federal Reserve’sinflation forecasts have indeed been more ac-curate than commercial forecasters’. In everycase, the MSE of the Federal Reserve forecastis lower than the commercial forecaster’s;

typically it is about 25 percent lower. Further,in a large majority of cases, the null hypoth-esis of equal forecast accuracy is rejected atthe 5-percent level or below.

These findings are consistent with the resultsof our tests for the information value of theFederal Reserve forecasts. In our basic regres-sions (Table 2), the coefficient on the FederalReserve forecast almost always exceeds the co-efficient on the commercial forecast. The nullhypothesis thatgC 5 gF is rejected at the5-percent level or below in about two-thirds ofthe cases, usually with at-statistic between 2.0and 3.5.

is the difference in the squared errors of the two forecastsfor a given observation, the estimate ofc is just the dif-ference between the commercial forecaster’s and theFederal Reserve’s MSEs. Thep-value reported in thetable is therefore thep-value for the test of the nullhypothesis thatc 5 0. The standard error ofc is cor-rected for heteroskedasticity and for serial correlationover h 1 1 quarters.

TABLE 4—TESTS OFFEDERAL RESERVE ADDITIONAL INFORMATION FOR INFLATION WITH MULTIPLE COMMERCIAL FORECASTS

pht 5 d 1 gBCphtBC 1 gDRIpht

DRI 1 gSPFphtSPF 1 gFpht

F 1 nht

Forecast horizon(Quarters) d gBC gDRI gSPF gF R2 N

Blue Chip and DRI0 0.12 0.10 0.60 0.26 0.86 97

(0.36) (0.22) (0.19) (0.18)1 1.15 20.82 0.46 1.13 0.82 97

(0.65) (0.43) (0.26) (0.16)2 0.97 20.57 0.30 1.04 0.70 97

(0.52) (0.29) (0.12) (0.20)3 1.46 21.05 0.70 1.02 0.75 97

(0.50) (0.31) (0.21) (0.23)4 1.17 20.80 0.49 1.01 0.56 93

(0.43) (0.34) (0.23) (0.27)5 2.66 21.43 1.02 0.76 0.45 69

(1.40) (0.71) (0.31) (0.38)6 1.24 0.59 20.04 20.21 0.27 38

(0.99) (0.48) (0.29) (0.19)

m (0–4) 1.38 20.89 0.63 0.96 0.92 93(0.37) (0.13) (0.16) (0.16)

Blue Chip, DRI, and SPF0 0.19 0.26 0.67 20.27 0.28 0.84 36

(0.51) (0.49) (0.21) (0.61) (0.23)1 0.83 20.97 0.65 0.07 1.11 0.84 36

(0.91) (0.76) (0.32) (0.47) (0.25)2 0.48 20.73 0.09 0.37 1.12 0.69 36

(1.09) (0.66) (0.54) (0.62) (0.29)3 2.74 21.29 1.34 20.36 0.75 0.68 36

(0.76) (0.45) (0.37) (0.40) (0.19)4 1.64 20.86 0.84 20.39 1.02 0.60 35

(1.15) (0.63) (0.40) (0.44) (0.34)

m (0–4) 1.93 21.02 1.19 20.48 0.92 0.93 35(0.40) (0.28) (0.21) (0.23) (0.21)

Notes:p denotes inflation, andpBC, pDRI, pSPF, andpF denote Blue Chip, Data Resources, Inc., Survey of ProfessionalForecasters, and Federal Reserve inflation forecasts;h and t index the horizon and date of the forecasts. The sample periodis 1980:1–1991:11. Numbers in parentheses are robust standard errors. The forecast horizonm (0–4) refers to the average of0 to 4 quarters ahead.


Real GNP.—Our final, and perhaps most im-portant, robustness check is to see if the FederalReserve’s information advantage for inflationextends to real GNP. Since inflation and realoutput are simultaneously determined, it wouldbe puzzling if the Federal Reserve had usefulinformation about one and not the other.

To see if the Federal Reserve possesses addi-tional information about real output, we rerun ourbasic regressions using real GNP growth in placeof inflation. The results are reported in Table

6. The table shows that the Federal Reserve cer-tainly possesses information about the course ofreal output that commercial forecasters would liketo have. The coefficient on the Federal Reserveforecast is always positive, almost always large,and usually statistically significant.

There are, however, two differences betweenthe results for output and those for inflation.First, the coefficient estimates are more variedfor output. For inflation, the typical estimate ofgF is close to one and the typical estimate ofgCis close to zero, which implies that someonewith access to both forecasts should discard thecommercial one. For output, a moderate numberof the estimatedgF’s are well below one and amoderate number of the estimatedgC’s are wellabove zero, suggesting that the optimal weighton the commercial forecast is positive.

The other substantial difference is that theFederal Reserve’s additional information atshort horizons is more pronounced for outputthan for inflation. For the contemporaneousquarter, the Federal Reserve appears to have alarge forecasting advantage over both BlueChip and SPF. One possible explanation for thisadvantage is that the Federal Reserve collectsand processes the index of industrial produc-tion. Therefore, at very short horizons it mayactually have more data about the state of theeconomy, rather than just be better at processingwidely available information.16

The fact that the Federal Reserve has a definiteforecasting advantage for real GNP (which mayderive from a genuine data advantage at shorthorizons) raises the possibility that the FederalReserve’s forecasting advantage for inflationcould be driven by its real-side advantage. Inparticular, perhaps the Federal Reserve’s superior

16 In a related exercise, we look at the Federal Reserveand commercial forecasts of CPI inflation. Despite the factthat the sample sizes in these regressions are substantiallysmaller than those for the GNP deflator because of datalimitations, the results are very similar: the Federal Reserveappears to have significant additional information about thisalternative measure of inflation. This information advantageis particularly striking at longer horizons. For example, thecoefficient on the Federal Reserve forecast in equation (2) islarger than 0.8 for every forecast of inflation four or morequarters ahead for each of the three commercial forecasterswe consider, and is almost always significant. For forecastsfor shorter horizons, the estimates ofgF are all positive, butonly a few of them are significant.

TABLE 5—OVERALL ACCURACY OF INFLATION FORECASTS

Forecasthorizon(Quarters)

Mean squared error

p-value NCommercialforecaster

FederalReserve

(Percentage points)Blue Chip

0 1.46 1.23 0.321 971 2.15 1.37 0.000 972 2.64 1.72 0.005 973 3.12 1.68 0.006 974 3.69 1.99 0.003 935 5.09 2.81 0.010 696 4.89 2.69 0.002 38

m (0–4) 1.22 0.50 0.011 93

DRI0 1.93 1.71 0.390 1701 3.98 3.13 0.034 1702 4.91 4.03 0.001 1683 5.44 4.37 0.061 1614 6.34 4.65 0.033 1465 7.51 5.50 0.077 1056 5.91 4.06 0.118 607 6.59 4.06 0.145 38

m (0–4) 2.28 1.61 0.015 146

SPF0 2.33 1.75 0.025 791 4.06 2.95 0.000 792 5.73 4.39 0.012 783 6.24 4.63 0.000 734 7.01 5.01 0.002 64

m (0–4) 2.51 1.70 0.001 64

Notes:The mean squared error is calculated as the averagesquared difference between forecasted and actual inflation.The sample periods are 1980:1–1991:11 for Blue Chip;1970:7–1991:11 for DRI; and 1968:11–1991:11 for SPF.The p-value is for the test of the null hypothesis that theFederal Reserve and commercial mean squared errors areequal. The forecast horizonm (0–4) refers to the average of0 to 4 quarters ahead.


inflation forecasts are simply derived from a Phil-lips curve that uses as inputs the superior real GNPforecasts.

To see if this is the case, we consider thefollowing simple extension of our regression inequation (2). We regress actual inflation at var-ious horizons not only on the commercial andFederal Reserve forecasts of inflation at thesame horizon, but also on the contemporaneousand one-quarter-ahead commercial and FederalReserve forecasts of real output growth. If theFederal Reserve’s forecasting advantage for in-flation is working through its near-term, poten-tially data-driven output forecasts, the FederalReserve inflation forecast should no longer be a

significant predictor of actual inflation. The re-sults of this extension are that the near-termoutput growth forecasts have little predictivepower for inflation and virtually no impact onthe significance of the Federal Reserve inflationforecast. This suggests that the Federal Reservehas a forecasting advantage for inflation relativeto commercial forecasters that is separate fromits advantage in forecasting real output.

III. Implications for the Behaviorof Interest Rates

We now turn to the implications of FederalReserve information for the behavior of interest

TABLE 6—TESTS OFFEDERAL RESERVE ADDITIONAL INFORMATION FOR REAL GNP GROWTH

yht 5 d 1 gCyhtC 1 gFyht

F 1 nht


Blue Chip0 1.00 (0.52) 20.82 (0.25) 1.47 (0.14) 0.61 971 0.32 (0.97) 0.55 (0.59) 0.26 (0.59) 0.19 972 1.20 (1.87) 0.05 (0.87) 0.48 (0.54) 0.08 973 1.78 (2.24) 20.00 (0.74) 0.22 (0.55) 0.01 974 21.02 (0.84) 20.57 (0.16) 1.96 (0.27) 0.38 935 20.64 (1.36) 20.20 (0.29) 1.32 (0.48a) 0.17 696 21.06 (1.82) 0.09 (0.43) 1.24 (0.35) 0.39 38

m (0–4) 21.12 (1.04) 0.49 (0.53) 0.87 (0.36) 0.56 93

DRI0 0.07 (0.39) 0.71 (0.22) 0.32 (0.24) 0.71 1691 20.14 (0.91) 0.55 (0.22) 0.42 (0.33) 0.37 1692 0.14 (1.09) 0.14 (0.29) 0.73 (0.39) 0.20 1673 20.04 (0.91) 0.18 (0.29) 0.65 (0.45) 0.12 1604 20.34 (0.71) 20.00 (0.21) 0.99 (0.33) 0.17 1455 21.72 (0.74) 20.03 (0.23) 1.38 (0.29) 0.22 1046 20.99 (1.21) 20.17 (0.27) 1.39 (0.48) 0.30 597 20.76 (1.72) 20.23 (0.72) 1.32 (0.29) 0.24 37

m (0–4) 20.10 (0.61) 0.29 (0.38) 0.65 (0.38) 0.49 145

SPF0 0.48 (0.29) 21.02 (0.47) 1.75 (0.40) 0.69 321 21.44 (0.61) 0.56 (0.53) 0.81 (0.52) 0.39 322 22.17 (1.11) 0.66 (0.53) 1.07 (0.66) 0.30 323 21.15 (1.50) 0.40 (0.28) 0.99 (0.44) 0.21 324 20.17 (1.01) 21.07 (0.55) 2.33 (0.46) 0.38 32

m (0–4) 21.75 (0.65) 0.42 (0.59) 1.20 (0.48) 0.64 32

Notes: yis the percentage change in real GNP, andyC andyF denote commercial and Federal Reserve forecasts of real GNPgrowth; h and t index the horizon and date of the forecasts. The sample periods are 1980:1–1991:11 for Blue Chip;1970:7–1991:11 for DRI; and 1981:08–1991:11 for SPF. Numbers in parentheses are robust standard errors. The forecasthorizonm (0–4) refers to the average of 0 to 4 quarters ahead.

a Standard error calculated using Newey-West procedure because the Hansen-Hodrick standard error cannot be computed.


rates. For asymmetric information to explainwhy interest rates at all horizons rise in responseto a monetary contraction, it must be the casethat some of the Federal Reserve’s additionalinformation is revealed by its actions and thatmarket participants respond to this informationrevelation by changing their forecasts of futureinflation. Furthermore, these effects must belarge relative to the observed movements ininterest rates.

A. Indicators of Federal Reserve Actions

The first step in this analysis is to derive anindicator of Federal Reserve actions. We con-sider two variants. The first is a simple dummyvariable derived from theWall Street Journal.Cook and Hahn (1989a and 1989b) catalog thedates from September 1974 to September 1979when theJournal reports that the Federal Re-serve deliberately moved the federal funds rate.From this catalog, we construct a dummy vari-able that is21 in the months when the FederalReserve loosened,11 in months when it tight-ened, and 0 in all other months.

We extend the sample period by replicatingCook and Hahn’s procedures for the monthsbetween March 1984 and December 1991. (Weskip the years surrounding the Volcker disinfla-tion because the Federal Reserve did not targetthe funds rate in this period.) To identify funds-rate changes, we check the front page of eachissue of theWall Street Journalfor some men-tion of Federal Reserve action or interest-ratechange. Occasionally, there was more than onefunds-rate change in a month. However, only inOctober 1987 was there both a tightening and aloosening in the same month. Therefore, in allbut this one month, assigning the dummy vari-able is straightforward. We deal with October1987 by excluding it from the sample.

This simple dummy variable may be a par-ticularly useful indicator of monetary actions. Itis possible that action of any sort is what revealsinformation. Thus, having an indicator that doesnot distinguish between large and small changescould be desirable. Furthermore, because thedates of actions are derived from the press, weare certain that this is information that commer-cial forecasters and other agents in the economyactually possessed.

An alternative indicator of policy actions that

we consider is the change in the Federal Re-serve’s actual federal funds-rate target. Thesedata are available for 1974:9–1979:9 and from1984:3 through the end of our sample period.We use the funds-rate target in effect at the endof the month as the monthly observation.17 Thechange in the target, therefore, reflects thechange from the end of the previous month tothe end of the current month.

The target series could be useful because itcalibrates the size of monetary actions. If commer-cial forecasters respond differently to changes inthe funds rate of different magnitudes, then it isuseful to know the size of the changes. The targetseries is also a useful complement to the dummyvariable derived from theWall Street Journalbe-cause it reflects what the Federal Reserve wasactually doing. Cook and Hahn (1989b) show thatwhile theJournal identifies most changes in thetarget, it misses some and misjudges the magni-tude of others. Particularly in analyzing the infor-mation revealed by Federal Reserve actions, it istherefore desirable to work with the Federal Re-serve’s own target information. At the same time,since most of the target information is revealed inthe press, the Federal Reserve series provides agood proxy for what market participants actuallyknew about the timing and magnitude of targetchanges.18

B. Information Revelation

The Federal Reserve’s information cannotmatter for the effects of policy actions on inter-est rates unless the actions reveal some of thatinformation. To investigate this issue, we con-sider the problem of market participants at-tempting to infer the information that the

17 The funds-rate target series is available in Glenn D.Rudebusch (1995). We construct observations for the end of1974:08 and 1984:02 (to be used in calculating changesover the next month) by combining the earliest observationof the funds-rate target in 1974:09 and 1984:03 and thereported change in the target.

18 For the 1980’s, it is quite difficult to derive a synthetictarget series from theWall Street Journal.In many in-stances, theJournal is confident that the Federal Reservehas moved, but it is unsure where the funds rate will cometo rest. Furthermore, theJournaloften reports the funds ratein comparison to a year ago, so it is unclear how large ashort-run change the newspaper observes.


Federal Reserve possesses that they do not. Aswith our examination of the existence of asym-metric information, we focus on informationabout inflation. Also as before, we use the com-mercial forecasts as our indicator of marketparticipants’ expectations.

Specification.—To see if market participantscould learn something about the Federal Re-serve’s additional information from monetaryactions, we regress the Federal Reserve forecastfor a particular horizon on a measure of FederalReserve actions and the contemporaneous com-mercial forecast for the same horizon. That is,we estimate equations of the form:

(4) phtF 5 c 1 uMt 1 fpht

C 1 vht ,

where phtF and pht

C are again the Federal Re-serve and commercial forecasts of inflationhquarters ahead, andMt is the Federal Reserve’smonetary-policy action in montht (measuredeither by our dummy variable or by the changein the funds-rate target). In this specification,the coefficientu shows whether, and by howmuch, monetary actions reveal that the FederalReserve forecast is systematically differentfrom what one would predict based on the com-mercial forecast. For example, a coefficient thatis large and positive would indicate that con-tractionary monetary-policy actions provide in-formation that the Federal Reserve inflationforecast is higher than usual relative to the com-mercial forecast. Because any information thatis publicly available at timet should be incor-porated in both the Federal Reserve and com-mercial forecasts, it is not necessary to includeany control variables in the regression.

The main issue that arises in the specification isthe relative timing of the inflation forecasts andFederal Reserve actions. Ideally, one would like tohave Federal Reserve and commercial forecaststhat were exactly contemporaneous and that weremade just before monetary actions. As describedin Section I, however, this is not feasible: FederalReserve and commercial forecasts are not madesimultaneously, and they are not made just beforeFederal Reserve actions.

Our best approximation to this ideal is thefollowing. We choose the timing of the com-mercial forecasts so that they are before themonetary-action variable for each observation.

The monetary-action variable refers to policychanges in montht. For the DRI and SPF fore-casts, which are made late in the month, wetherefore use the forecast in montht 2 1 as thecontrol variable. For the Blue Chip forecast,which is made at the beginning of the month,we use the forecast in montht as the controlvariable.

For the Federal Reserve forecast, we choosethe timing so that it is slightly after the com-mercial forecast used as the control variable. Asdescribed in Section I, although the FederalReserve forecasts are made at different times ofthe month, the majority of them are made in thefirst half of the month. This suggests that wheneither the DRI or SPF forecast int 2 1 is usedas the control variable, the Federal Reserveforecast int is the appropriate dependent vari-able. When the Blue Chip forecast int is used asthe control variable, the Federal Reserve fore-cast int is the appropriate dependent variable.

We use the Federal Reserve forecast slightlyafter the commercial forecasts to counteract alikely bias against finding signal revelation. Be-cause both the commercial and Federal Reserveforecasts are often made weeks before the mon-etary actions, the Federal Reserve may base itsactions on information not contained in its lastofficial forecast. Therefore, an action may sig-nal that the Federal Reserve forecast is evenmore different from that predicted based on thecommercial forecast than the analysis of pub-lished forecasts would suggest.19 Taking theFederal Reserve forecast just slightly after the

19 Because the commercial forecast is also before themonetary action, it is possible that the commercial forecast-ers also receive additional information that causes them torevise their forecasts. In this case, our focus on publishedforecasts could cause a bias toward finding signal revela-tion. However, the Federal Reserve presumably bases itsactions mainly on its own forecasts rather than those ofcommercial forecasters. Thus, times when its estimates ofinflation increase after its last published forecast but com-mercial forecasters’ do not are more likely to be followed bytightening than times exhibiting the reverse pattern. To theextent that this occurs, the actions are signaling a deviationin the usual relationship between Federal Reserve and com-mercial estimates of inflation; but our tests, which are basedon the published forecasts, will not capture this. As a result,if the Federal Reserve and commercial forecasts were madeat the same time, but both preceded the Federal Reserve’sactions, the tests would be biased against finding informa-tion revelation.


commercial forecast can help to counteract thisproblem.20

Results.—Table 7 reports the results. In PanelA, policy actions are measured using the dummyvariable; in Panel B, they are measured using thechange in the funds-rate target. As before, thestandard errors are computed allowing for het-eroskedasticity and for serial correlation overh 11 quarters. The sample periods used are deter-mined by the availability of the data.

The results support the hypothesis that shiftsto tighter policy signal that the Federal Reserveforecasts of inflation are unusually high giventhe commercial forecasts. Almost all of theestimates ofu are positive, and a substantialnumber are significantly greater than zero atconventional levels. None of the estimates aresignificantly less than zero.

As before, a convenient way of summarizingthe evidence from the different quarters is toconsider forecasts of average inflation over thenexth quarters.21 Table 7 therefore also reportsthe results using average forecasts up to fourquarters ahead, the longest horizon for which allthree commercial forecasts are available. In allsix variants considered, the estimated relation-ship is positive. Thet-statistic on the measure ofpolicy actions ranges from 1.5 to 2.9. The re-sults also suggest that the magnitude of theassociation is substantial. When we use the

dummy variable as the indicator of monetarypolicy, the average point estimate ofu for thisone-year horizon is 0.16. Thus, the estimatessuggest that a typical move to tighter policyindicates that the Federal Reserve forecast ofinflation over the coming year is between one-and two-tenths of a percentage point higher thanone would expect given the commercial fore-cast. When we use the change in the funds-ratetarget as a monetary indicator, the correspond-ing figure is 0.27: an increase in the funds-ratetarget of one percentage point signals a gap ofabout a quarter of a percentage point betweenthe Federal Reserve inflation forecast and whatone would expect given the commercial fore-cast. Thus, Federal Reserve actions appear to beimportant signals of its additional information.

These results shed further light on the source ofthe Federal Reserve’s information advantage. Asdescribed in the previous section, one possiblereason that the Federal Reserve could have usefulinformation about inflation is that it has superiorinformation about future monetary policy. As dis-cussed there, the fact that the Federal Reserve hasuseful information about inflation just one or twoquarters ahead already casts strong doubt on thishypothesis. The direction of the relationship be-tween the Federal Reserve’s information and itsactions provides a further piece of evidence. If theFederal Reserve has additional information aboutfuture inflation because it knows more about itslikely policy actions, then times when its inflationforecasts are unusually high should on average befollowed by moves to looser policy. However,such times are in fact followed by moves to tighterpolicy. This is consistent with the view that theFederal Reserve has additional information aboutthe economy not stemming from its knowledgeabout future policy. Times when the Federal Re-serve’s inflation forecasts are unusually high areon average times when it has received news thatinflation will be higher than expected, and when itis therefore about to tighten to dampen rises ininflation.

C. Expectations Response

The previous analysis shows that monetaryactions reveal some of the Federal Reserve’sinformation about future inflation. For this toexplain the response of interest rates to mone-tary actions, market participants must revise

20 For completeness, we have also examined the casewhere the Federal Reserve forecast usually precedes thecommercial forecast. For DRI and SPF, this means that weconsider the Federal Reserve forecast in the same month asthe commercial forecast; for Blue Chip, it means that weconsider the Federal Reserve forecast in the precedingmonth. Our analysis implies that this specification is unam-biguously biased against finding a signaling effect of policyactions. Consistent with this analysis, for DRI and SPF—where the Federal Reserve forecasts typically precede thecommercial forecasts by several weeks and the policy ac-tions by over a month—we obtain results that are qualita-tively similar to those from our main specification, butconsiderably weaker. For Blue Chip—where the FederalReserve forecasts usually precede the commercial forecastsby almost a month and the policy actions by more than amonth—we find no consistent relationship between policyactions and the Federal Reserve forecasts, controlling forthe commercial forecasts.

21 Thus, we estimate regressions of the formp htF 5 c 1

uMt 1 fp htC 1 vht, where bars over the variables indicate

averages up to horizonh. The results reported are forh 54, but they are similar for other horizons.


TABLE 7—ESTIMATES OF REVELATION OF INFORMATION ABOUT INFLATION

phtF 5 c 1 uMt 1 fpht

C 1 vht

Forecast horizon(Quarters) c u f R2 N

A. Dummy Variable

Blue Chip0 20.71 (0.71) 0.21 (0.12) 1.13 (0.18) 0.59 611 20.45 (0.47) 20.13 (0.14) 1.01 (0.11) 0.60 612 20.07 (0.63) 0.02 (0.10) 0.93 (0.16) 0.54 613 0.55 (0.90) 0.13 (0.08) 0.78 (0.20) 0.48 614 0.48 (0.66) 0.16 (0.05) 0.79 (0.14) 0.57 615 0.24 (0.44) 0.10 (0.05) 0.84 (0.09) 0.68 476 20.49 (0.61) 20.10 (0.07) 1.01 (0.14) 0.74 27

m (0–4) 20.43 (0.63) 0.07 (0.04) 1.02 (0.16) 0.76 61

DRI0 0.63 (0.29) 0.17 (0.14) 0.91 (0.07) 0.85 1001 0.86 (0.33) 0.29 (0.15) 0.87 (0.07) 0.84 1002 0.81 (0.34) 0.29 (0.12) 0.87 (0.07) 0.85 1003 1.01 (0.31) 0.25 (0.11) 0.82 (0.08) 0.84 1004 0.86 (0.38) 0.15 (0.11) 0.84 (0.09) 0.81 1005 0.67 (0.29) 0.18 (0.07) 0.90 (0.07) 0.89 856 0.50 (0.35) 0.08 (0.12) 0.92 (0.08) 0.87 527 1.50 (0.64) 0.47 (0.22) 0.66 (0.18) 0.80 28

m (0–4) 0.54 (0.32) 0.19 (0.13) 0.92 (0.07) 0.93 100

SPF0 20.57 (0.41) 0.13 (0.19) 1.13 (0.08) 0.88 471 20.52 (0.38) 0.12 (0.18) 1.09 (0.08) 0.86 472 20.42 (0.44) 0.29 (0.17) 1.08 (0.09) 0.80 473 20.47 (0.40) 0.20 (0.09) 1.06 (0.07) 0.86 474 0.36 (0.36) 0.27 (0.09) 0.92 (0.08) 0.81 46

m (0–4) 20.49 (0.26) 0.22 (0.10) 1.09 (0.05) 0.94 46

B. Change in Funds-Rate Target

Blue Chip0 20.69 (0.72) 0.31 (0.20) 1.11 (0.18) 0.57 621 20.48 (0.45) 20.29 (0.22) 1.02 (0.11) 0.60 622 20.02 (0.59) 0.18 (0.16) 0.92 (0.14) 0.55 623 0.54 (0.82) 0.33 (0.09) 0.78 (0.19) 0.49 624 0.48 (0.62) 0.44 (0.09) 0.79 (0.13) 0.60 625 0.36 (0.46) 0.34 (0.11) 0.81 (0.09) 0.69 476 20.39 (0.60) 20.02 (0.17) 0.98 (0.13) 0.73 27

m (0–4) 20.41 (0.60) 0.17 (0.06) 1.01 (0.15) 0.76 62

DRI0 0.55 (0.27) 0.05 (0.32) 0.93 (0.07) 0.85 1011 0.72 (0.30) 0.54 (0.22) 0.90 (0.06) 0.83 1012 0.68 (0.29) 0.60 (0.21) 0.90 (0.06) 0.85 1013 0.88 (0.29) 0.42 (0.22) 0.85 (0.07) 0.84 1014 0.77 (0.38) 0.16 (0.17) 0.86 (0.10) 0.80 1015 0.60 (0.27) 0.36 (0.10) 0.92 (0.06) 0.90 866 0.47 (0.27) 0.32 (0.19) 0.92 (0.07) 0.87 527 1.11 (0.67) 0.71 (0.47) 0.74 (0.19) 0.77 28

m (0–4) 0.44 (0.28) 0.35 (0.18) 0.95 (0.07) 0.92 101

SPF0 20.61 (0.38) 0.04 (0.30) 1.13 (0.08) 0.88 471 20.56 (0.36) 0.05 (0.30) 1.10 (0.08) 0.86 472 20.62 (0.40) 0.53 (0.32) 1.12 (0.09) 0.80 473 20.59 (0.39) 0.29 (0.17) 1.08 (0.07) 0.85 474 0.22 (0.37) 0.21 (0.10) 0.95 (0.08) 0.79 46

m (0–4) 20.62 (0.23) 0.32 (0.12) 1.11 (0.05) 0.93 46

Notes:pF and pC denote Federal Reserve and commercial inflation forecasts;h and t index the horizon and date of theforecasts.M is the indicator of monetary-policy actions. The sample periods are 1984:2–1991:11 for Blue Chip; and1974:8–1979:8 and 1984:2–1991:11 for DRI and SPF. Numbers in parentheses are robust standard errors. The forecasthorizonm (0–4) refers to the average of 0 to 4 quarters ahead.


their expectations of inflation to reflect the in-formation revealed by the actions. Since weview commercial forecasts as a proxy for or akey input into market expectations, this meansthat commercial forecasters must raise their ex-pectations of inflation when the Federal Reservetightens, rather than lower them as more con-ventional models of the effects of monetarypolicy would lead one to predict. In this sectionwe investigate whether this is the case.

Specification.—A straightforward way to ana-lyze how forecasters respond to Federal Reserveactions is to regress the next commercial forecaston an indicator of Federal Reserve actions and thelast commercial forecast before the action. A pos-itive coefficient on the action would indicate thatthe forecasters raise their inflation forecasts inresponse to a monetary tightening, controlling fortheir initial prediction.

The obvious complication is that the currentcommercial forecast is not the only other relevantpredictor of the next forecast. In particular, boththe monetary action and the next forecast could bethe result of information received after the currentforecast was made. For example, suppose thatthere is unfavorable news about inflation. Thencommercial forecasters may raise their inflationforecasts in response to this news, and the FederalReserve may tighten. The rise in the forecasts,however, would not be a response to the tighten-ing. Thus, in the absence of controls, the coeffi-cient estimate could be biased upward.

To address this possibility, we include as anadditional control variable the change in the Fed-eral Reserve’s inflation forecast in the intervalbetween the current and subsequent commercialforecasts. The change in the Federal Reserve fore-cast should reflect general information that be-comes available during the period between thetwo forecasts.22 Therefore, we estimate regres-sions of the form:

(5) ph,t 1 1C 5 h 1 lMt 1 kpht

C

1 r~ph,t 1 1F 2 pht

F ! 1 yh,t 1 1 ,

whereMt is again an indicator of Federal Re-serve actions. A positive value ofl would in-dicate that inflation expectations respond tomonetary actions in a way consistent with theexistence and revelation of Federal Reserve ad-ditional information.

Timing is again very important. We need toensure that the revision in the Federal Reserveforecast included as a control variable corre-sponds to the interval between the current andsubsequent commercial forecasts. Similarly, weneed to ensure that the current and subsequentcommercial forecasts that we examine bracketthe monetary action. Because the various fore-casts differ in when they are made during themonth, taking account of timing involves mak-ing minor adjustments in the time subscriptsgiven in equation (5).23

The timing depicted in equation (5) is mostaccurate for the regressions using the Blue Chipforecasts. Because these forecasts are made atthe beginning of the month,ph,t 1 1

C will be aforecast made soon after a monetary action inmontht andpht

C will be made before the action.Similarly, since the Federal Reserve forecaststend to be made in the first half of the month,the revision betweent andt 1 1 should reflectnew general information received by the com-mercial forecaster between forecasts.

Because the DRI forecasts are done at the endof each month, the forecasts for montht 1 1

22 The change in the Federal Reserve forecast reflects thearrival not just of new public information, but also of newinformation observed only by the Federal Reserve. On theone hand, this means that the change in the Federal Reserveforecast is a noisy measure of new public information. Tothe extent that the Federal Reserve acts on the basis of thenew public information, the fact that we are controlling forthis information imperfectly means that the coefficient onthe Federal Reserve’s action is biased up. On the other hand,to the extent that the Federal Reserve acts on the basis of its

new private information and commercial forecasters re-spond to those actions, by controlling for the change in theFederal Reserve forecast we tend to understate the impor-tance of its actions to commercial forecast revisions. Onecan show that in the natural baseline case where the FederalReserve puts the same weight on new public and privateinformation in choosing its action, the two sources of biasjust balance.

23 Similarly, theh subscripts need to be adjusted some ofthe time. In particular, to ensure that both the initial andsubsequent commercial forecasts concern inflation in thesame quarter, the dependent variable in (5) is the commer-cial forecast in montht 1 1 of inflation h quarters aftermontht. Thus when montht is the last month of a quarter,the dependent variable is in factph2 1,t 1 1

C . Likewise, theFederal Reserve forecasts for botht andt 1 1 are forecastsof inflation h quarters after montht.


reflect all of the events of and information re-leased during montht 1 1. As a result, onewants to see if the DRI forecasts in montht 11 respond to monetary actions in montht 1 1,not montht. Because the Federal Reserve fore-casts are more often done early in the month,one needs to control for the change in the Fed-eral Reserve forecast fromt 1 1 to t 1 2 tocapture the new information that DRI may re-ceive between the end of montht when it makesthe initial forecast and the end of montht 1 1when it makes its subsequent forecast.

The Survey of Professional Forecasters pre-sents even more complicated timing issues. TheSPF is only done at the end of the middle monthof each quarter. Thus, the next forecast aftermontht is in montht 1 3. For the same reasonsmentioned for DRI, the appropriate control vari-able is therefore the change in the Federal Reserveforecast from montht 1 1 to montht 1 4. Inaddition, to minimize the possibility that the Fed-eral Reserve actions are responses to informationthat becomes available between the two forecastdates rather than to its forecast as of the initialforecast date, we consider the relationship be-tween the SPF forecast int 1 3 and monetary-policy actions only in montht 1 1.24

Results.—Table 8 presents the results. Sincethe data are the same as those used in the

information-revelation regressions, the sampleperiods are the same as in those regressions.25

Panel A shows the results for the dummy vari-able for Federal Reserve actions, and Panel Bshows the results for the change in the funds-rate target. The estimates ofr, the coefficient onthe change in the Federal Reserve forecast, arepositive in the vast majority of cases, but areusually not significant. Not surprisingly, theestimates ofk, the coefficient on the initialcommercial forecast, are close to one and over-whelmingly significant.26

Our main interest, however, is inl, the co-efficient on the measure of policy actions. Theestimates ofl support the hypothesis that con-tractionary monetary actions cause commercialforecasters to raise their inflation forecasts. Avery large majority of the estimates are positive,and a substantial number of them are significantat conventional levels. None of the estimates aresignificantly less than zero. There is certainlyvariation in the strength of the finding, however.For the Blue Chip and SPF forecasts, the esti-mated coefficients are positive in every case,and for Blue Chip, they are often significant.For the DRI forecasts, in contrast, about a thirdof the estimates are negative, and only a few aresignificantly larger than zero. Table 8 alsoshows the results using revisions of averageforecasts up through four quarters ahead ratherthan revisions of forecasts for individual quar-

24 We have also investigated an alternative way of ad-dressing the problem that both the Federal Reserve’s actionsand commercial inflation forecasts could be responding toinformation released between the times of the initial fore-casts and the Federal Reserve’s actions. The alternative is tocontrol for the main pieces of information released early inthe interval between the two commercial forecasts. Relativeto our main approach of controlling for the change in theFederal Reserve forecast, this approach has an advantageand a disadvantage. The advantage is that, because it doesnot require data on Federal Reserve forecasts, it permits alarger sample. The disadvantage is that, because one cannotcontrol for all publicly available information, it can onlypartially address the problem.

The specific variables we control for are the percentagechanges in payroll employment, average hourly earnings ofproduction workers, and average weekly hours of produc-tion workers. Although this change in specification notice-ably alters the results of many of the individual regressions,it has virtually no impact on the fraction of the estimatesthat are positive or their average size. Because of the largersample sizes, however, the standard errors are generallysmaller; as a result, more of the estimates are significantlylarger than zero.

25 Because the Federal Reserve rarely makes long-termforecasts in two consecutive months, the sample sizes forthe six-quarter horizon using Blue Chip and the seven-quarter horizon using DRI are less than fifteen. We thereforedo not consider these horizons. Also, because the SPFforecasts are made only once a quarter, it is not possible toconsider the next forecast after montht of inflation forhorizonh 5 0 (the contemporaneous quarter): by the timeof the next forecast (t 1 3), inflation for the initial quarterhas been realized and thus is no longer being forecast. Wetherefore consider the responses of contemporaneous fore-casts to monetary actions only for Blue Chip and DRI.Finally, because theory predicts that a forecast should not bepredictable given the previous forecast, and because theestimated residuals do not show any consistent pattern ofserial correlation, the standard errors in Table 8 are cor-rected for heteroskedasticity but not for serial correlation.

26 Because the other variables in the regression are notknown when the initial forecast is made, the hypothesis thatthe forecast is rational does not imply thatk should equalone. We therefore do not impose this restriction. However,imposing it has little effect on the estimates or significanceof the other coefficients.


TABLE 8—ESTIMATES OF RESPONSE OFINFLATION FORECASTS TOMONETARY-POLICY ACTIONS

ph,t 1 1C 5 h 1 lMt 1 kpht

C 1 r(ph,t 1 1F 2 pht

F ) 1 yh,t 1 1

Forecast horizon(Quarters) h l k r R2 N

A. Dummy Variable

Blue Chip0 20.12 (0.16) 0.14 (0.06) 1.01 (0.05) 0.11 (0.07) 0.87 311 0.15 (0.35) 0.05 (0.03) 0.95 (0.09) 0.17 (0.09) 0.87 312 0.11 (0.20) 0.03 (0.03) 0.96 (0.05) 0.10 (0.10) 0.93 313 0.14 (0.19) 0.05 (0.02) 0.96 (0.04) 0.17 (0.07) 0.95 314 0.19 (0.13) 0.07 (0.02) 0.95 (0.03) 0.07 (0.06) 0.97 315 0.11 (0.12) 0.04 (0.03) 0.96 (0.03) 0.10 (0.06) 0.97 27

m (0–4) 0.09 (0.18) 0.05 (0.03) 0.97 (0.04) 0.39 (0.12) 0.94 31

DRI0 1.24 (0.23) 0.31 (0.12) 0.65 (0.07) 0.14 (0.15) 0.68 271 0.19 (0.16) 0.12 (0.07) 0.95 (0.04) 20.14 (0.13) 0.94 502 20.05 (0.32) 20.01 (0.12) 1.01 (0.08) 20.03 (0.28) 0.86 503 0.33 (0.24) 0.02 (0.11) 0.92 (0.06) 0.07 (0.14) 0.88 504 0.57 (0.24) 0.15 (0.14) 0.85 (0.06) 20.31 (0.37) 0.82 505 0.55 (0.17) 0.22 (0.07) 0.89 (0.03) 0.14 (0.18) 0.94 486 0.56 (0.35) 0.17 (0.14) 0.88 (0.09) 0.96 (0.52) 0.88 28

m (0–4) 0.41 (0.16) 0.03 (0.04) 0.87 (0.05) 0.33 (0.21) 0.92 27

SPF1 20.01 (0.46) 0.28 (0.19) 1.00 (0.10) 0.35 (0.14) 0.87 402 20.58 (0.44) 0.08 (0.19) 1.11 (0.09) 20.10 (0.11) 0.85 403 20.22 (0.27) 0.16 (0.11) 1.04 (0.05) 0.02 (0.12) 0.90 404 1.13 (0.27) 0.33 (0.13) 0.80 (0.05) 0.23 (0.19) 0.83 39

m (1–4) 0.08 (0.28) 0.30 (0.13) 0.99 (0.06) 0.03 (0.17) 0.92 39

B. Change in Funds-Rate Target

Blue Chip0 20.06 (0.18) 0.30 (0.15) 0.99 (0.05) 0.10 (0.07) 0.86 311 0.19 (0.35) 0.15 (0.09) 0.94 (0.09) 0.16 (0.10) 0.87 312 0.16 (0.22) 0.09 (0.08) 0.95 (0.05) 0.10 (0.10) 0.93 313 0.22 (0.17) 0.16 (0.05) 0.94 (0.04) 0.16 (0.07) 0.95 314 0.27 (0.13) 0.19 (0.05) 0.93 (0.03) 0.05 (0.06) 0.97 315 0.14 (0.11) 0.13 (0.06) 0.96 (0.02) 0.10 (0.06) 0.97 27

m (0–4) 0.14 (0.20) 0.13 (0.09) 0.95 (0.05) 0.37 (0.13) 0.94 31

DRI0 1.27 (0.24) 0.80 (0.26) 0.63 (0.06) 0.11 (0.14) 0.69 271 0.14 (0.17) 0.17 (0.17) 0.96 (0.04) 20.13 (0.14) 0.94 502 0.01 (0.25) 20.27 (0.43) 0.99 (0.06) 0.05 (0.21) 0.87 503 0.32 (0.24) 20.20 (0.31) 0.92 (0.06) 0.09 (0.14) 0.88 504 0.47 (0.23) 20.16 (0.30) 0.87 (0.06) 20.26 (0.36) 0.81 505 0.43 (0.16) 0.40 (0.09) 0.92 (0.03) 0.18 (0.19) 0.94 486 0.40 (0.27) 0.20 (0.26) 0.91 (0.07) 1.04 (0.55) 0.88 28

m (0–4) 0.37 (0.16) 0.01 (0.09) 0.88 (0.05) 0.36 (0.22) 0.92 27

SPF1 20.03 (0.45) 0.11 (0.34) 1.00 (0.10) 0.38 (0.13) 0.86 402 20.62 (0.36) 0.09 (0.45) 1.12 (0.07) 20.09 (0.13) 0.85 403 20.29 (0.24) 0.19 (0.28) 1.05 (0.05) 0.02 (0.15) 0.90 404 1.05 (0.31) 0.23 (0.36) 0.81 (0.06) 0.31 (0.18) 0.80 39

m (1–4) 20.06 (0.27) 0.40 (0.33) 1.01 (0.06) 0.04 (0.22) 0.91 39

Notes:pC and pF denote commercial and Federal Reserve inflation forecasts;h and t index the horizon and date of theforecasts.M is the indicator of monetary-policy actions. Because the time within the month that the forecasts are made variesacross forecasters, the actual time and horizon subscripts for the inflation forecasts and the monetary-policy variable also varyacross forecasters; see text for details. The sample periods are 1984:2–1991:11 for Blue Chip; and 1974:8–1979:8 and1984:2–1991:11 for DRI and SPF. Numbers in parentheses are robust standard errors. The forecast horizonsm (0–4) andm(1–4) refer to the averages of 0 to 4 quarters ahead and 1 to 4 quarters ahead, respectively.


ters.27 All of the point estimates ofl are posi-tive, but only for one is thet-statistic over two.

The magnitudes of the estimated effects areclose to what one would expect given our previousfindings about the information content of policyactions. For the dummy variable, the averagepoint estimate for the average forecast revisionover the next four quarters is 0.13. This impliesthat a report in theWall Street Journalof a rise inthe federal funds rate raises commercial forecastsof inflation over the next year by between one-and two-tenths of a percentage point. For compar-ison, the corresponding figure from the informa-tion-revelation regressions in Table 7 is 0.16. Thisfigure implies that a contractionary monetary ac-tion signals that the Federal Reserve forecast isalso between one- and two-tenths of a percentagepoint above what one would expect given thecommercial forecast.

For the funds-rate target, the average pointestimate for the impact of a policy action oncommercial inflation forecasts up to four quar-ters ahead is 0.18. In the previous section, wefound that a rise of 100 basis points in thefunds-rate target signals a Federal Reserve fore-cast roughly 27 basis points above what onewould expect given the commercial forecast.Our results here therefore indicate that commer-cial forecasters change their forecasts by abouttwo-thirds of this amount.

D. Implications for the Impact of FederalReserve Actions on the Term Structure

The asymmetric-information hypothesis sug-gests that interest rates at all horizons rise inresponse to monetary contractions because mar-ket participants raise their expectations of infla-tion. In this section, we compare the predictedchanges in interest rates due to the changes inexpected inflation associated with monetary ac-tions with the observed changes in interest ratesat various horizons following monetary actions.This comparison provides a way of gaugingwhether the asymmetric-information effects

identified in earlier sections can explain a sub-stantial amount of the mysterious behavior ofthe term structure following monetary actions.We begin with the relatively straightforwardcase of short-term rates, and then turn to themore difficult case of long-term rates.

Of course, asymmetric information concern-ing inflation is not the only possible explanationof policy’s impact throughout the term struc-ture. For example, if inflation is very sluggish, acontractionary action stemming from a changein the Federal Reserve’s inflation goals couldimply long-lasting increases in real rates andonly very gradual decreases in inflation, andthus increases in long-term nominal rates. Sim-ilarly, there could be asymmetric informationnot about inflation, but about the equilibriumreal rate. That is, a contractionary action couldsignal that the Federal Reserve has informationthat the current and future real interest ratesconsistent with normal output are higher thanpreviously believed, and thus cause nominalrates to rise. Finally, long-term nominal ratesmight overreact to changes in short-term rates.That is, the rational expectations theory of theterm structure might fail.

Our goal is not to provide a complete analysisof how much these or other possible mecha-nisms contribute to the response of interest ratesto policy actions. Rather, we ask the narrowerquestion of how much of the response is con-sistent with the hypothesis that actions revealFederal Reserve information about inflation.Note, however, that the hypothesis based onsluggish inflation predicts that Federal Reserveactions should signal that its inflation forecastsare below those of commercial forecasters andthat commercial forecasters should revise theirexpectations of inflation down in response tocontractionary policy actions. Both predictionsare contradicted by our findings in subsectionsB and C of this section. And given the strongevidence in Section II of the existence of asym-metric information about inflation, the hypoth-esis that this asymmetry is central to the impactof policy actions on the term structure seems atleast as plausible as the alternatives based onasymmetric information about equilibrium realrates or on overreaction. Thus our hypothesisdeserves serious consideration as a candidateexplanation of the response of interest ratesthroughout the term structure to policy actions.

27 That is, we estimate regressions of the form:p h,t 1 1C

5 h 1 lMt 1 kp htC 1 r(p h,t 1 1

F 2 p htF ) 1 yh,t 1 1, where

bars over the variables indicate averages up to horizonh.The results reported are forh 5 4, but again they aresimilar for other horizons.


Short-Term Interest Rates.—Table 8, dis-cussed above, gives estimates of the impact ofpolicy actions on expected inflation as measuredby commercial forecasts in the current quarter andeach of the next four quarters. Finding the actions’impact on expected inflation at horizons up to ayear is thus just a matter of calculating the appro-priate averages from these estimates.

For simplicity, we assume that policy actionsoccur in the middle of a quarter. To compute theimpact of an action on expected inflation overthe life of a 3-month Treasury bill, we thereforeaverage the estimated impacts on expectedinflation in the current quarter and in thenext quarter. That is, we calculateDp3

e 5(l0 1 l1)/ 2, wherep3

e is expected inflationover a 3-month horizon andl0 and l1 are esti-mates ofl for h 5 0 andh 5 1 in equation (5)reported in Table 8. Similarly, we calculate anaction’s impact on expected inflation over thelives of 6-month and 12-month Treasury bills asDp6

e 5 (l0 1 2l1 1 l2)/4 and Dp12e 5

(l0 1 2l1 1 2l2 1 2l3 1 l4)/8.28

Table 9 reports the results. As before, wemeasure actions by both the dummy variableand the change in the funds-rate target. Forcomparison, the final column shows Cook andHahn’s estimates of the impact of a 100-basis-point change in the funds-rate target on theTreasury bill rate at that horizon.

The results suggest that policy actions have a

noticeable impact on expected inflation at shorthorizons. The results using the change in thefunds-rate target suggest that a 100-basis-pointincrease in the target raises expected inflationover the coming 3 months by about 20 basispoints, and over the coming 6 and 12 months byabout 10 basis points. The overall rise in Trea-sury bill rates for all three horizons is about 50basis points. Thus, between a fifth and almosthalf of the response of short-term rates to policyactions appears to reflect changes in expectedinflation. Not surprisingly, however, most of theresponse reflects changes in real rates.

The results using the dummy variable are sim-ilar. An increase in the funds-rate target raisesexpected inflation over the coming 3 months byabout 15 basis points, and over the coming 6 and12 months by about 10 basis points.29

Policy Actions and Expected Inflation atLonger Horizons.—Discerning the impact ofpolicy actions on expected inflation at horizons

28 We have no estimate of actions’ impact on the Surveyof Professional Forecasters’ expectation of inflation in thecurrent quarter. For both Blue Chip and DRI, the impact onexpected inflation in the current quarter is much larger thanthe impact in later quarters. For SPF, we therefore conser-vatively assume that the effect for the current quarter is thesame as the effect for the next quarter.

29 If a policy action becomes expected between the timeof the initial commercial forecast and the action itself, therevisions in expected inflation (and the consequent changesin interest rates) occur at the time the action becomesanticipated, not when it actually occurs. Cook and Hahn,however, consider changes in interest rates just on the daysof policy actions. Thus, the results here and in Table 11 mayoverstate how much of the changes in interest rates inresponse to policy actions stem from the revelation ofFederal Reserve information. During Cook and Hahn’s sam-ple period, however, policy actions were frequent and didnot take place at regular intervals, and often occurred attimes other than those of regular FOMC meetings. Moregenerally, policy actions were not subject to anything ap-proaching the degree of speculation that they are today. Asa result, the assumption that actions did not become ex-pected between the time of the previous commercial fore-cast and when they actually occurred appears to be areasonable first approximation.

TABLE 9—ESTIMATED IMPACT OF MONETARY-POLICY ACTIONS ON EXPECTED INFLATION AT SHORT HORIZONS

Forecast horizon(Months)

Dummy variable100-basis-point change

in funds-rate target Change inTreasurybill rateaBlue Chip DRI SPF Blue Chip DRI SPF

(Percentage points)3 0.09 0.21 0.28 0.22 0.49 0.11 0.556 0.07 0.13 0.23 0.17 0.22 0.10 0.54

12 0.06 0.09 0.21 0.16 0.00 0.14 0.50

a From Cook and Hahn (1989a, Table 3). These estimates show the effect of a 100-basis-point change in the funds-ratetarget on the relevant Treasury bill rate on the day of the change.


beyond a year is both more important and moredifficult. It is more important because the ac-tions’ impact on long-term rates is puzzling. Itis more difficult because we have little directevidence about the actions’ effect on expectedinflation beyond a year, and no direct evidenceat all about their effect beyond seven quarters.Thus, any estimates of their effect on expectedinflation at long horizons must be indirect.

One piece of indirect evidence comes fromthe behavior of inflation. If the Federal Reservehas a narrow target rate for inflation and bringsinflation back to that range rapidly after a de-parture, long-term expected inflation must beclose to the Federal Reserve’s target regardlessof what is happening in the near term. But if theFederal Reserve brings inflation back to its nor-mal level only slowly, market participants arelikely to revise their expectations of long-terminflation in response to news about short-terminflation.

The actual behavior of inflation is consistentwith the view that the Federal Reserve bringsinflation back to normal only slowly after ashock. Standard Box-Jenkins analysis suggeststhat inflation for our full sample period (1968IV–1991 IV) is well described as an IMA(1, 1)process. For example, both the Akaike andSchwarz criteria point to this specification. TheMA coefficient is20.42 (with a standard errorof 0.10). Thus, in response to a generic 1-per-centage-point innovation in inflation, expecta-tions of inflation in all subsequent quartersshould rise by 0.58 percentage points. Estimat-ing other low-order ARMA processes for thechange in inflation yields similar results. Fur-ther, as a check for the possibility of slow meanreversion, we also estimate AR-8 and AR-12processes for the change in inflation. We findthat these, too, suggest that inflation is verypersistent.30 And as we show below, it is themedium-term rather than the long-term behav-ior of inflation that is crucial for the impact ofpolicy actions on interest rates.

A second piece of indirect evidence comesfrom examining whether the Federal Reserve’smedium-term forecasts contain useful informa-tion about inflation at longer horizons. We ask

whether a market participant using a medium-term commercial inflation forecast in predictinginflation at longer horizons could improve onthat forecast if he or she had access to theFederal Reserve’s medium-term forecast.

Specifically, we consider an individual inmontht trying to forecast inflation from 4 quar-ters after montht to 8 quarters after, from 8 to12 quarters after, and from 12 to 16 quartersafter. We regress actual inflation over theseperiods on a constant, a commercial medium-term inflation forecast, and the comparable Fed-eral Reserve forecast. Our interest is in whetherthe coefficient on the Federal Reserve forecastis positive; that is, we want to know whether theFederal Reserve’s medium-term inflation fore-cast helps predict inflation two, three, and fouryears in the future.

We use the forecast in montht of inflationfour quarters after montht as our medium-terminflation forecast. This forecast is available for alarge number of observations for all of ourforecasters. As before, we correct the standarderrors for heteroskedasticity and for serial cor-relation overh 1 1 quarters. Thus, for example,when we consider inflation from between 8 and12 quarters ahead, we correct for heteroskedas-ticity over 13 quarters. One implication is thatthe standard errors should be interpreted withextreme caution: the justification for the stan-dard errors is asymptotic, and the forecast ho-rizons are substantial compared with our sampleperiods. This is especially true for the Blue Chipforecasts, where we have only 11 years of data.

Table 10 reports the results. For the shortBlue Chip sample, the results show little valuein the Federal Reserve forecasts for predictinginflation two to four years ahead. But for DRIand SPF, the estimates suggest that the FederalReserve forecasts contain considerable informa-tion. All of the point estimates are above one-half, and most are close to one. Moreover, thet-statistics (which, as indicated above, should beviewed as highly approximate) suggest thatmany of the estimates are statistically signifi-cant. Thus, the bulk of the evidence suggeststhat the Federal Reserve’s medium-term fore-casts contain useful information about longer-term inflation.

Long-Term Interest Rates.—The indirect ev-idence supports the view that monetary actions

30 Other authors also find that inflation is highly persis-tent. See, for example, Robert B. Barsky (1987).


affect inflationary expectations at fairly distanthorizons. We therefore want to go a step fartherand ask whether the magnitudes involved arelarge enough to account for much of the impactof monetary actions on long-term interest rates.Specifically, we want to extrapolate our findingsin Table 8 to longer horizons, and then estimateby how much actions’ effects on the path ofexpected inflation affect various long-termrates.

This exercise is clearly just a back-of-the-envelope calculation. We have no conclusiveevidence that policy actions affect expected in-flation at long horizons, and there are manypossible ways of extrapolating our findings forshort horizons to obtain quantitative estimatesof the effect at long horizons. Nonetheless, wethink it is useful to derive at least rough esti-mates of the likely effect of monetary actions onexpected inflation at long horizons and, thus, ofthe behavior of long-term interest rates associ-ated with the revelation of Federal Reserve in-formation.

We proceed as follows. For each commercialforecaster, we pool the different forecast hori-zons and reestimate the regressions underlyingTable 8 by nonlinear least squares, constrainingthe l’s to follow an AR-1 process. That is,loosely speaking, we fit an AR-1 process to theestimatedl’s in Table 8. Our estimates of the

l’s at horizons beyond those for which we havedirect evidence are just thel’s implied by theestimated process.

We then use thesel’s to find the impact ofpolicy actions on the interest rates on Treasurybonds of different maturities through their im-pact on the path of expected inflation. We ac-count for the fact that Treasury bonds are notpure discount bonds, so that changes in ex-pected inflation at short horizons have largereffects than changes at long horizons.31

The nonlinear least squares estimates implythat the effect of policy actions on expectedinflation are quite persistent. For Blue Chip,the immediate effect of a 100-basis-point risein the target is a rise in expected inflation of30 basis points, and the long-run effect is arise of 14 basis points. When the dummyvariable is used instead of the change in thetarget, the immediate effect of a tightening isa rise of 14 basis points, and the long-runeffect is a rise of 5 basis points. For SPF, the

31 Specifically, we consider a bond with an interest rateequal to the average for Cook and Hahn’s sample period(1974–1979) for that maturity. We assume that the termstructure is initially flat at that rate, and that it changes bythe change in the path of expected inflation. We then findhow this change affects the price of the bond and, thus, itsimplied yield.

TABLE 10—TESTS OFFEDERAL RESERVE ADDITIONAL INFORMATION FOR INFLATION AT LONG HORIZONS

pht 5 d 1 gCp4tC 1 gFp4t

F 1 nht


Blue Chip5–8 2.17 (0.72) 0.18 (0.14) 0.10 (0.07) 0.26 939–12 2.64 (1.06) 0.17 (0.31) 20.04 (0.20) 0.07 9213–16 2.93 (0.97) 0.14 (0.34) 20.10 (0.34) 0.02 84

DRI5–8 0.80 (1.22) 20.82 (0.49) 1.67 (0.57) 0.37 1469–12 1.84 (1.47) 20.59 (0.43) 1.19 (0.48) 0.18 14513–16 2.98 (1.58) 20.51 (0.29) 0.86 (0.36a) 0.08 137

SPF5–8 2.09 (1.36) 20.64 (0.29) 1.32 (0.27) 0.31 649–12 3.07 (1.69) 20.41 (0.43) 0.85 (0.52) 0.12 6413–16 4.26 (1.95) 20.41 (0.52) 0.60 (0.55) 0.04 61

Notes:p denotes inflation, andpC and pF denote commercial and Federal Reserve inflation forecasts;h and t index thehorizon and date of the forecasts. The sample periods are 1980:1–1991:11 for Blue Chip; 1970:7–1991:11 for DRI; and1968:11–1991:11 for SPF. Numbers in parentheses are robust standard errors.

a Standard error calculated using Newey-West procedure because the Hansen-Hodrick standard error cannot be computed.


estimates suggest virtually no time variationin the effect on expected inflation. Expectedinflation rises essentially permanently by 15basis points using the change in the target andby 21 basis points using the dummy variable.For DRI, the results vary depending on theindicator of monetary policy actions used.When policy actions are measured using thechange in the funds rate target, the estimatesimply that the impact of a 100-basis-point risein the target on expected inflation falls rap-idly from an increase of 82 basis points in thecurrent quarter, to essentially no effect twoquarters ahead, to a long-run decrease of 2basis points. However, when we use thedummy variable in place of the change inthe target, the results for DRI are similar tothose for the other two forecasters. The esti-mates imply that the immediate effect of apolicy tightening is a rise of 31 basis pointsand that the long-run effect is a rise of 10basis points.

Table 11 reports the effects on interest rateson bonds of different maturities through thischannel. As in Table 9, the final column givesCook and Hahn’s estimates of the effect of a100-basis-point change in the target on theyield of a Treasury bond of the correspondingmaturity. For the two cases where the esti-mated impact on expected inflation is essen-tially constant over time, interest rates at allhorizons rise by the amount of the rise inexpected inflation. For the other four cases,the estimated impact is a more complicatedfunction of the estimated effect on the path ofexpected inflation.

The results indicate that the impact of pol-

icy actions on expected inflation through therevelation of Federal Reserve informationmay account for much of the effect of policyactions on long-term rates. In particular, us-ing either Blue Chip or SPF, this channelaccounts for a rise of over 10 basis points inlong-term interest rates in response to a 100-basis-point rise in the funds-rate target. Thisrepresents over half of the overall response of5-year and 7-year bond rates found by Cookand Hahn, and essentially all of the responseof 10-year and 20-year rates. Because theeffects of a 100-basis-point rise in the funds-rate target are much less persistent for DRI,the results using the DRI forecasts suggestessentially no impact through this channel.However, when we use the dummy variable tomeasure policy actions, the results for allthree forecasters suggest that the actions havesubstantial effects on long-term rates.

These results do not depend on the actions’impact on expected inflation at very long hori-zons. Because Treasury bonds are not pure dis-count bonds, short-term and medium-termexpected inflation are more important to theirvalue than long-term expected inflation. For ex-ample, about half of policy actions’ estimatedimpacts on the 20-year bond rate shown inTable 11 stem from their effect on expectedinflation over the first six years, and an addi-tional quarter of the impacts comes from theireffect on expected inflation over the followingfive years. The relative unimportance of verylong-term expectations increases the plausibil-ity of our back-of-the-envelope calculation,since our estimates of the impact of informationrevelation on expected inflation are surely less

TABLE 11—ESTIMATED IMPACT OF MONETARY-POLICY ACTIONS ON EXPECTED INFLATION AT LONG HORIZONS

Forecasthorizon(Years)

Dummy variable100-basis-point change

in funds-rate target Change inTreasury

bond rateaBlue Chip DRI SPF Blue Chip DRI SPF

(Percentage points)3 0.05 0.11 0.21 0.15 0.04 0.15 0.295 0.05 0.11 0.21 0.15 0.02 0.15 0.217 0.05 0.11 0.21 0.14 0.01 0.15 0.19

10 0.05 0.11 0.21 0.14 0.01 0.15 0.1320 0.05 0.10 0.21 0.14 20.00 0.15 0.10

a From Cook and Hahn (1989a, Table 3). These estimates show the effect of a 100-basis-point change in the funds-ratetarget on the relevant Treasury bond rate on the day of the change.


speculative for the short and medium run thanfor the very long run.

IV. Conclusion

The most important finding of this paper isthat the Federal Reserve appears to possessinformation about the future state of the econ-omy that is not known to market participants.Our estimates suggest that if they had accessto the Federal Reserve’s forecasts of inflation,commercial forecasters would find it nearlyoptimal to discard their forecasts and adoptthe Federal Reserve’s. This information ad-vantage appears to exist for real output aswell as for inflation.

The existence of this information asymmetryhas important implications for the behavior ofinterest rates. The tests discussed above suggestthat Federal Reserve actions reveal some of itsadditional information and forecasters respondby changing their expectations of inflation. As aresult, the information revelation associatedwith monetary actions can explain why interestrates at even long horizons rise when the Fed-eral Reserve tightens policy.

Our finding of substantial asymmetric in-formation between the Federal Reserve andthe public may also have implications for avariety of other studies in monetary econom-ics. First, as mentioned in the introduction,many models of central-bank behavior em-phasize the potential importance of an infor-mation advantage for the monetary authority.For example, in models with rational expec-tations and flexible prices, activist monetarypolicy can stabilize real output only if themonetary authority has additional informationabout the state of the economy (Sargent andWallace, 1975; Barro, 1976). To give anotherexample, Barro and Gordon (1983), Canzo-neri (1985), and Cukierman and Meltzer(1986) argue that in settings where optimalmonetary policy is not dynamically consis-tent, asymmetric information between themonetary authority and the public about thebenefits of expansionary policy has importantimplications for the conduct of policy, themonetary authority’s desire for secrecy, andthe relation between economic conditions andpolicy actions. Our results bear on the impor-

tance of all of these models of asymmetricinformation.

Second, an even broader literature is con-cerned with the possibility of asymmetric infor-mation in financial markets, and of actionsproviding signals of that information. In thecase of inflation and interest rates, it is easy toidentify a participant that may have additionalinformation (the Federal Reserve) and one im-portant set of its actions (changes in its funds-rate target). Even more important, the FederalReserve and commercial inflation forecasts pro-vide a potential record of the informed party’sadditional information. As a result, this settingmay be particularly fruitful for investigatingthis general class of models. As we have de-scribed, in this case there is overwhelming ev-idence of the existence of asymmetricinformation and considerable evidence that ac-tions provide signals of that information andthat those signals are important to the actions’effects. This suggests that asymmetric informa-tion and signaling deserve serious considerationin the analysis of financial markets more gen-erally.

Finally, a large literature dating back to atleast Christopher A. Sims (1980) and Ben S.Bernanke and Alan S. Blinder (1992) attemptsto identify the effects of monetary policy byexamining the response of the economy to thecomponent of a policy instrument, such as thefederal funds rate, that is orthogonal to some setof publicly available information. Our resultssuggests that there is a fundamental problemwith this approach. The component of monetarypolicy orthogonal to publicly available informa-tion reflects not just random variations in pol-icy, but also the Federal Reserve’s responses toinformation that it has but the public does not.As a result, the estimates of policy’s effectsfrom this approach are contaminated with theeffects of the shocks that are causing thechanges in policy. Thus, the existence of a sig-nificant information advantage for the FederalReserve may have important implications notjust for the behavior of interest rates and theo-retical analyses of central-bank behavior, butfor a wide range of empirical investigations ofmonetary policy.

Our finding of substantial asymmetric in-formation between the Federal Reserve andthe public may also have implications for the


debate over Federal Reserve reporting prac-tices. The Federal Reserve could eliminate itsinformation advantage by releasing the GreenBook forecasts as soon as they are made.Provided that the information content of theGreen Book forecasts remained the same, im-mediate release would benefit all those whouse forecasts. Immediate release would presum-ably also increase the transparency of monetarypolicy-making by showing more of the motivationbehind FOMC decisions. It could thus have thebenefit of reducing the financial-market volatilityassociated with speculation about Federal Reserveactions and motives.

The obvious complication is that immediaterelease could change the information content ofthe Green Book forecasts. Given the prestige ofthe Federal Reserve and the forecasts’ impor-tance to monetary policy, the forecasts wouldsurely attract a great deal of attention if theywere released without delay. This could lead theFederal Reserve staff to change how it made theforecasts. For example, it could cause the staffto be less willing to depart from the consensusof private forecasts, to put more weight oneasily documented model simulations and lesson its judgments, or to report more optimisticforecasts.

Even if early release did not change how thestaff made the forecasts, it would take years,and perhaps decades, for the statistical record tobe long enough for this to be clear. Thus, re-gardless of whether early release changed theinformation content of the forecasts, the knowl-edge that the forecasts were being releasedcould cause users of the forecasts to have lessconfidence in them. In the case of monetarypolicy, if the FOMC became less sure of theforecasts’ value, it might place more weight onother, potentially less reliable, sources of infor-mation in policy-making. Thus, the finding thatthe Federal Reserve forecasts contain valuableinformation about future economic develop-ments is not enough to settle the question ofwhether it would be desirable for the FederalReserve to release those forecasts.

REFERENCES

Barro, Robert J. “Rational Expectations and theRole of Monetary Policy.”Journal of Mone-

tary Economics, January 1976,2(1), pp.1–32.

Barro, Robert J. and Gordon, David B. “Rules,Discretion and Reputation in a Model ofMonetary Policy.”Journal of Monetary Eco-nomics, July 1983,12(1), pp. 101–21.

Barsky, Robert B. “The Fisher Hypothesis andthe Forecastability and Persistence of Infla-tion.” Journal of Monetary Economics, Jan-uary 1987,19(1), pp. 3–24.

Bernanke, Ben S. and Blinder, Alan S.“TheFederal Funds Rate and the Channels ofMonetary Transmission.”American Eco-nomic Review, September 1992,82(4), pp.901–21.

Canzoneri, Matthew B. “Monetary Policy Gamesand the Role of Private Information.”Amer-ican Economic Review, December 1985,75(5), pp. 1056–70.

Cook, Timothy and Hahn, Thomas. “The Effectof Changes in the Federal Funds Rate Targeton Market Interest Rates in the 1970s.”Jour-nal of Monetary Economics, November1989a,24(3), pp. 331–51.

. “The Credibility of the Wall StreetJournal in Reporting the Timing and Detailsof Monetary Policy Events.” Federal ReserveBank of Richmond Working Paper No. 89-5,December 1989b.

Cukierman, Alex and Meltzer, Allan H. “A The-ory of Ambiguity, Credibility, and Inflationunder Discretion and Asymmetric Informa-tion.” Econometrica, September 1986,54(5),pp. 1099–128.

Ehrbeck, Tilman and Waldmann, Robert.“Why Are Professional Forecasters Biased?Agency versus Behavioral Explanations.”Quarterly Journal of Economics, February1996,111(1), pp. 21– 40.

Gordon, Robert J. Macroeconomics, 6th Ed.New York: HarperCollins, 1993.

Hansen, Lars P. and Hodrick, Robert J.“ForwardExchange Rates as Optimal Predictors of Fu-ture Spot Rates: An Econometric Analysis.”Journal of Political Economy, October 1980,88(5), pp. 829–53.

Lamont, Owen. “Macroeconomic Forecasts andMicroeconomic Forecasters.” National Bureauof Economic Research (Cambridge, MA)Working Paper No. 5284, October 1995.

Newey, Whitney K. and West, Kenneth D. “ASimple, Positive Semi-definite, Heteroske-


dasticity and Autocorrelation Consistent Co-variance Matrix.”Econometrica, May 1987,55(3), pp. 703–08.

Romer, Christina D. and Romer, David H. “WhatEnds Recessions?” in Stanley Fischer andJulio J. Rotemberg, eds.,NBER macroeco-nomics annual 1994. Cambridge, MA: MITPress, 1994, pp. 13–57.

Rudebusch, Glenn D.“Federal Reserve Inter-est Targeting, Rational Expectations, andthe Term Structure.”Journal of MonetaryEconomics, April 1995, 35(2), pp. 245–74.

Sargent, Thomas J. and Wallace, Neil.“ ‘Ratio-nal’ Expectations, the Optimal Monetary In-strument, and the Optimal Money SupplyRule.” Journal of Political Economy, April1975,83(2), pp. 241–54.

Scharfstein, David S. and Stein, Jeremy C.“Herd Behavior and Investment.”Ameri-can Economic Review, June 1990,80(3),pp. 465–79.

Sims, Christopher A. “Macroeconomics and Re-ality.” Econometrica, January 1980,48(1),pp. 1–48.


Federal Reserve Information and the Behavior of Interest …dromer_aer2000.pdf · Federal Reserve...

Documents

Transcript of Federal Reserve Information and the Behavior of Interest …dromer_aer2000.pdf · Federal Reserve...