Do Central Banks' Forecasts Take Into Account Public Opinion and Views?

BoardofGovernorsoftheFederalReserveSystem InternationalFinanceDiscussionPapers Number1080 May2013 Docentralbanks’forecaststakeintoaccountpublicopinionandviews? RicardoNunes NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author. Recent IFDPs are available ontheWebatwww.federalreserve.gov/pubs/ifdp/.

Do central banks’ forecasts take into account public opinion and views? RicardoNunes BoardofGovernorsoftheFederal ReserveSystem Firstversion:July2012;Thisversion:March2013 Abstract The Federal Reserve through the Federal Open Market Committee (FOMC) regularly releases macroeconomicforecaststothegeneralpublicandtheUScongresswiththepurposeofexplaining the likely evolution of the economy and the appropriate stance of monetary policy. Immediately beforedoingso,theFOMCreceivesaforecastproducedbytheFederalReservestaffwhichremains private for five years. The literature has pointed out that, despite the informational advantage of the FOMC, its forecast differs from and is not always more accurate than the staff forecast. This finding has raised concerns regarding the loss of relevant information and the usefulness of the FOMCforecasts.ThispaperbringsevidencethattheFOMCforecastalsoincorporatesotherpublicly availableforecastsandviews,andthattheweightattributedtopublicforecastsislargerthanwhatis optimalgivenameansquarederrorobjective.Thesefindingsareconsistentwithi)theinstitutional role of the FOMC in being representative of a variety of public views, ii) the academic literature recommendationtouseequalweightsandnottooverfitspecificforecastsbasedonpastperformance. Thestatisticalmodelcanalsoaccountforseveralempiricalregularitiesoftheforecasts. Keywords:monetarypolicydesign,centralbanks’forecasts. JELcodes:E52,E58,E31. Acknowledgments: I am grateful to Isabel Correia, Martin Ellison, Neil Ericsson, Jon Faust, Jordi Gali, Luca Guerrieri, Chris Gust, Thomas Laubach, Jesper Linde, David Lopez-Salido, Albert Marcet, CaterinaMendicino,DavidReifshneider,JohnRogers,PedroTeles,seminarparticipantsattheUniversity of Oxford, Federal Reserve Board, Universidade Catolica Portuguesa, Bank of Portugal, and several economists at the Federal Reserve Board for comments. John Weber, Walker Ray, and Erica Reisman providedsuperbresearchassistance.Theviewsexpressedinthispaperaresolelytheresponsibilityofthe authorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoardofGovernorsoftheFederal ReserveSystemorofanyotherpersonassociatedwiththeFederalReserveSystem. E-mail:ricardo.p.nunes@frb.gov

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 3 1. Introduction Economicprojectionsareoftenpresentedtothepublicanditspoliticalrepresentatives by central banks as part of their mandates. The public disclosure of forecasts is a common practice of central banks which are designed to be transparent and accountable to the democratic societies that they represent. These forecasts help society understand the likely evolution of the economy and the trade-offs that the societyfacesaswellasthepolicychoicesthatbestrepresentitsinterests. MorespecificallyintheUnitedStates,theFederalReserveregularlydisclosesto the public the forecasts produced by the Federal Open Market Committee (FOMC) – the body that sets the Federal Funds rate. The FOMC meetings devote substantial amountoftimepreparingtheseforecastsandtheydrawonconsiderableexpertiseand resources. The FOMC forecast is then heavily discussed and scrutinized in the U.S. Congress and the media. Given the importance of the FOMC forecast, both as part ofappropriatemonetarypolicyandasanaccountableandtransparentpolicy-making process of democratic societal choice, it is worthwhile to ascertain the factors that affectthisforecast. Romer and Romer (2008) and other subsequent papers show a puzzling result concerning the usefulness of the FOMC forecast. The staff of the Federal Reserve prepares the so-called Greenbook forecasts.1 The Greenbook forecast is prepared specifically for and one week prior to the FOMC meetings. This forecast constitutes a key input into the FOMC forecast, and the FOMC could simply adopt it without changes.TheFOMChasaninformationaladvantageasitreleasesitsforecastlaterand alsohasinsiderknowledgeoftheirownpreferencesregardingtheinterestratesetting. Romer and Romer (2008) find that despite these advantages, the FOMC forecast is not more accurate than the Greenbook forecast. Optimal predictions of inflation and unemploymentplaceazero,oraverylowweight,ontheFOMCforecast. ThispaperexaminesthedeterminantsoftheFOMCforecast.Inordertodoso,itis criticaltousetheGreenbookforecastasacontrolbecauseitispreparedjustbeforeand specifically for the FOMC. A key characteristic that I explore is the public exposure and the institutional role of the FOMC. The Greenbook forecast remains private for fiveyearsandisnotmandatedtofulfilanypublicrole.Followingthislineofthought, IfindthattheFOMCforecastseemstoreflectorincorporateotherpubliclyavailable forecastsandinformation,namelyIexploreboththeroleofprivatesectorandWhite Houseforecasts. The statistical model I estimate can account for several empirical regularities of the forecasts. More precisely, the statistical results can explain the puzzle of Romer andRomer(2008).ThemodelcanalsocapturethecasesinwhichtheFOMCproduces better forecasts than the Greenbook, as well as the cases in which it produces better forecaststhantheSPFandtheWhiteHouse. 1. The Greenbook has recently been merged with the Bluebook and relabeled “Tealbook”. For convenience,IwillcontinuetousethetermGreenbookthroughoutthispaper.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 4 TheWhiteHouseforecastdataareobtainedfromtheBudgetoftheUnitedStates Government and the Mid-Session Review. Private sector forecast data are available through the Survey of Professional Forecasters. Because all the variables in our regressionsareforecasts,thestatisticalframeworkisstraightforwardandinterpretable. There are a variety of reasons for which the FOMC forecast may reflect other publicly available information besides what is already contained in the Greenbook. Thispaperopensavenuesforfutureresearchintotheunderlyingmotivationsfordoing so, as several possible explanations are observationally equivalent. The goal of this paperisnottodistinguishbetweenthesebuttoestablishwhethertheFOMCdoesor doesnottakeintoaccountpubliclyavailableforecastsofrelevanteconomicagentsand institutions. Having said that, the results are consistent with the role assigned in the FederalReserveActtotheFOMC.2 TheFOMCisadiversecommitteeinstitutionally designed to represent the public and a variety of views, and the Greenbook could potentiallybeoneamongseveralviews.ThefindinginthispaperthattheFOMCdoes notdepartrandomlyfromtheGreenbook,andinsteadtakesitheavilyintoaccountbut alsoincorporatesotherpublicinformationisacomfortingresultfortheFOMCgiven itsmandate.Inaddition,theFOMCexplicitlyandpubliclycompareditsforecastswith the ones of the White House and the private sector both in public statements and officialdocuments,beingverytransparentindoingso. Ialsoexaminewhethertheweightsattributedtopublicforecastsreflectanoptimal pooling of information. I find that the weights on public forecasts are largerthan the optimalweightscomputedfromameansquarederror(MSE)perspective.According tothisperspective,thepublicforecastsconsideredseemtobeoverweighted.However, theforecastingliteratureoftenrecommendsgivingequalweightstorelevantforecasts rather than following a strict MSE perspective (e.g. Zarnowitz (1992), Clements and Hendry (1998, 2002)). The rationale is that the methods of each forecaster as well as the structure of the economy change frequently and it may be difficult to identifygoodforecastersbasedonpastperformance.Theliteratureshowsheresome tensionbetweenusingMSEorequalweights,andtheFOMCforecastseemstofollow good forecasting practice and draw on both – the weights estimated in the statistical regressionsareinbetweenbothapproaches. Severalpapershavecompareddifferentforecastsbyexaminingtheirsimilarityand ranking their forecasting performance (Romer and Romer (2000), Reifschneider and Tulip(2007), and Faustand Wright (2009) among manyothers). Studies focus often ontheinflationforecastbecauseitisthevariablemostrelatedtothemissionofcentral banks and for which they have specific expertise. Consistent with these arguments, there is a consensus that the Greenbook inflation forecast seems to be the most efficient.Otherpapershaveexaminedtheeffectsofthepublicopinion(Tootell(1999)) andregionalfactors(MeadeandSheets(2005))ontheFOMCmembers’policyvotes, and the patterns of forecast disagreement among individual members of the FOMC (BanternghansaandMcCracken(2009)).Noneofthesepapers,however,providesan 2. SeetheFederalReserveAct,Section4,Article10,11,12andSection10,Article1.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 5 explanation for the difference between the FOMC and Greenbook forecasts; to the bestofmyknowledgethepresentpaperisthefirsttoempiricallyaddressthisissue. It is beyond the scope of the present paper to examine welfare or normative questions. Such questions are as interesting as they are ill posed since they depend on the objective functions of society and the central bank. For instance, Ellison and Sargent(2012)provideatheoreticalexplanationwheretheFOMCbehavesoptimally. The authors arguethat the FOMC is a robustpolicymaker, as in Hansen and Sargent (2008).UnliketheGreenbook,theFOMCforecastsfollowtheprobabilitiesofrobust policy rather than aim to be optimal in a pure forecasting sense. In fact, the results foundherecouldalsobejustifiedbyclaimingthattheFOMCdidnotfollowapurely MSE forecasting perspective. The mechanisms pointed out here and in Ellison and Sargent (2012) are not contradictory but complementary. Most likely many factors playarole.Alatersectiondiscussestheinsightsthatcanbegainedfrombothpapers. The paper is organized as follows: Section 2 describes the forecasts and key statistics. Section 3 provides the main results and Section 4 addresses their implications. Section 5 presents robustness analysis, including the use of individual levelforecastsoftheFOMCparticipants.Section6concludes. 2. Theforecasts The FOMC forecast: The course of monetary policy in the US is decided at the FOMC meetings. The Humphrey-Hawkins Full Employment and Balanced Growth Act requires the Federal Reserve to submit to Congress twice a year a document – theMonetaryPolicyReport(MPR)–discussingtheconductofmonetarypolicyand the outlook on the economy. To this effect, preceding the MPR submission and in conjunction with FOMC meetings, each Federal Reserve Bank president plus seven membersoftheFederalReserveBoardsubmittheirforecasts. The range and central tendency of these forecasts are released to the public as part of the MPR and are discussed in a congressional hearing. I use the midpoint of the central tendency and, if not available, the midpoint of the range. A later section examinestheindividualresponsesoftheFOMCmembers. Starting in July 1979 the FOMC forecasts are prepared in February and July of eachyear.TheforecastsinFebruaryareforthecurrentyear;theforecastsinJulyare for both the current year and next year. The forecasts for inflation and real growth are for fourth quarter over fourth quarter. The forecasts for unemployment are for thefourthquarterlevel.Thedefinitionsoftheforecastedvariableschangedovertime, whichposesanextradifficultyinmatchingdefinitionsacrosssources.3 3. The initial inflation definition was GNP inflation, in February 1989 changed to CPI, in February 2000changedtopersonalconsumptionexpenditures(PCE),inJuly2004changedtocorePCE,andafter February2008changedtobothPCEandcorePCE.TheinitialrealgrowthdefinitionwasGNP,inFebruary 1992changedtoGDP.Theunemploymentdefinitiondidnotchange.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 6 The Fed staff forecast (Greenbook): The staff of the Board of Governors of the Federal Reserve System produces these forecasts one week before each FOMC meeting. All FOMC participants have access to them. The Greenbook takes five years to become public and the data sample in this paper observes that condition. AllvariablesforecastedbytheFOMCareforecastedbythestaff. TheWhiteHouseforecast:TheforecastsoftheWhiteHouseareincludedbecause theMPRfrequentlyshowstheseforecastsandcomparesthemdirectlywiththeFOMC forecast. These comparisons are present from the inception of the MPR until 2000.4 TheforecastsoftheWhiteHouseandtheAdministrationareinterconnectedwiththe Administrationpoliciesandgoals.TheAdministrationopinionontheevolutionofthe economy are disclosed to the public in a variety of formats: speeches, interviews, debates, articles, press releases, and so on. Even though all of these sources are importantforthispaper,oneneedstorelyonasystematicprocesstorecordtheWhite Houseforecasts. Thedata-setwasconstructedbymanuallycollectingtheforecastsfromtheBudget of the United States Government and the Mid-Session Review.5 Around January or February of each year, the budget for the subsequent fiscal year is presented. The forecastsforseveralyearsarecontainedinthechapter“EconomicAssumptions”.The tablewithforecastsalsocontainsanoteusuallyindicatingthatinformationonlyupto Novemberofthepreviousyearisused. ForthecorrespondingJulyforecastsoftheFOMC,IusetheMid-SessionReview of the Budget, which is available around June or July of each year. The “Economic Assumptions”tablealsocontainsanoteusuallyindicatingthatinformationonlyupto MayorJuneisused.Thetimingoftheforecastsisaddressedlaterintheanalysis. ThenumberofvariablesbeingforecastedislessthanthoseintheGreenbookbut morethanthoseintheFOMC.Forinflation,onecanmatchtheforecastsofinflation until July 1999.6 A later section extends the sample for the inflation results. For real growthandunemployment,onecanmatchthedefinitionsintheentiresample. The private sector forecast: Comparisons with private sector forecasts are never present in the MPR but occasionally appear during the testimony and hearings. For instance,testifyingattheSenateBankingCommitteeon14February2008Chairman 4. The MPRs are available at http://fraser.stlouisfed.org/historicaldocs/680/. After 2000 explicit comparisonsarenolongerpresent,asthedefinitionsofvariablesbeingforecastedbytheWhiteHouse andFOMCnolongercoincidesexactly.Alatersectionshowsthattheresultsstillholdafter2000. 5. TheseforecastsgatherinputsfromtheCouncilofEconomicAdvisors,DepartmentoftheTreasury, OfficeofManagementandBudget,andalsofromtheDepartmentofCommerce(BureauofEconomic AnalysisandEconomicsandStatisticsAdministration),andtheDepartmentofLabor(BureauofLabor Statistics). 6. Unfortunately,thebudgetdocumentsdonotcontainPCEinflation,thedefinitionoftheFOMCafter 2000asexplainedinfootnote3.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 7 Bernanke referred that the forecasts to be released in the following weeks were “reasonablyconsistent”withprivate-sectorforecasts.7 I use the Survey of Professional Forecasters (SPF) because of several reasons. First, it is the oldest survey of macroeconomic forecasts in the United States conductedataquarterlyfrequency.Second,therespondentsproduceregularforecasts of economic variables as part of their routine tasks in the business world and Wall Street.Third,thissurveycontainsforecastsforseveralvariablesatdifferenthorizons thatallowustomatchtheFOMCdefinitions.Alateranalysisshowsresultswiththe BlueChipforecasts. Forinflation,onecanmatchtheFOMCinflationforecastuntilJuly1999.8Forreal growthandunemploymenttheentiresampleofFOMCforecastsismatched.TheSPF forecasts from the fourth quarter of the preceding year and the second quarter of the corresponding year match the February and July FOMC forecasts, respectively. The professional forecasters need to submit their responses at late in the second or third week of the middle month of each quarter.9 The results of the survey are released to thepublicaroundthefourthweekofthemiddlemonthofthequarter. Outcomes:Foractualoutcomes,ItrytomatchwhatthestaffandtheFOMCwere trying to forecast. For NIPA variables I use the final estimates released after three monthsandfornon-NIPAIusethedataasoriginallyreleased. 2.1. Preliminaryanalysis Mean Squared Errors: The results are particularly relevant for the inflation forecast. First, the Greenbook inflation forecast seems to be quite accurate. Second, the Fed has specific expertise in monitoring and forecasting inflation – unlike real growth and unemployment, which are affected to a greater extent by other policies and institutions. Third, as shown in Svensson (1997, 1999), the modern framework of inflation targeting can be implemented and depends crucially on the evolution and monitoringoftheinflationforecast. Table 1 shows the Mean Squared Errors (MSE) for the updated sample. The Greenbook performs quite well in this dimension. With respect to inflation, the Greenbook outperforms the others. This result is not surprising given that the main mandateofcentralbanksistostabilizeand,therefore,forecastinflation.Also,inflation is mainly a monetary phenomenon and central banks should have a very good sense offuturedevelopmentsinthatarea. With respect to unemployment and real growth, Table 1 does not dictate a clear winner. The forecasts are similar and none is found to perform much better. Unlike 7. ThecorrespondingMPRisdated27February2008;thereaderisreferredtoTheWallStreetJournal articleReddy(2008). 8. IntheSPF,PCEandcorePCEinflationareonlyavailableafter2007,whichdoesnotallowtomatch thedefinitionsoftheFOMCimmediatelyafter2000asexplainedinfootnote3. 9. This timing always allows the participants to have access to the advance report of the Bureau of EconomicAnalysis.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 8 inflation, real growth and unemployment depend to a greater extent on policies and factors outside the Fed’s control. Unsurprisingly, others can forecast those two variablesequallywell. TABLE1. MeanSquaredErrors WhiteHouse MSE G F W Fˆ rank{G,F,Fˆ} rank{W,F,Fˆ} Inflation 0.7570 0.9331 1.2649 0.8609 X X Unemp. 0.4539 0.4744 0.6109 0.4787 X X Realgrowth 1.7870 1.6884 2.4482 1.8374 X SPF MSE G F S Fˆ rank{G,F,Fˆ} rank{S,F,Fˆ} Inflation 0.6658 0.8157 1.3629 0.7748 X X Unemp. 0.4015 0.3895 0.5087 0.3943 X X Realgrowth 1.7901 1.6181 1.8501 1.6907 X X Notes:ThetablereportsMeanSquaredErrorsforGreenbook(G),FOMC(F),WhiteHouse(W),SPF(S),andFOMC predictedbythemodelinsection3(Fˆ).Thesampleisthesameforallvariablesineachpanel.Theupperandlower paneldisplaydifferentvaluesevenforcommonseriesbecausethesampledoesnotcoincideassomevaluesfortheSPF forecastscannotbeconstructedintheearlierpartofthesample.ThefourthcolumndisplaysacheckmarkiftheMSE rankingofFandGisreplicatedbyFˆandG;thefifthcolumnisanalogoustothefourthbutconsiderstheMSEranking relativetopublicforecastsratherthanG. Average Forecast: The upper panel of Table 2 shows the averages of the Greenbook (G), FOMC (F), and White House forecasts (W), and the outcome t t t of the variables being forecasted. The fourth column displays a checkmark if the FOMC forecast is in between the Greenbook and the White House. The FOMC averageforecastisalwaysinbetweentheothertwo.Thispatternispresentevenwhen disaggregatingtheforecastsbydateandhorizon.ThelowerpanelofTable2performs thesameanalysiscomparingtheGreenbook,FOMC,andSPFforecasts(S).Thesame t patternsarepresent. Even though, the average forecast of F, S, and W are similar, there is variability over time that can be explored. Table A.1 in the appendix shows the standard deviations of the difference between two forecasts. For instance, for real growth s(W (cid:0)G)isequalto0.8. t t SignPredictions:Table3examineswhetherhavingaccesstotheGreenbookand either the SPF or the White House forecast would help predict whether the FOMC forecastwouldbehigherorlowerthantheGreenbook.IftheFOMCbasesitsforecast on the Greenbook but also incorporates some information from the SPF, one would think that (S (cid:0)G)>0 should imply (F (cid:0)G)>0, and (S (cid:0)G)<0 should imply t t t t t t (F (cid:0)G)<0. Even with such a direct approach, Table 3 shows that the success rate t t

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 9 TABLE2. AverageForecastComparison WhiteHouse Average G F W Inbetween? Outcome Overall Inflation 4.1731 4.3054 4.3887 X 3.9319 Unemp. 6.2318 6.1685 6.0965 X 6.0694 RealGrowth 2.6178 2.6882 2.8553 X 2.7303 Inflation Jan.Cur. 4.1295 4.2312 4.3550 X 3.8712 Jul.Cur. 4.2093 4.3512 4.4000 X 4.1093 Jul.Nex 4.1785 4.3304 4.4095 X 3.8121 Unemp. Jan.Cur. 6.3000 6.2231 6.2222 X 6.1222 Jul.Cur. 6.2393 6.2228 6.1571 X 6.1143 Jul.Nex 6.2393 6.1451 5.9893 X 6.0750 RealGrowth Jan.Cur. 2.5416 2.6389 2.6741 X 2.7552 Jul.Cur. 2.5400 2.5804 2.6357 X 2.6911 Jul.Nex 2.7202 2.7946 3.2250 X 2.7410 SPF Average G F S Inbetween? Outcome Overall Inflation 3.9657 4.0900 4.2877 X 3.7416 Unemp. 6.1825 6.1291 6.1153 X 6.0000 RealGrowth 2.6431 2.7094 2.7414 X 2.8354 Inflation Jan.Cur. 4.1295 4.2312 4.4549 X 3.8712 Jul.Cur. 4.2093 4.3512 4.3983 X 4.1093 Jul.Nex 3.4996 3.6285 3.9728 X 3.1686 Unemp. Jan.Cur. 6.3000 6.2231 6.2146 X 6.1222 Jul.Cur. 6.2393 6.2228 6.1641 X 6.1143 Jul.Nex 5.9920 5.9225 5.9534 5.7400 RealGrowth Jan.Cur. 2.5416 2.6389 2.6774 X 2.7552 Jul.Cur. 2.5400 2.5804 2.7183 X 2.6911 Jul.Nex 2.8680 2.9300 2.8363 3.0835 Notes:ThetablereportstheaverageforecastoftheGreenbook(G),FOMC(F),White House (W), SPF (S), and the outcome. The fourth column displays a check if the FOMCforecastisinbetweentheothertwo.Thesampleisthesameforallvariables ineachpanel.Theupperandlowerpaneldisplaydifferentvaluesevenforcommon seriesbecausethesampledoesnotcoincideassomevaluesfortheSPFforecastscannot beconstructedintheearlierpartofthesample.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 10 is above 75% when predicting the direction of the FOMC forecast relative to the Greenbook. The remaining columns in the table show that the results are robust to only incorporating predictions in which (S (cid:0)G) are above or below a certain threshold. t t Taking as an example the unemployment forecast, the last column shows that when (S (cid:0)G)isbelowthe25thorabovethe75thpercentile,in92%ofallobservationsone t t cancorrectlypredictthedirectionoftheFOMCforecastrelativetotheGreenbook. TABLE3. PercentageofCorrectSignPredictions PredictingSign PredictingSign PredictingSign percentiles(40-60) percentiles(25-75) W S W S W S Inflation 75.9259 75.4717 73.4694 73.3333 76.6667 75.0000 Unemp. 81.0811 83.7838 85.0000 83.8710 92.3077 88.3721 RealGrowth 76.7123 76.9231 78.9474 77.4194 78.5714 82.0513 Notes:Thetablereportsthepercentageofcorrectpredictionsofthesignof(Ft (cid:0)Gt).Thefirstcolumn plotsthepercentageofcorrectpredictions(Wt (cid:0)Gt)>0and(Wt (cid:0)Gt)<0implying(Ft (cid:0)Gt)>0 and(F(cid:0)G)<0,respectively.Thethirdcolumncomputesthepercentageofcorrectpredictionswith (Wt (cid:0)Gt)>(Wt (cid:0)Gt)60thpercentileand(Wt (cid:0)Gt)<(Wt (cid:0)Gt)40thpercentileimplying(Ft (cid:0)Gt)>0 and(Ft (cid:0)Gt)<0,respectively.Thefifthcolumndoestheanalysiswiththe25thand75thpercentile. Thesecond,fourth,andsixthcolumnmakethesameanalysisfortheSPFforecast. 3. MainResults The analysis in the previous section was indicative of the channels being examined. Thissectionperformsamoreformaleconometricevaluation.Themainspecification examines why the FOMC deviates from the Greenbook forecast instead of just adopting it. In other words, the question is whether movements in the dependent variable (F (cid:0)G) reflect other forecasts. To this effect, the two equations below are t t estimatedseparately: (F (cid:0)G)=a+b(W (cid:0)G) (1) t t t t (F (cid:0)G)=a+b(S (cid:0)G): (2) t t t t Table 4 presents the results. The equations are estimated with ordinary and weighted least squares. The WLS regression captures the different timings of the forecasts.Newey-WeststandarderrorswiththreelagsarereportedwhenusingWLS. A constant is included in the regressions because the average forecast was already examined in Table 2. Here, a crucial and stricter point is being tested, whether timevariationsintheFOMCforecastreflecttime-variationsinpublicforecasts. TheresultsshowthatindeedthedifferencebetweentheFOMCandtheGreenbook can be explained by, or is correlated with, the White House and the SPF. In the inflationforecasttheweightputonthenon-Greenbookforecastisroughly0.25.The

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 11 TABLE4. Regressionresults constant W S R2of(F-G) R2ofF Inflation W OLS 0.0791 (0.0333) 0.2466 (0.0527) - 0.2673 0.9878 WLS 0.0783 (0.0484) 0.2519 (0.0677) - 0.2608 0.9878 S OLS 0.0569 (0.0382) - 0.2093 (0.0574) 0.1892 0.9856 WLS 0.0443 (0.0722) - 0.2420 (0.1136) 0.2526 0.9856 Unemployment W OLS -0.0076 (0.018) 0.4114 (0.0454) - 0.4971 0.9874 WLS -0.0005 (0.0225) 0.4026 (0.063) - 0.4748 0.9873 S OLS -0.0272 (0.0165) - 0.3897 (0.0397) 0.5530 0.9888 WLS -0.0256 (0.0191) - 0.3882 (0.0656) 0.5408 0.9888 RealGrowth W OLS 0.0005 (0.031) 0.2945 (0.0372) - 0.4307 0.9605 WLS -0.0021 (0.0412) 0.2942 (0.0333) - 0.4226 0.9605 S OLS 0.0417 (0.0312) - 0.2505 (0.0423) 0.3100 0.9602 WLS 0.0367 (0.0419) - 0.2409 (0.0715) 0.2920 0.9602 Notes:Thetablereportstheestimatesofequations(1)-(2).WandSdenoteWhiteHouseandSPFforecasts,respectively. TheequationsareestimatedbothwithOLSandWLS.Newey-Weststandarderrorswiththreelagsarereportedinthe WLSregression. weightgoesupslightlyforrealgrowth,andregardingunemploymenttheweightgoes up to roughly 0.40. In all specifications the coefficients are statistically significant. TwoR2 measuresarereported,thefirstrelatedtoexplaining(F (cid:0)G),thesecondin t t explainingF.Obviously,explaining(F (cid:0)G)isadauntingtaskasthetwoforecasts t t t areformulatedwithinaoneweekinterval.Despitethesedifficulties,theeconometric modelpresentedhereisstatisticallysignificant. 3.1. Discussionofresults It is not the claim of this paper that the results in Table 4 are causal. Causality and correlationmayleadtoobservationallyequivalentFOMCforecasts.Forinstance,the FOMC mandate of representing the public may lead the FOMC to actively want to understand some public views regarding the evolution of the economy, and to incorporatetheviewsthatitagreeswith.ButitcouldalsobethattheFOMCalready hadsharedviewswithotherforecasters.Reversecausalityseemslessplausiblegiven our results. Below I discuss the timings of the forecasts and I present additional evidenceonthisissue. Discussionoftimings: ItisalwaysthecasethattheGreenbookiscompletedroughlyoneweekbeforethe FOMC meetings. The SPF is a quarterly survey forecasting several periods ahead. I chose the quarter in which the forecast is completed such that the SPF forecast

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 12 is finished and publicly available before the Greenbook and the FOMC forecasts.10 Thistiminghelpstoestablishthatitisthevariable(F (cid:0)G)reflectinginformationin t t (S (cid:0)G) instead of the opposite.11 Also, this choice of timing for the SPF puts this t t forecastataninformationaldisadvantage,makingithardertoobtaintheresultclaimed herebecausetheFOMCshouldputevenmoreweightontheGreenbook. Regarding the White House forecast, the dates of the documents indicate that it iscompletedbeforetheGreenbook.Namely,theEconomicAssumptionstablesinthe budgetdocumentshaveafootnoteindicatingthatonlyinformationuptoacertaindate isincorporated.ThatdateisbeforetheGreenbook.Thistiminghelpstoestablishthat itisthevariable(F (cid:0)G)reflectinginformationin(W (cid:0)G)insteadoftheopposite. t t t t The results also suggest that reverse causality is much less likely. For instance, the FOMC mean forecast is in between the White House and the Greenbook mean (Table 2). Such evidence is at odds with the explanation that it is the White House incorporating the other two forecasts. The same pattern can be found in the results of Table 4. If one postulates that the variable (W (cid:0)G) reacts to (F (cid:0)G) then it t t t t does so with a coefficient larger than one. Such reasoning is implausible because it wouldimplythatif,forinstance,theGreenbookforecastsinflationtobe3%andthe FOMC forecasts 3.5%, then the White House would extrapolate and forecast 4%. While extrapolative models of inflation are not necessarily wrong, it is hard to argue that such behavior is likely or optimal for the US inflation time series. The timings andstatisticalevidencedonotseemtosuggestreversecausality. Furtherevidence: Even if the White House forecast is completed before the Greenbook, it may happenthatitisonlypublishedafter.ForthatreasononecannotclaimthattheFOMC hadknowledgeoftheforecastintheBudget.Also,insomeyearsthedatesinthethe EconomicAssumptionstablesaremissingorareindicativeonly.Thisissueisinfact notproblematic.TheforecastsandviewsoftheAdministrationarediscussedwidely in the media and in policymaking circles. A strong evidence of this claim is that the MPRmakesreferencetotheWhiteHouseforecastsevenintimeswhentheywerenot publicly available through the Budget, as they were available through the media and othersources. For instance, the MPR dated 20th of July 1983 specifically shows the White HouseforecasteventhoughtheMid-SessionReviewoftheBudgetwasonlyreleased on the 25th of July 1983. Another example is contained in the MPR of July 1993, “The Administration has not yet released the midyear update to its economic and budgetary projections. However, statements by Administration officials suggest that 10. Thestudiesexaminingtheaccuracyofdifferentforecaststrytouseforecastsformedatroughlythe sametimeperiod.Thattreatmentoftimingscanbeproblematicforthecurrentstudy.Takingasanexample theFebruaryFOMCmeetings,ifIusedtheSPFforecastsformedinthefirstquarteroftheyearratherthan thelastquarterofthepreviousyear,theSPFparticipantscouldalreadyhaveaccesstotheFOMCforecasts andblurtheresults. 11. ThisresultisstillconsistentwiththeFOMCforecasts(andmonetarypolicyactions)influencingthe SPFforecastsatalaterdate.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 13 the revised forecasts for real growth and inflation in 1993 and 1994 are not likely to differsignificantlyfromthoseoftheFederalReserve.” In addition to this evidence, further analysis is presented to clarify the results. OnemaybeconcernedthattheFOMCmaynotknowexactlyandineveryperiodthe White House forecast contained in the Budget. This issue translates in econometric termsintoaproblemofmeasurementerror.Alsoonewouldliketoobserveperfectly the perception of the FOMC regarding the White House forecasts as well as the FOMC perception of other Administration signals. Instead, one observes only the WhiteHouseforecastscontainedintheUSbudget.Asitiswellknown,measurement errorcreatesadownwardbiasintheestimates.ThefactthattheestimatesinTable4 aresignificantis,therefore,supportiveofthechannelsproposedinthispaper. Table5presentsfurtherevidence.Iuseaknowncorrectionformeasurementerror in regressions with one variable. The method of group averages was first advocated byWald(1940)andisdescribed,forinstance,inGreene(2000).Thismethodusesan instrumentalvariablebasedongroupsoftheoriginalvariable.Forexample,withthree groups one creates the instrumental variable -1,0,1 if the variable is below the 33th percentile, in between, or above the 67th percentile, respectively. The table reports theresultswhentwoandthreegroupsareused,inbothcasestheinstrumentsarenot weak.12 Themainresultsstillhold. TABLE5. FurtherEvidence constant W Firstst.F-stat. Inflation IV 2groups 0.0651 (0.0354) 0.3119 (0.0776) 50.7000 3groups 0.0748 (0.034) 0.2669 (0.0664) 94.1480 Unemployment IV 2groups -0.0044 (0.0186) 0.4350 (0.0589) 115.6200 3groups 0.0004 (0.0183) 0.4702 (0.0523) 250.6580 RealGrowth IV 2groups 0.0038 (0.032) 0.2806 (0.0533) 74.9500 3groups 0.0065 (0.0314) 0.2695 (0.0463) 143.1370 Notes:Thetablereportstheestimatesofequations(1)-(2).Thefirstpanelusesinstrumentalvariableswith themethodofgroupaverages. Summaryoffurtherevidence: 12. Theresultsarealsorobusttorunningtheregressionwhenthenumberofgroupsequalsthenumber ofobservations,inwhichcasetheinstrumentalvariableistherankingoftheobservations.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 14 Making causal arguments in economics is extremely hard. Given that the data is not a pure time-series, it is impossible to use Granger causality and similar tests. Such tests are also subjective once forward looking variables are involved, as is the case here. Having said that, the results do not suggest that the causality runs in the opposite direction for three reasons. First, the timings of forecasts are such that the non-Fed forecasts are completed before the FOMC meetings. Second, the estimated coefficientsalsoconfirmtheclaimsinthepaper,otherwiseonewouldneedtoappeal to extrapolative forecasting in the US. Third, the instrumental variable regressions providefurtherevidence. With this evidence at hand one should also clarify that the results do not dismiss the explanations of Ellison and Sargent (2012). One can argue that the forecasts are independent and the regression coefficient b is capturing the similarity between the robustnessdegreeoftheagentsproducingtheforecasts. Ifthisistheexplanation,thenthispaperisstillsurprising.First,theWhiteHouse would also have to behave as a robust policymaker, instead of trying to be more optimistic in order to win the elections or push certain political agendas. Second, a crucial observation is that the private sector does not decide on monetary policy and hasnoincentivetopublishpolicymakingrobustforecasts.Third,onecouldclaimthat robust forecasting is the norm for all agents; if that is so it would be interesting to examine why the Fed staff is so different and what determines different degrees of robustness. 4. AdditionalResults 4.1. ReplicatingMSE Inowexaminewhetherthestatisticalmodelcanaccountfortheobservedpatternsof the MSEs. To examine this issue, one can first compute the FOMC forecast implied by the model, denoted as Fˆ. In other words, Fˆ is the fitted values of F based on the OLSestimatesofequations(1)and(2). The MSE of Fˆ can gauge the improvement or deterioration in the MSE caused bythechannelsmodeledinthispaper.Themodelissuccessfulalongthisdimension. For all cases except one, Table 1 shows that the ranking of MSEs of F relative to G is replicated by the ranking of MSEs of Fˆ relative to G (described in second to last columninTable1).Forinstance,theMSEoftheinflationforecastFˆ isalwaysworse thantheGreenbook.ThemodelcanalsoexplaincasesinwhichtheFOMCperforms better than the Greenbook; also the last column in Table 1 shows that the model can capturethebetteraccuracyoftheFOMCrelativetotheotherpublicforecasts. 4.2. Forecastaccuracyandoptimalweights Anaturalquestionarises:isitoptimalfortheFOMCtoincorporatetheWhiteHouse and SPF forecasts? Part of the answer can be seen in Table 1. The FOMC inflation

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 15 forecast is worse than the Greenbook and therefore, if MSE forecast accuracy were theobjective,the FOMC wouldhavebeen better offby just adopting the Greenbook inflation forecast. However, the pooling of forecasts usually leads to an improved forecastbutdeterminingtheoptimalweightsmaynotbestraightforward. An important step is to compare the weights actually used (b) with the optimal weights. This comparison allows one to analyze more carefully whether the FOMC reflectstoomuchortoolittlethenon-Greenbookforecasts.Optimalityhereisdefined from a forecasting perspective, the weights that would minimize the MSE of the FOMCforecast. Table 6 reports the optimal weights, which are directly comparable with the weights b in Table 4. For all cases except one, the weights on the non-Fed forecasts are larger than the optimal ones. This result shows that for all variables – inflation, unemployment, and real growth – publicly available forecasts are overweighed. For instance, for inflation the weight on the White House forecast is 0.25 whereas the optimalweightis0. Onepossiblereasonfortheoverweightofthenon-Greenbookforecastsrelatesto the institutional design of the FOMC. The FOMC is a plural body that is designed to represent a variety of views, and in that sense it is comforting to know that the differencebetweentheFOMCandGreenbookisnotnoise,andinsteadreflectsother publicviews.Therearealsootherpotentialreasonsbasedonforecastinggroundsthat Iinvestigatebelow. First, I investigate if the Greenbook was a poor forecast at the beginning of the sample, in which case the FOMC could have taken substantial time to learn that the Greenbook is accurate. Such patterns can occur in a model of learning as in Marcet and Sargent (1989) and Marcet and Nicolini (2003). Figure 4.2 shows the evolution overtimeoftheoptimalMSEweight–therecursiveestimatesusingdatauptothedate inthehorizontalaxis.13 Thefiguresshowthattheoptimalweightsonpublicforecasts vis-à-vis the Greenbook were always low or converged quite fast to a low number. Therefore,apotentially“badstart”oftheGreenbookforecastdoesnotseemtobethe reasonforoverweighingpublicforecasts. Second,Iinvestigateiftheforecastingliteratureadvocatesotherapproachesrather than using optimal MSE weights.14 Indeed, there are several approaches advocated, andfittingMSEweightsissometimesnotrecommendedbecauseofseveralarguments (e.g.Zarnowitz(1992),ClementsandHendry(1998,2002)amongothers). Onereasonisthatforecastersorforecastschangetheirmethodsandviewsoften; alsothestructureoftheeconomychanges,andagoodmethodforatime-periodmay not necessarily be valid for another time-period. Hence, the success of a particular forecast may be occasional and fortuitous or intuitive. And a particular’s forecast 13. Theappendixshowsthefigureswheretheconstantisomittedfromtheregressionsandtheweightb isrestrictedtobebetween0and1.Theresultsaresimilar. 14. Excluding the constant or minimizing the MSE in the 10% largest errors leads to the same conclusions.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 16 TABLE6. OptimalWeights constant W S R2 Inflation W -0.2399 (0.1148) -0.0062 (0.1816) - 0.8499 S -0.1723 (0.1179) - -0.1611 (0.177) 0.8434 Unemployment W -0.1833 (0.076) -0.1550 (0.1914) - 0.8083 S -0.1742 (0.0694) - 0.1235 (0.1666) 0.8115 RealGrowth W 0.1164 (0.1525) -0.0162 (0.1831) - 0.4382 S 0.1513 (0.1472) - 0.4176 (0.1995) 0.4222 Notes:Thetablereportstheweightsthatwouldbeoptimaltouseinthefullsample. record may not be reliable as a basis for inferences on how it will perform in the future.Whennomodelcoincideswithanon-constantdatagenerationprocess(DGP), ClementsandHendry(2002)showthataveragingmaythendominateoverestimated weightsinthecombination.15 In results not shown, the optimal MSE weights do show some variability. While excluding the Volker disinflation alone does not lead to significant changes, there is increased uncertainty in the optimal MSE weights after 1990. Indeed, after 1990 the White House inflation forecast is slightly under-weighted. This literature and results suggestthattheremaybeperiodsinwhichtheFOMCforecastissuperiororinferior totheGreenbook.Andthatrankingmaylargelydependonhowperfectorimperfect otherpublicforecastsandviewsare,aswelltheiroptimalweight,whichisnoteasyto determineex-ante. TheforecastingliteratureshowsthatthereisatensionbetweenusingoptimalMSE weights based on past record versus using equal weights among several forecasts. Bothapproacheshaveadvantagesanddisadvantages,andcanbedefendedorcriticized accordingly.Interestingly,itseemsthattheweightscharacterizingtheFOMCforecast are in between these two approaches. For five cases out of six, the weights shown in Table 4 are in between the optimal MSE weights shown in Table 6 and the equal weightsof0.5.16 15. Anotherreasonrelatestoherding,whereforecasterstrynottodiffertoomuchamongthemselves. TheGreenbookistheonlyforecastthatcannotbemadeavailableinreal-time.Accordingly,theFOMC forecastmayattachlessweighttotheGreenbookthantopublicforecastsbasedonaherdingmotive. 16. Alatersectionshowstheresultswhencombiningmorethantwoforecasts.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 17 Figure1. Optimalweightsb. White House SPF 0 0 −0.5 −2 −1 −4 1979 1984 1989 1994 1999 1979 1984 1989 1994 1999 Inflation Inflation 3 0 2 −0.5 1 −1 0 −1.5 1979 1984 1989 1994 1999 2004 1979 1984 1989 1994 1999 2004 Unemployment Unemployment 0 5 −2 0 −4 −5 −6 1979 1984 1989 1994 1999 2004 1979 1984 1989 1994 1999 2004 Real Growth Real Growth Notes:Thefigureplotstheoptimalweightsinequations(1)-(2)usingOLS.Thegraphsontheleftandright refertotheWhiteHouseandSPFforecasts,respectively. 4.3. Replicationofregressionspredictingactualvalues Thissubsectionexamineswhetherthestatisticalmodelisabletocapturesomestylized factsintheliterature.InordertodeterminetheforecastingpoweroftheFOMCrelative totheGreenbook,thefollowingregressionisusuallyestimated: X =a+bG +cF +e; (3) t t t t

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 18 whereX istheoutcome.ForinflationRomerandRomer(2008)findthattheweight t onF issmallornegative.Forunemploymentthesamehappensbuttoalesserextent, t andforrealgrowththeweightsshowthatnoneoftheforecastshasaclearadvantage.17 Table 7 examines whether the model can account for the findings of Romer and Romer (2008). The table displays their original regression using the updated sample of this paper (first two rows in the table). Then it substitutes the forecast F by the t forecast as predicted by the models (Fˆ). The weights on Fˆ and G broadly replicate t t t thepatterns of theweights on F and G;namely theweight on Fˆ issmall and lower t t t thantheweightonG. t 5. Robustness 5.1. Extendingtheinflationregressionandsub-sampleanalysis Because the Greenbook inflation forecast seems to be better than the other non-Fed forecasts, the results gain extra interest with respect to this variable. However, the change in definitions of the FOMC inflation forecast limited the analysis until July 1999, while for the other variables one can use all the data up to the five year lag of Greenbookconfidentiality. Thissectionextendstheinflationforecastinthefollowingway.Until1999onecan usethedefinitionofvariablesasbefore.FromFebruary2000untilFebruary2004,the variable (W (cid:0)G) refers to CPI and the variable (F (cid:0)G) refers to PCE. From July t t t t 2004 onwards, the variable (W (cid:0)G) refers to CPI and the variable (F (cid:0)G) refers t t t t to PCE core. The spirit of the regression is the same as before, but the definitions in theindependentanddependentvariablesdonotmatchexactly.Table8showsthatthe resultslargelyholdintheextendedsample,aswellasfrom2000onwardsonly. TableA.3intheappendixexaminessub-sampleanalysisandshowsthattheresults stillhold. 5.2. ResultswithindividualFOMCresponses ThestatisticalanalysispresentedsofaremployedthecentraltendencyoftheFOMC forecasts. The central tendency is less prone to extreme responses that may reflect some strategic behavior by some FOMC participants. I am not discarding strategic behaviorattheindividuallevelresponses,whattheresultspresentedsofarsuggestis that the central tendency seems to be incorporate information from the White House andSPFforecasts. 17. Table A.2 in the appendix reproduces these results for the updated sample used here. The table alsoshowstheresultsofthesameregressionswithGt andWt orGt andSt,whichconfirmsthelimited usefulnessofWt andSt.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 19 TABLE7. Roleinpredictingactualvalues constant G Non-Staff R2 Inflation F OLS -0.1822 (0.2501) 1.0548 (0.4173) -0.0668 (0.3986) 0.8500 WLS -0.2216 (0.1635) 1.3537 (0.3926) -0.3322 (0.3791) 0.8996 Fˆ= f(G;W) OLS -0.1792 (0.2633) 0.9969 (0.7496) -0.0114 (0.744) 0.8499 WLS -0.2237 (0.1914) 0.8828 (0.6355) 0.1239 (0.6217) 0.8956 Fˆ= f(G;S) OLS 0.0011 (0.2964) 1.8428 (0.8637) -0.8723 (0.8754) 0.8435 WLS -0.0718 (0.222) 1.7455 (0.5258) -0.7516 (0.5386) 0.8921 Unemployment F OLS 0.1356 (0.335) 0.9136 (0.3306) 0.0390 (0.339) 0.8048 WLS 0.1236 (0.5701) 0.7373 (0.478) 0.2205 (0.4394) 0.8324 Fˆ= f(G;W) OLS 0.3629 (0.3581) 1.6987 (0.5048) -0.7910 (0.5315) 0.8099 WLS 0.3072 (0.5299) 1.5402 (0.7774) -0.6184 (0.7782) 0.8312 Fˆ= f(G;S) OLS 0.3791 (0.3393) 0.9312 (0.449) -0.0222 (0.4706) 0.8126 WLS 0.2801 (0.5358) 0.8243 (0.5657) 0.1039 (0.563) 0.8277 RealGrowth F OLS 0.4970 (0.3235) 0.2649 (0.4004) 0.5728 (0.4316) 0.4488 WLS 0.5777 (0.4672) 0.2432 (0.5349) 0.5651 (0.5723) 0.4829 Fˆ= f(G;W) OLS 0.9877 (0.3532) 1.7340 (0.6494) -1.0403 (0.7007) 0.4517 WLS 0.9848 (0.5227) 1.7001 (0.8988) -1.0120 (0.9898) 0.4833 Fˆ= f(G;S) OLS 0.7698 (0.4666) 0.4539 (1.067) 0.3196 (1.1695) 0.4291 WLS 0.7576 (0.748) 0.3857 (1.461) 0.3867 (1.6435) 0.4655 Notes: The table reports the estimates of equation (3). All the regressions include as dependent variable the actualoutcomeandasindependentvariabletheGreenbookforecast.Theregressionsalsoincludeasindependent variableeithertheFOMCforecast(F),theFOMCforecastspredictedbyequations(1)and(2)(Fˆ=f(G;W)and Fˆ=f(G;S)). TABLE8. Extendingtheinflationregression constant W S R2of(F-G) R2ofF 1979- W OLS 0.0636 (0.0275) 0.2358 (0.0457) - 0.2428 0.9896 2000- W OLS 0.0220 (0.0546) 0.2190 (0.1077) - 0.1646 0.8752 1979- S OLS 0.0460 (0.0313) - 0.2097 (0.049) 0.1902 0.9874 2000- S OLS 0.0041 (0.055) - 0.2370 (0.0986) 0.2330 0.8554 Notes:Thetablereportstheestimateswiththeinflationforecastfortheextendedsample.Thefirstandthirdrows considerthesamplefrom1979untiltheavailabilityoftheGreenbook.Thesecondandfourthrowsconsiderthesample after2000wherethedefinitionsofinflationdonotcoincide.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 20 This subsection uses the individual responses data described in Romer (2010). On the one hand, the individual responses are released only after a ten year lag and, therefore, are less interesting to analyze than the central tendency that is released immediately.Ontheotherhand,theindividualresponsescontainadditionalstatistical informationwithsixteentoeighteenindividualforecastsperMPR. Table 9 reports the results of a panel data fixed effects regression where the individual elements are the Fed regional banks and the individual governors.18 The main results are still present. The coefficients attached with the non-Greenbook forecasts are statistically significant and are even larger than the values reported in Table4.Thelargercoefficientsinthepaneldataarelargelyexplainedbythesample period.TableA.4intheappendixreportsthecentraltendencyregressionsforthesame sampleperiodwhereonecanobservethatthecoefficientsbecomequitesimilartothe onesobtainedinthepaneldata. TABLE9. PanelDataRegressionResults W S R2of(F-G) Inflation W 0.6651 (0.0481) - 0.3922 S - 0.4572 (0.0352) 0.3796 Unemployment W 0.4571 (0.0308) - 0.4557 S - 0.4492 (0.0288) 0.4888 RealGrowth W 0.3789 (0.0421) - 0.3210 S - 0.3058 (0.0296) 0.3721 Notes:Thetablereportsthepaneldataestimatesofequations(1)-(2).WandSdenotes WhiteHouseandSPFforecasts,respectively.Theequationsareestimatedwithfixedeffects. PaneldataNewey-Weststandarderrorswiththreelagsarereported. 6. Conclusion This paper aimed at contributing to a better understanding of the forecasts that central banks produce and present to the public. The reliability of the forecasts that central banks present to the public and its representatives provides a strong pilar for transparentandaccountablecentralbanksaspartofdemocraticsocieties. ThispaperfocusesontheUnitedStatesgiventheamountofdatathatisavailable. The Greenbook forecast is produced just before the official forecast is released, but 18. Thispanelisunbalancedbecausethereisturnoveramonggovernors.Theresultsareverysimilar whenthegroupofgovernorsistreatedasacluster.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 21 is kept private for five years; using the Greenbook forecast as a control is crucial for the analysis. Also, for the United States one can rely on a large amount of forecasts representingotherviewsintheeconomy. I find that the FOMC forecast does not discard the Greenbook forecast, in fact the weight given to the Greenbook is large. In addition, the difference between the FOMCandtheGreenbookforecastisnotrandomorwhitenoise.IfindthattheFOMC also takes into account, or reflects, forecasts and views of the public and relevant institutions. The statistical model can capture several characteristics of the forecasts in terms of accuracy and their relative ranking. For instance, the model captures that the FOMC forecast does not always improve relative to the Greenbook, but remains thebestforecastamongthesetofforecaststhatarepublishedimmediately. This paper opens avenues for future research questions regarding the underlying motivations of the FOMC because several possible explanations are observationally equivalent.Thispaperfocusedfirstinexaminingwhethercertainresultsareorarenot present. Having said that, two key explanations are consistent with the results that I findinthispaper. First, the FOMC is institutionally designed to reflect or represent a variety of viewsintheeconomy.ThefindingthattheFOMCincorporatesviewsfromthegeneral publicanditsrepresentativesisconsistentwithitsmandate.Somemayhavepreferred thatofficiallypublishedcentralbankforecastsresemblemoreheavilyorrelyontheir internalforecastsandmethods.Othersmayfindcomfortthatpublicviews,regardless of how perfect or imperfect they may be, play a role in institutions that are designed torepresentandservethepublic. Second, the forecasting literature recommends pooling different forecasts and views.Practiceandtheorysuggestthatcombiningforecastsaddsvalue,andcaneven dominatethebestindividualforecast.However,determininginreal-timetheweights to attribute to each forecast is not straightforward. The literature recommends using optimalweightsfromaMSEperspectiveorusingequalweightsforrelevantforecasts. Both approaches can be defended or criticized. The weights obtained in this paper seemtobeinbetweenthosetwoapproaches. ***

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 22 References Banternghansa,ChanontandMichaelW.McCracken(2009).“Forecastdisagreement amongFOMCmembers.”WorkingPapers2009-059,FederalReserveBankofSt. Louis. Clements,MichaelP.andDavidF.Hendry(1998).ForecastingEconomicTimeSeries. Cambridge,CambridgeUniversityPress. Clements, Michael P. and David F. Hendry (2002). “Pooling of Forecasts.” EconometricsJournal,5,1–26. Ellison, Martin and Tom Sargent (2012). “A defence of the FOMC.” International EconomicReview,forthcoming. Faust, Jon and Jonathan H. Wright (2009). “Comparing Greenbook and Reduced FormForecastsUsingaLargeRealtimeDataset.”JournalofBusiness&Economic Statistics,27(4),468–479. Greene,WilliamH.(2000).EconometricAnalysis.PrenticeHall. Hansen, L.P. and T.J. Sargent (2008). Robustness. Princeton, New Jersey, Princeton UniversityPress. Marcet, Albert and J.P. Nicolini (2003). “Recurrent Hyperinflations and Learning.” AmericanEconomicReview,93(5),1476–1498. Marcet, Albert and T.J. Sargent (1989). “Convergence of Least-Squares Learning mechanisms in Self-Referential Linear Stochastic Models.” Journal of Economic Theory,48,337–368. Meade,EllenEandDNathanSheets(2005).“RegionalInfluencesonFOMCVoting Patterns.”JournalofMoney,CreditandBanking,37(4),661–77. Reddy,Sudeep(2008).“TheEconomy:BernankeOpentoaSizableRateCut—Job andHousingMarketsMayWorsen,PosingRisksToGrowth,FedChiefSays.”The WallStreetJournal,15February2008. Reifschneider, David and Peter Tulip (2007). “Gauging the uncertainty of the economic outlook from historical forecasting errors.” Finance and economics discussionseries,BoardofGovernorsoftheFederalReserveSystem(U.S.). Romer, Christina D. and David H. Romer (2000). “Federal Reserve Information and theBehaviorofInterestRates.”AmericanEconomicReview,90(3),429–457. Romer,ChristinaD.andDavidH.Romer(2008).“TheFOMCversustheStaff:Where CanMonetaryPolicymakersAddValue?”AmericanEconomicReview,98(2),230– 35. Romer,David(2010).“ANewDataSetonMonetaryPolicy:TheEconomicForecasts of Individual Members of the FOMC.” Journal of Money, Credit and Banking, 42(5),951–957. Svensson, Lars E. O. (1997). “Inflation forecast targeting: Implementing and monitoringinflationtargets.”EuropeanEconomicReview,41(6),1111–1146. Svensson,LarsE.O.(1999).“Inflationtargetingasamonetarypolicyrule.”Journal ofMonetaryEconomics,43(3),607–654. Tootell, Geoffrey M. B. (1999). “Whose monetary policy is it anyway?” Journal of MonetaryEconomics,43(1),217–235.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 23 Wald,Abraham(1940).“TheFittingofStraightLinesifBothVariablesareSubjectto Error.”TheAnnalsofMathematicalStatistics,11(3),pp.284–300. Zarnowitz, Victor (1992). “Has Macro-Forecasting Failed?” Cato Journal, 12, 129– 164.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 24 Appendix: A.1. StandardDeviations TABLEA.1. StandardDeviationsofForecastDifferences s(F (cid:0)G) s(S (cid:0)G) s(W (cid:0)G) t t t t t t Inflation 0.2858 0.5878 0.5990 Unemp. 0.2192 0.4136 0.3757 RealGrowth 0.3604 0.7357 0.8032 Notes:Thetablereportsthestandarddeviationsof(Ft (cid:0)Gt),(St (cid:0)Gt),(Wt (cid:0)Gt).F,W,andSdenotetheFOMC,White House,andSPFforecasts,respectively. A.2. OptimalWeights

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 25 FigureA.1. Optimalweightsb. White House SPF 1 1 0.5 0.5 0 0 1979 1984 1989 1994 1999 1979 1984 1989 1994 1999 Inflation Inflation 1 1 0.5 0.5 0 0 1979 1984 1989 1994 1999 2004 1979 1984 1989 1994 1999 2004 Unemployment Unemployment 1 1 0.5 0.5 0 0 1979 1984 1989 1994 1999 2004 1979 1984 1989 1994 1999 2004 Real Growth Real Growth Notes:Thefigureplotstheoptimalweightsinequations(1)-(2),buttheconstantiseliminatedfromtheregressionand thecoefficientbislimitedtobebetweenzeroandone.ThegraphsontheleftandrightrefertotheWhiteHouseand SPFforecasts,respectively.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 26 A.3. Roleinpredictingactualvalues TABLEA.2. Roleinpredictingactualvalues constant G Non-Staff R2 Inflation F OLS -0.1822 (0.2501) 1.0548 (0.4173) -0.0668 (0.3986) 0.8500 WLS -0.2216 (0.1635) 1.3537 (0.3926) -0.3322 (0.3791) 0.8996 W OLS -0.1801 (0.251) 0.9883 (0.1948) -0.0028 (0.1835) 0.8499 WLS -0.2140 (0.1729) 0.9755 (0.1735) 0.0312 (0.1566) 0.8956 S OLS -0.0485 (0.2723) 1.1531 (0.1789) -0.1826 (0.1832) 0.8435 WLS -0.1051 (0.2062) 1.1757 (0.122) -0.1819 (0.1303) 0.8921 Unemployment F OLS 0.1356 (0.335) 0.9136 (0.3306) 0.0390 (0.339) 0.8048 WLS 0.1236 (0.5701) 0.7373 (0.478) 0.2205 (0.4394) 0.8324 W OLS 0.3689 (0.3598) 1.2331 (0.1962) -0.3254 (0.2186) 0.8099 WLS 0.3075 (0.5299) 1.1708 (0.3205) -0.2490 (0.3133) 0.8312 S OLS 0.3797 (0.3442) 0.9176 (0.1667) -0.0087 (0.1834) 0.8126 WLS 0.2774 (0.5366) 0.8878 (0.2328) 0.0403 (0.2185) 0.8277 RealGrowth F OLS 0.4970 (0.3235) 0.2649 (0.4004) 0.5728 (0.4316) 0.4488 WLS 0.5777 (0.4672) 0.2432 (0.5349) 0.5651 (0.5723) 0.4829 W OLS 0.9872 (0.353) 1.0000 (0.1766) -0.3063 (0.2063) 0.4517 WLS 0.9869 (0.524) 0.9859 (0.2307) -0.2978 (0.2912) 0.4833 S OLS 0.7832 (0.4301) 0.6934 (0.2101) 0.0800 (0.2929) 0.4291 WLS 0.7718 (0.6989) 0.6792 (0.2438) 0.0932 (0.396) 0.4655 Notes:Thetablereportstheestimatesofequation(3).Alltheregressionsincludeasdependentvariabletheactual outcomeandasindependentvariabletheGreenbookforecast.Theregressionsalsoincludeasindependentvariable eithertheforecastsoftheFOMC(F),theWhiteHouse(W),ortheSPF(S).

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 27 A.4. Sub-sampleAnalysis TABLEA.3. RegressionResults:Sub-sampleanalysis After1985 constant W S R2of(F-G) R2ofF Inflation W OLS 0.0054 (0.0279) 0.3341 (0.0771) - 0.3197 0.9470 WLS 0.0070 (0.0356) 0.3531 (0.0993) - 0.3242 0.9468 S OLS -0.0577 (0.0323) - 0.3729 (0.0837) 0.3317 0.9513 WLS -0.0632 (0.0315) - 0.3858 (0.0807) 0.3410 0.9513 Unemployment W OLS -0.0155 (0.0193) 0.4641 (0.0577) - 0.5065 0.9720 WLS -0.0110 (0.0259) 0.4507 (0.0876) - 0.4865 0.9717 S OLS -0.0317 (0.0178) - 0.4995 (0.0541) 0.5827 0.9745 WLS -0.0294 (0.0227) - 0.4989 (0.0766) 0.5780 0.9745 RealGrowth W OLS 0.0062 (0.03) 0.3099 (0.0481) - 0.3973 0.9055 WLS 0.0064 (0.0455) 0.3074 (0.0514) - 0.3871 0.9056 S OLS 0.0218 (0.0273) - 0.3977 (0.0486) 0.5228 0.9240 WLS 0.0080 (0.0335) - 0.3893 (0.0504) 0.5431 0.9242 After1990 constant W S R2of(F-G) R2ofF Inflation W OLS -0.0248 (0.0328) 0.4852 (0.1334) - 0.3461 0.9270 WLS -0.0186 (0.0334) 0.4856 (0.1307) - 0.3529 0.9270 S OLS -0.0628 (0.0378) - 0.2958 (0.0887) 0.3081 0.9288 WLS -0.0708 (0.0355) - 0.2965 (0.0883) 0.3124 0.9288 Unemployment W OLS -0.0289 (0.0224) 0.5147 (0.0702) - 0.5285 0.9730 WLS -0.0249 (0.0302) 0.4971 (0.0942) - 0.5128 0.9725 S OLS -0.0093 (0.0205) - 0.5293 (0.0592) 0.6347 0.9813 WLS -0.0097 (0.0279) - 0.5229 (0.086) 0.6229 0.9812 RealGrowth W OLS 0.0690 (0.0355) 0.4266 (0.0619) - 0.4972 0.9136 WLS 0.0763 (0.05) 0.4297 (0.0587) - 0.5113 0.9135 S OLS 0.0306 (0.0331) - 0.4187 (0.0541) 0.5653 0.9264 WLS 0.0182 (0.0402) - 0.4043 (0.059) 0.5731 0.9266 Notes:Thetablereportstheestimatesofequations(1)-(2).WandSdenotesWhiteHouseandSPFforecasts,respectively.The equationsareestimatedbothwithOLSandWLS.Newey-WeststandarderrorswiththreelagsarereportedintheWLSregression. Theupperpanelonlyincludesthesampleafter1985,thelowerpanelincludesthesampleafter1990.

Nunes Docentralbanks’forecaststakeintoaccountpublicopinionandviews? 28 TABLEA.4. RegressionResults:Sub-samplecoincidingwithPaneldata constant W S R2of(F-G) R2ofF Inflation W OLS -0.0045 (0.0264) 0.6377 (0.1082) - 0.6124 0.9319 WLS -0.0076 (0.031) 0.6109 (0.0798) - 0.5901 0.9324 S OLS -0.0498 (0.0326) - 0.4093 (0.0865) 0.5043 0.9147 WLS -0.0705 (0.0286) - 0.4181 (0.0824) 0.5257 0.9143 Unemployment W OLS -0.0673 (0.0267) 0.4795 (0.0716) - 0.6158 0.9812 WLS -0.0663 (0.0343) 0.4779 (0.0949) - 0.6209 0.9812 S OLS -0.0390 (0.0243) - 0.4754 (0.0606) 0.6875 0.9857 WLS -0.0420 (0.0301) - 0.4894 (0.087) 0.7292 0.9859 RealGrowth W OLS 0.1249 (0.0414) 0.3711 (0.1066) - 0.3023 0.8782 WLS 0.1366 (0.0501) 0.3972 (0.0725) - 0.3660 0.8759 S OLS 0.0945 (0.0401) - 0.2752 (0.0701) 0.3550 0.8841 WLS 0.0887 (0.0436) - 0.2733 (0.0574) 0.3494 0.8843 Notes:Thetablereportstheestimatesofequations(1)-(2).WandSdenotesWhiteHouseandSPFforecasts,respectively.The equationsareestimatedbothwithOLSandWLS.Newey-WeststandarderrorswiththreelagsarereportedintheWLSregression. Thesamplegoesfrom1992to2001.