1

No.2023-20

BANKOFKOREA

BOKISSUENOTE

Jun.27,2023

Therelationshipbetweenthe

levelofinflationandattention

toit,anditsimplications

ChaeMinSok*,ChoiJongHo**

?1Weproxyfortheattentionthatinflationreceivesviathefrequencyofinternetsearchesforwordsrelatedtopricesandconsumers’perceptiongaponinflation.Wechose“prices”,“pricegrowthrate”and“inflation”aswordsrelatedpricesandmeasurethefrequencyofsearchesforthesewordsonGoogleandNaver1).Wederivethegapbetweenconsumers’perceptionandtheriseintheconsumerpriceindex(CPI)inthelastoneyearbasedontheConsumerSurveyresultsandtheactualriseintheCPI.

?2Empiricalstudyshowswhentheinflationrateexceedsacertainthreshold,economicagentsturnincreasinglyattentivetowardsinflation.Whentheinflationraterunsbeyondacertainlevel,internetsearchesonthepricesrelatedwordsincreasesharply,andthedisparitybetweenconsumers‘perceptionontheinflationrateandtheactualratelessensasconsumersbecomemoreacuteaboutpricemovements.Precedentstudiesdiscoveredthecommonphenomenonofattentiontoinflationacceleratingwhentheinflationratehitsaboveacertainlevelinmanycountries.

?3Whenattentiontoinflationrisesduetonoticeablegainininflation,informationonpricesandtheshockimmediatelyimpacteconomicagents’expectationstomakeinflationexpectationscanbecomeunstable.Accordingtotheresultsofathresholdvectorautoregression(TVAR)analysis,whentheinflationrateishigherthantheattentionthreshold,priceshockshavedeeperandlongerlastingimpactoninflationexpectations.

1)ThemostpopularandrepresentativesearchengineinSouthKorea.

2

?4Undertheenvironmentwheretheinflationrateisexpectedtohoverbeyondthetargetlevelforsometime,thecentralbankmustputfocusonstabilizingpriceuntiltheinflationratecomesdownnearthetarget.

*PriceResearchTeam,ResearchDepartment(minsok.chae@bok.or.kr)

**PriceResearchTeam,ResearchDepartment(jonghochoi@bok.or.kr)

■Disclaimer:Theviewsexpressedhereinarethoseoftheauthors,anddonotnecessarilyreflecttheofficialviewsoftheBankofKorea.Whenreportingorcitingthispaper,theauthors’namesshouldalwaysbeexplicitlystated.

■WearegratefultoChoiChangHo,director-generaloftheResearchDepartment,KangTaeSoo,headofPriceandEmploymentbureau,andKimJungSung,headofthePriceResearchTeam.

3

Ⅰ.Introduction

1.Theextentofattentiontoinflationbyeconomicagentsbecomesoneofthecentralelementsindefiningpricestability.FormerU.S.FederalReserveChairmanAlanGreenspandefinedpricestabilityasastateinwhichthelevelandvolatilityofinflationarelow,sothathouseholdsandfirmsdonothavetopaymuchattentiontoinflationintheirdecision-makingprocesses.(Greenspan,2001).Sinceittakesacertainamountoftimeandefforttoobtaininformationonprices,householdsandfirmstendtopaylittleattentiontotheinflationratewhenpricegrowthisslow.(Frankel,2019).Ontheotherhand,whenpricegrowthaccelerates,inflationbecomessalient,asfailuretoaccuratelyreflectthecurrentinflationrateandfutureprospectsinwagenegotiationsorpricesettingscanleadtogreatlosses.(Blanchard,2022).

2.Whenhighinflationpersistsasrecentlyandattentiontoinflationheightens,inflationexpectationscanbecomeunstable.Thisisbecausegreaterattentiontopricelevelscanprompteconomicagentstoquicklyreflectchangesinthepricesituationintheformationofexpectations.Theunstablenessininflationexpectationsaffectsthedecision-makingofeconomicagents,suchaswagenegotiations,pricesettings,andinvestmentdecisions,andultimatelyimpactstheactualinflationrate.

WhentheCPIrisesatahighrate,inflationexpectationsbecomemorevolatile.

[Figure1]CPIriseandinflationexpectations

volatility1)

Note:1)Standarddeviationofconsumers’

expectationsoninflationleveloverrecent

1-yearhorizon.

Source:BOK.

3.Weneedacloserstudyonthecorrelationsbetweentheinflationlevel,attentiontoinflation,andinflationexpectations.Inthispaper,wefirstmeasuretheattentiontoinflationthroughvariousmeans,andthenfindanswerstothefollowingtwoquestions.

.Howdoesthelevelofinflationchangetheeconomicagents’attentiontoinflation?

.Howdoestheresponseofinflationexpectationstopriceshockdifferbetweentheperiodsofhighandlowattentiontoinflation?

4

Measurementofattentiontoinflation

4.Therecanbevariouswaystomeasurepublicattentiontoinflation.Thispaperexamines①internetsearchesforwordsrelatedtoprices,and②thedifferencebetweenconsumers’perceptiononinflationrateandtheactualinflationrate(=perceptiongaponinflation).Whenthereisagreaterattentiontopricelevels,economicagentsmorefrequentlylookuprelatedwordsontheinternet.Agentswithhigherinterestininflationtendtobemoreaccurateinassessingthepricegains,suggestingthathigherattentivenesstowardsinflationcanreducetheperceptiongap.(Cavalloetal.,2017).

5.Wechoose‘prices’,‘priceincreaserate’,and‘inflation’aswordsrelatedprices2),andmeasurethefrequencyofsearchesforthesewordsonGoogleandNaver.Thetwoplatformscanbemutuallycomplementaryduetodifferingmeritsanddisadvantages.Googleprovidesrecordsfrom20043),enablinganexaminationoveralongerhorizon.Ontheotherhand,thedatafromNaverisavailableonlyfrom2016,butisusedbyagreaternumberofKoreans.4)The

peakinmonthlyfrequencyofsearchesisindexedas100.[Figure2]showsthesearchfrequency,whichremainedrelativelylowaftershowinglargefluctuationduring2007~2011,hasincreasedsignificantlysince2021.5)

Searchforinflationsurgesfrom2021

[Figure2]Searchesforpricesrelated

words1)

(SearchesonGoogle)

(SearchesonNaver)

Note:1)Prices,priceincreaserate,inflation.

Source:GoogleTrends,NaverDatalab.

2)Wealsoexaminedthefrequencyofsearchesforthelargergroupofwordsthatadditionallyincludesconsumer“priceindex“,“CPI“,“coreprices“,andtheirrateofincreases,butthetrendandcorrelationwithinflationratedidnotchangemuch.

3)Westudiedthedatafrom2007forhighercredibility,givengreatervolatilityinearlierdata.

4)NavertoppedinMAUtallyfromJan.toMay.2023withashareof58.8%,whileGooglewassecondwithashareof31.4%(InternetTrendWebsite).

5)In[Figure2],searchesonpricesrelatedwordspeakedinJune2022atbothGoogleandNaver.ThesearchessurgefromJune13to16reflectingconcernsoverhigherinflationastheCPIgaininMayintheU.S.toucheda41-yearhighof8.6%anditcausedjittersoveracceleratedmonetarypolicytighteningbytheFedthatsentWallStreetstockpricestumblingover2%onJune13,whileKorea’sCPIdataforMay2022,whichwasreleasedinJune3,showedtheYoYgaininCPIhittingabove5%forthefirsttimein13yearsorsinceSept.2008.

6.Wethenmeasuretheperceptiongaponinflationbyexaminingthedifferencebetweenconsumers’perceptionofCPIgainoverthepast1yearandtheactualinflationrate.6)Accordingto[Figure3],whichillustratestheperceptiongapsince2013whenthequestiononCPIgainoverthepast1yearwasincludedintheConsumerSurvey,thegaphassignificantlynarrowedfrom2021.7)

Perceptiongaponinflationhassharplynarrowedfrom2021

[Figure3]Perceptiongaponinflation1)

Note:1)Consumers’perceptiononCPIgainoverthe

past1year–actualCPIincreaserate.Source:Authors’calculations.

Therelationshipbetweenthelevelofinflationandattentiontoit

7.Wefirststudythechangeininternetsearchesforwordsrelatedtopricesinresponsetotherisesinconsumerprices.

Weestimatetheequation(1)basedonthemethodologyproposedbyKorenoketal.(2022)toexaminetherelationshipbetweensearchesforinflation(At)andprior-month

YoYCPIgain(πt—1).ThelinearrelationshipbetweensearchesforinflationandCPIriseisallowedtovarydependingonwhetherthelatterisgreaterthanthethresholdparameter(Y)ornot.WeestimateYbyfindingtheonewhichminimizestheresidualsumofsquares(RSS).Prior-monthCPIriserateisusedbecauseweassumethatinflationchangescanaffectattentionlevelafteralagoftimegiventhatCPIdataisreleasedinthefollowingmonthandthatittakestimetoprocessandaccumulateinformationoninflation.Evenwhentheseassumptionsareoverlooked,theestimationresultsweresimilar.8)

At=c+β1πt—1*D1t+β2πt—1*D2t+Et(1)

D1t:1ifπt—1≤Y,and0otherwise.D2t:1ifπt—1持Y,and0otherwise.

6)Theactualinflationratemeanstheaverageofthelast12monthsofYoYCPIgrowth.

7)ThejumpinthegapfromJunetoOct2022stemsfromtheuptickinattentiontowardsinflationfromJune2022thatcausedsharpcorrectionofconsumers’perception.

8)Assearchesforinflationalsorelatetotheinflationrateofthecurrentmonthanditisavailableinrealtime,itcanbehelpfultopredictinflation.Althoughfollow-upstudywouldbenecessaryformoreaccurateanalysis,thecomparisonbetweensimpleinflationpredictionautoregressive(AR)modelandamodelenhancedwithsearchesforinflationshowedthatthemodelwithinformationonsearchesresultedwithlowerroot-mean-squareerror(RMSE)by6.3%todemonstratehigherpredictability.

。

8.Theequation(1)estimationresultsshowwhenCPIriseexceedsacertainthreshold,searchesforinflationaccelerate,whichimplieslargerattentiontoinflation.IntheanalysiswithGoogledata,whenCPIgainwasbelow2.0%,therewasinsignificantrelationshipbetweensearchesforinflationandCPIrise.Whenhigher,therelationshipbecamesalient.InthecaseofNaver,searchesforinflationincreasedwhenCPIgaintopped2.3%.

[Table1]EstimationresultsoftherelationshipbetweenCPIriseandsearchesforinflation

c

B1

B2

出(%)

Google

28.4***

-0.3

***

6.3

***

2.0

Naver

22.4***

-2.6

***

5.0

***

2.3

Note:1)*,**and***respectivelymeanssignificanceat10%,5%and1%.Inthecaseof出,itisaboutF-testofthenullhypothesisthatthethresholdmodelprovidesnostatisticalimprovementover

ano-thresholdlinearmodel.

Source:Authors’calculations.

Searches

forinflation

increase

whenthe

inflation

rateruns

beyond

acertain

threshold

[Figure4]CPIincreaserateandsearches

forinflation

(SearchesonGoogle)

(SearchesonNaver)

Soure:Authors’calculations.

ComparisoninPredictability

Subject

Basicmodel(AR)

Modeladdedwithsearchesforinflation

RMSE

0.3847

0.3605

Note:Weestimatedthemodel(theandMay2020,andthengetfromJune2020toMay2023.higherthepredictability.

orderisselectedbasedontheBIC)usingdataacrossJan2007

RMSEfromtheestimatedvalue(t+1|t)andrealvalue(πt+1)

RMSEcanbedefinedasfollows,andthelowerthenumber,the

RMSE=v-T—1R(πt+1—t+1|t)2

Source:Author’scalculations.

‘

7

9.Accordingtopreviousstudiesusinginternetsearchesforinflation,theriseinpublicattentiontoinflationwhenthegainsexceedacertainthresholdisacommonfeatureinmanycountries.TheattentionthresholdsvarybetweencountriesasemergingeconomieslikeBrazil(7.1%)andChile(3.9%)arehigherthanadvancedeconomiesliketheU.S.(3.5%),Germany(2.6%),andU.K.(2.2%)(Korenoketal.,2022).

10.Wethenexaminehowtheriseininflationaffectsconsumers’perceptiongaponinflation.BasedonthemethodologyofGoodspeed(2023),weobservetherelationshipbetweenperceptiongap(Rt)andpriormonthYoYCPIgain(πt—1)byestimatingtheequation(2).Asabove,thethresholdvalue(Y)thatseparatestwo

regimesisestimatedbyfindingthevaluewhichminimizestheresidualsumof

squares.

Rt=c1+c2*Dt+Et(2)

Dt:1ifπt—1持Y,and0otherwise.

11.Theequation(2)estimationresultsshowwhenCPIgainexceedsacertainthreshold,consumersaremoreaccurateinperceivingthepast1-yearinflationlevel.Thefindingsuggestswhentheinflationrategoesbeyondacertainlevel,economicagentsbecomemoreattentivetowardspricemovement.WhenCPIgainwasbelow2.6%,therewasagapof1.3%pbetweentheperceivedCPIriserateforthelast1yearandtheactualrate.WhenCPIgainrisesbeyond2.6%,thegapnarrowedto0.4%p.

[Table2]Estimationresultsofthe

relationshipbetweenCPIriseandperceptiongaponinflation

c1

c2

(%)

1.3***

-0.9***

***

2.6

Note:1)*,**and***respectivelymeans

significanceat10%,5%and1%.Inthecaseof,itisabouttheF-testofnullhypothesisthatthethresholdmodelprovidesnostatisticalimprovementoverano-thresholdlinearmodel.

Source:Authors’calculations.

Whentheinflationrateexceedsacertainthreshold,consumersperceivetheinflationratemoreaccurately

[Figure5]CPIincreaserateandperceptiongaponinflation

Source:Authors’calculations.

Differencesintheimpactofpriceshocksoninflationexpectationsacrosstheattentionregimes.

12.Whenattentiontoinflationincreaseswithinflationrunningaboveacertainlevel,informationandshockrelatedtopricescanimmediatelyimpacteconomicagents’expectations,thusinflationexpectationscanbecomeunstable.Toempiricallyconfirmthetheory,usingdatafromQ12007toQ4

2023,weestimateThresholdVARwithfourvariables:GDPgaprate(gt)9),wageriserate(wt,YoY)10),CPIriserate(πt,YoY)andinflationexpectations(π)11).stequals1whenprior-quarterCPIincreaserateexceeds

theattentionthresholdlevelsestimatedabove,and0otherwise.Underthismodel,thecoefficientscanchangewhenattentiontoinflationrisesastheinflationrateincreases.

(3)

yt=lc1+B1，jyt—j+Ω/2et]st

+lc2+B2，jyt—j+Ω/2et](1—st)

yt=lgt，wt，πt，π],

st:1ifπt—1持Y,and0otherwise.

13.Theresultsshowwhenattentiontoinflationrisesastheinflationratesrunbeyondacertainlevel,thepriceshockcanhavebiggerandlongerlastingimpactoninflationexpectations.TheresponseofinflationexpectationsagainstthepriceshockwasgreaterfortwotothreequarterswhenCPIgrowthexceedsthethresholdlevelestimatedabovethanwhenitisbelowthelevel.Additionally,theperiodthatpriceshockmeaningfullyimpactsoninflationexpectationsisestimatedthreetofourquarterslongerwhentheinflationrateishigherthanthethreshold.

Responseofinflationexpectationstowardspriceshockislargerandlastslongerinhighattentionregime

[Figure6]Changeofinflationexpectationsagainst1%priseinCPIgrowthrateunderdifferingattentionregime

(Basedonsearchthreshold(2.0%))

(Basedonperceptiongapthreshold(2.6%))

Note:Darkenedareameans68%confidenceinterval.

Source:Authors’calculations.

9)PotentialGDPrateiscalculatedusingHPfilter.

10)Hourlywageforworkersemployedbycompanieswithmorethanfivehires(=aggregatewage/totalhoursworked).

11)Consumers’inflationexpectationsfor1year.

。

9

Conclusion

14.Findingsupholdtheplausibilityof

increasedvolatilityininflationexpectations