纽约联储-美联储透明度和政策预期错误：文本分析方法（英）-47页

NO.1081FedTransparencyandNOVEMBER2023PolicyExpectationErrors:ATextAnalysisApproachEricFischerRebeccaMcCaughrinSakethPrazadMarkVandergonFedTransparencyandPolicyExpectationErrors:ATextAnalysisApproachEricFischer,RebeccaMcCaughrin,SakethPrazad,andMarkVandergonFederalReserveBankofNewYorkStaffReports,no.1081November2023https://doi.org/10.59576/sr.1081AbstractThispaperseekstoestimatetheextenttowhichmarket-impliedpolicyexpectationscouldbeimprovedwithfurtherinformationdisclosurefromtheFOMC.Usingtextanalysismethodsbasedonlargelanguagemodels,weshowthatifFOMCmeetingmaterialswithfive-yearlaggedreleasedates—likemeetingtranscriptsandTealbooks—wereaccessibletothepublicinrealtime,marketpolicyexpectationscouldsubstantiallyimproveforecastingaccuracy.Mostofthisimprovementoccursduringeasingcycles.Forinstance,atthesix-monthforecastinghorizon,themarketcouldhavepredictedasmuchas125basispointsofadditionaleasingduringthe2001and2008recessions,equivalenttoa40-50percentreductioninmeansquarederror.ThispotentialforecastingimprovementappearstoberelatedtoincompleteinformationabouttheFed’sreactionfunction,particularlywithrespecttofinancialstabilityconcernsin2008.Incontrast,havingenhancedaccesstomeetingmaterialswouldnothaveimprovedthemarket’spolicyrateforecastingduringtighteningcycles.JELclassification:E43,E52,E58,C80Keywords:interestrate...