Inflation Nowcasting: Frequently Asked Questions These questions and answers accompany the technical working paper Nowcasting U.S.

Transcription

1 Inflaion Nowcasing: Frequenly Asked Quesions These quesions and answers accompany he echnical working paper Nowcasing US Headline and Core Inflaion by Edward S Knoek II and Saeed Zaman See he paper for more deails and references Wha is nowcasing? Nowcasing is a combinaion of wo erms: now and forecasing Economic indicaors are ofen released wih a considerable delay, bu here is usually inense ineres in knowing wha a given indicaor is before is official release Nowcasing refers o he predicion of he presen, he very recen pas, and he very near fuure in essence, rying o fill in ha missing daa poin wih an accurae esimae Ofen, nowcass ake advanage of higher-frequency daa ha are informaive abou some oher economic indicaor o predic wha he nex release of he indicaor of ineres will say Why is inflaion nowcasing useful? Inflaion is a opic of perennial ineres because i influences he behavior and plans of consumers, businesses, financial markes, pension funds, governmens in essence, everyone in an economy The Federal Reserve pays close aenion o inflaion because of is impac on he economy and because he Federal Open Marke Commiee has been explicily asked by Congress hrough he Federal Reserve Ac o promoe he goals of maximum employmen and price sabiliy, he so-called dual mandae When making decisions, consumers and businesses may have o forecas he inflaion rae far ino he fuure For example, if a consumer is hinking abou aking ou a loan, i helps o know how quickly wages and prices will be rising during he life of he loan afer all, i will be much easier o service he loan wih sronger wage and price growh Unforunaely, inflaion ends o be difficul o predic accuraely Bu some recen research finds ha forecass of inflaion in he fuure can be improved by having more accurae esimaes of near-erm inflaion, or nowcass These nowcass serve as an imporan jumping-off poin for hinking abou where inflaion will be in he fuure Which inflaion measures do you nowcas? We generae nowcass for four inflaion measures: (1) inflaion in he price index for personal consumpion expendiures, more commonly referred o as inflaion; (2) inflaion in he price index excluding food and energy, also known as core inflaion; (3) inflaion in he consumer price index (); and (4) inflaion in he excluding food and energy, also known as core inflaion The model repors seasonally adjused, monh-over-monh inflaion raes in hese four measures (expressed as nonannualized percen changes) and quarerly inflaion raes in hese four measures (expressed a seasonally adjused annualized raes, or SAAR) The model also repors year-over-year inflaion raes in hese four measures (based on nonseasonally adjused daa for inflaion and core inflaion and seasonally adjused daa for inflaion and core inflaion) Wha is so special abou your inflaion nowcass? Using hisorical comparisons from 1999 o 2015, inflaion nowcass from he Cleveland Fed s model have been highly accurae Our model s nowcass easily ouperform a variey of nowcass from alernaive saisical models, especially over he course of a monh or quarer Bu perhaps even more noable is he model s performance compared wih he bes available benchmarks: nowcass from surveys of professional forecasers The model s nowcass for inflaion end

2 o be more accurae han he consensus (average) nowcass from he Blue Chip Economic Indicaors survey Is nowcass for and inflaion also end o be more accurae han he median nowcass from he Federal Reserve Bank of Philadelphia s Survey of Professional Forecasers (SPF) These resuls are somewha surprising, because professional forecasers can and do use a range of models and judgmen o capure he special facors ha affec near-erm inflaion rends, and here is addiionally evidence of he wisdom of he masses (or, more formally, he law of large numbers) when making forecass Meanwhile, he model s nowcass for core inflaion end o be jus as accurae as hose in he SPF, bu he model is available a a daily frequency, while he SPF is released only once per quarer Finally, we also compare he model s inflaion nowcasing accuracy wih publicly available nowcass from he Greenbook (he in-deph analysis of economic condiions produced by he Federal Reserve Board of Governors saff for Federal Open Marke Commiee meeings) While Greenbook nowcass end o be highly accurae, hey are only released o he public wih a five-year lag The Cleveland Fed s model nowcass have hisorically ended o be quie similar o hose in he Greenbook, bu hey are available in real ime, wihou delay Are he esimaes updaed daily? Yes, every business day around 10:00 am Easern ime Wha is he iming of he and inflaion daa releases? In he Unied Saes, he Bureau of Labor Saisics (BLS) ypically releases he for a given monh around he middle of he following monh (for example, he January is released around mid-february) The Bureau of Economic Analysis (BEA) ypically releases he oher major measure of consumer prices, he price index, around he end of he following monh (for example, he January price index is released around he end of February, afer he for January has been released) Which high-frequency daa in paricular are used o produce he inflaion nowcass? Daily Bren crude spo oil prices and weekly reail gasoline prices How many daa series are used o produce he inflaion nowcass? The nowcass are produced using en daa series Eigh daa series are monhly Five of hese monhly series come from he BLS: he, he core, he for food, he for food a home, and he for gasoline (All of he series we use are seasonally adjused) Three of he monhly series come from he BEA: he price index, he core price index, and he price index for food and beverages purchased for off-premises consumpion One daa series is weekly: reail gasoline prices, released each Monday by he Energy Informaion Adminisraion One daa series is daily: Bren crude spo oil prices, from he Financial Times or he Energy Informaion Adminisraion Could you very briefly describe how he nowcass are made? In a nushell, here are four pars o he model The firs par nowcass (or, perhaps more accuraely, forecass) core inflaion To do so, we find ha esimaes based enirely on he recen pas do a fairly good job The second par forecass food price inflaion Again, we primarily use he recen pas o forecas fuure food price inflaion The hird par nowcass gasoline price inflaion We use a combinaion of curren gasoline prices and curren oil prices, under he assumpion ha oday s oil prices are informaive abou where gasoline prices are likely o head in he fuure, and hen we seasonally adjus he daa Finally, he fourh par combines he nowcass and forecass of core

3 inflaion, food price inflaion, and gasoline price inflaion o come up wih nowcass of inflaion in eiher he or price index Why do he core inflaion nowcass change so lile and so infrequenly? Why do he headline inflaion nowcass move around so much and so ofen? Nowcass change when new daa arrive ha are differen han wha was expeced This can be eiher because of he frequency wih which new daa arrive or he volailiy of ha new daa The core inflaion nowcass use very few daa sources, which arrive infrequenly; herefore, hey only change infrequenly Nowcass of core inflaion are based only on pas core inflaion; hus, he nowcass can only change when new daa are released (or when pas daa are revised) However, if he daa come in exacly as expeced, hen he core inflaion nowcass will no acually change Nowcass of core inflaion are primarily based on pas core inflaion, bu hey can also be affeced by he availabiliy of daa: if we have one more core observaion han core for some monh, hen we map monh s core observaion o core in monh Thus, core nowcass can only change when we ge new daa on eiher he or he price index bu again, if he daa come in exacly as expeced, hey acually will no change Core inflaion also ends o be relaively sable, so revisions o core inflaion nowcass can be small when hey do occur By conras, gasoline prices play an imporan role in headline and inflaion nowcass, and gasoline price nowcass depend in urn on oil prices Oil prices move around almos every day, and hey can be quie volaile and hus difficul o predic The headline inflaion nowcass can change when new or price index daa are released; bu beween releases, hey ofen move around based on incoming oil and gasoline price daa How accurae are your nowcass? Nowcass are a ype of forecas and hus are always subjec o considerable uncerainy and imprecision Afer all, we are using a very small number of series (en) and simple saisical echniques o make informed guesses abou economic indicaors before hey are released By conras, he official daa releases are usually he produc of ens of housands of observaions, powerful compuaional echniques, and a very large number of hours worked On average, he model s nowcass end o be more precise han a range of compeing nowcass, bu he ouperformance is no absolue The accuracy of each individual nowcas depends on a range of facors, including he measure under consideraion and he poin in he monh or quarer ha he nowcas is being made On average, nowcass made wih more informaion laer in a monh or quarer end o be more accurae Bu shocks are ever-presen ha can push inflaion away from he nowcased value, especially if hey occur o inflaion componens ha we do no ry o model How do I compare he model s hisorical nowcass wih he acual daa? The charing ool on he nowcasing websie allows you o plo a sequence of nowcass for previous monhs or quarers along wih he acual inflaion release (for simpliciy, we display he firs available daa release corresponding o he quarer or monh being nowcased) This allows he user o see how close or far he model s nowcass were compared wih wha was acually released by he saisical agency (he BLS or he BEA) Typically, nowcass early in he monh or quarer are less precise han nowcass made laer in he monh or quarer, because he early nowcass are based on less informaion han laer nowcass Wha s coming nex? We are consanly seeking ways o improve his websie and is funcionaliy One iem on he o-do lis is o provide an opion ha includes hisorical nowcasing error bands in he chars ha vary wih

4 he exac informaion se available a a given poin in ime These bands would sar ou wide a he beginning of a monh or quarer and hen shrink over ime as more informaion accumulaes Whom should I conac if I have any furher quesions, commens, or suggesions? We welcome your quesions, commens, and suggesions for ways o make his websie more useful and o enhance our lis of FAQs While we canno respond o every query, we plan o make updaes o he sie and he FAQs, and your inpu would be much appreciaed us a mailo:inflaioncenral@clevfrborg Is he same model used in he nowcasing working paper and he nowcasing websie (echnical)? Yes, he basic model is he same, bu here is one difference The paper uses a single series on food inflaion (specifically, he for food) in consrucing boh and inflaion esimaes On he websie, he inflaion esimaes are consruced using a differen measure of food inflaion inflaion in he price index for food and beverages purchased for off-premises consumpion This change is mehodologically more consisen wih how he price index is consruced, and i would have been grea o include i in he paper Bu unforunaely, a long real-ime hisory of his paricular series is no available Afer deciding o include he price index for food and beverages purchased for off-premises consumpion, we added one more series o he model he for food a home Monhly inflaion in he for food a home has hisorically been similar o monhly inflaion in he price index for food and beverages purchased for off-premises consumpion As a resul, he for food a home can provide some addiional informaion a imes abou he price index for food and beverages purchased for off-premises consumpion Could you walk hrough he model (echnical)? For more on he model, see he working paper Here, we very quickly go hrough he key elemens of he model used o compue he inflaion nowcass, focusing more on pracical implemenaion han echnical deails Le T denoe quarers and denoe monhs, such ha wihin some quarer T here are monhs =1, 2, 3 Quarerly inflaion is usually measured a seasonally adjused annualized T 4 raes as T 100[( PT / PT 1) ], where PT (1/ 3)( PT, 1 PT, 2 PT, 3) Monhly inflaion is expressed in nonannualized erms as 100( P / P 1 ) Thus, if we have nowcass or forecass of monhly inflaion raes, we can fill in he missing price levels for he monhs wihin a given quarer o compue quarerly inflaion We denoe nowcass or forecass for a variable x by using he noaion f(x) In wha follows, we se J=12, τ=24, and τ L =60 inflaion Suppose we have monhly core inflaion daa hrough monh 1, We recursively forecas monhly core inflaion using J-monh moving averages, so he forecas for monh is f ( ) (1/ J) j1,, J j, and forecass for monhs +1, +2, follow Core inflaion Suppose we have monhly core inflaion daa hrough monh Core 1, If we have core inflaion hrough monh (daa, no a forecas) bu are missing core inflaion in monh, we bridge from core inflaion o core inflaion Tha is, using a rolling window of lengh τ, we esimae he regression model

5 e, hen use he esimaed coefficiens and he acual reading on Core 0 1 Core o obain f ( ) If we do no have more core inflaion daa han we have core inflaion daa, or if we ve already bridged all he available core inflaion daa, hen we Core recursively forecas monhly core inflaion f ( k ), k 0, using J-monh moving averages food inflaion Suppose we have monhly food inflaion hrough monh 1, Food Food As wih core inflaion, we recursively forecas monhly food inflaion f ( k ), k 0, using J-monh moving averages food inflaion Suppose we have monhly inflaion in he price index for food Food Off Premises and beverages purchased for off-premises consumpion hrough monh 1, If Food a Home we have monhly inflaion in he for food a home (no jus food) in monh,, Food Off Premises Food a Home we bridge i over o he -equivalen concep by seing f ( ) If we do no have more daa han daa, or if we ve already bridged i, we recursively Food Off Premises forecas f ( k ), k 0, using J-monh moving averages (NSA) inflaion Leing P denoe he average of available weekly gasoline prices wihin a monh before any seasonal adjusmen, we compue monhly gasoline inflaion (NSA) (NSA) (NSA) 100( P / P ) We seasonally adjus he series o make i useful: 1 if is monhly inflaion in he (seasonally adjused) for gasoline in monh j, we j define he seasonal facor in monh o be sf (1/ 3) ( (NSA) ) j 1 year, 2 years, 3 years j j and hen use i o derive our measure of seasonally adjused monhly gasoline inflaion: (NSA) Oil ˆ sf To augmen our gasoline price daa, le P be he average of available daily oil prices wihin a monh Mechanically, we exend he monhly oil price series Oil by one observaion by seing P o he las available daily observaion Using a rolling window of lengh τ L, we esimae he firs-sage regression 1 P P e ; le (NSA) Oil (NSA) gp ( ) denoe he prediced gasoline prices based on he regression coefficiens and he acual monhly oil prices Nex, we esimae he second-sage regression (NSA) (NSA) (NSA) (NSA) [ P g( P )] a[ P g( P )] e, using he same rolling window We mach he lengh of 1 1 gp ( ) o he lengh of he oil price series using he (NSA) 1 (NSA) firs-sage regression coefficiens, and hen derive forecass of f( P ) based on he second-sage regression coefficien Finally, we compue monhly inflaion in he non-seasonally adjused gasoline prices and hen seasonally adjus he daa as described above o produce a se of f ( k ), k 0 Due o release lags, we ypically have one or wo more monhs of gasoline inflaion nowcass or forecass han we have and inflaion daa inflaion Suppose we have monhly inflaion daa hrough monh 1, Food We esimae he regression model e using a rolling window of lengh τ In nowcasing monhs, +1, ec, if we have an esimae of f ( ) for some k 0, we use i along wih he coefficiens we jus esimaed and he k 1

6 esimaes of f and Food ( k ) f produced earlier and o derive he nowcas of ( k ) f ( k) If we do no have an esimae of f ( k ), we recursively forecas f ( k), k 0, using J- monh moving averages inflaion Suppose we have monhly inflaion daa hrough monh 1, If we have monhly inflaion daa (no a forecas) hrough monh, we bridge i o nowcas inflaion by esimaing he regression model e using a rolling window 0 1 of lengh τ, hen use he esimaed coefficiens and he acual reading on o obain f ( ) Going forward for some monh +k, if we have an esimae of f ( k ), we firs esimae he Core Food Off Premises regression model e using a rolling window of lengh τ, and hen use hose esimaed coefficiens along wih esimaes of Food Off Premises Core f ( k ), f ( k ), and f ( k ) produced earlier o derive he nowcas of f ( k) Finally, if we do no have an esimae of f ( k ), we recursively forecas f ( k) using J-monh moving averages