The drivers of physical demand for gold

EvaluatingEstimatorConsistency Hausman1978proposesatestthatevaluatestheknownconsistencyofanestimator eoretically, theprocedureisbasedontheexpectationthatforastandardregressionofthetype:

s e t , o r a m i x t u r e o f b o t h ?

W h i l e m o s t r e s e a r c h e r s f o c u s o n t h e i n fl u e n c e ofmacroeconomi cvariablesonthepriceofgold,thispaperinvestigatestherelationshipb e t w

e e n asetofmacroeconomicvariablesandthephysicaldemandfortheyel lowmetalacrossam u l t i t u d e ofcountries.Differentpaneland non- panelmodelsareusedandtestedforgoodnessoffi t inordertoderiveempi ricalinsightsintothedriversofphysicaldemand.

termyieldsande c o n o m i c unce rtainty, whiletheexac t opposite iso bse rvedfor in du s trial golddemand , whereap o s i t i v e r e l a t i o n s h i

Resultsfortotalgolddemandindicateapositiverelationshipwithshort-p w i t h e c o n o m i c a c t i v i t y i s o b s e r v e d F u r t h e r m o r e , r e s u l t s i n d i

c a t e a r i s i n g l u x u r y demandlinkedtoincreasesinnationalwealth,andt owardsapositiverelationshipbetweeni n v e s t m e n t de man d fo r g o l d an

d b ot h i n fl a t i o n an d e c o n o m i c u n c e rt ai n t y M ore s p e c i fi c a l l y , we b r

e a k a c o m m o n m y t h b y p r o v i n g t h a t g l o b a l i n v e s t o r s p r o t e c t

t h e m s e l v e s f r o m i n fl a t i o n b y i n v e s t i n g intophysicalgoldrat herthanthroughbuyingjewellery.

Keywords:g o l d ; p h ys i c a l d e m a n d

However,theallegedsafetycharacterofgoldistheverydefinitionoftheasset’snature;o n e w o u l d t h i n k t h a t t h i s w o u l d o n l y

t r u l y c o m e t o l i g h t b y m e a n s o f

a physicalinvestmentintogold.Whileindeedanexposuretogoldthroughholdingitinaninvestor’sportfolioisbeneficialformultiplereasons(seeBaurandLucey(2010)andBattenetal

(2014)),therealsafetyofgoldliesinholdingitphysicallyasalastresortassetinextremesituations(StarrandT r a n (2008)).Financialresearchongoldcanb e dividedintod i ff e r e n t categories,eachconsideringdifferentaspectsofthepreciousmetals(O’Connoretal

(2015)).Averypredominantfieldisontherelationshipbetweengoldandinflation;hereanallegedrelationshipisbelievedtoexistbasedongold’sdefinitionasboth:aninternationalcurrencyandaproductionasset.Ifgoldisconsideredtobeaninternationalc u r r e n c y , anincreaseinexpectedinflationwouldleadtoareductionoftheanticipatedpurchasingpower,whichwouldleadtoinvestorsdrivingdowntheirproportionofcasha n d investingold,hencepushingthepriceupwards(Luceyetal

(2016)).Ontheotherhand,ifgoldisconsideredtobearegularasset,thenitspricewouldrisealongsidether a t e ofinflationsincethedefinitionofinflationisthatthedollarpriceofatypicalgoodr i s e s (Jaffe(1989)).Thereactiontoinflationfrominvestorsisthereforeproactivewhilethereactionfromproducersisreactive-

anobviousdifferenceinthebehaviourofdemandshouldthereforebeobservable.Asimilarreasoningcan beappliedforthesafehaventheoryproposedbyBaurandLucey(2010):goldoffers protectiontoinvestorsduringfinancialturmoils,whichshouldpositivelyimpactinvestorsdemandwhileitshould,ifanyt hing , diminishthedemandfromproducerswhoarefacinganeconomicdownturn.Again,adifferentimpactoninvestorandproducerdemandcanbeexpected

540 |

ICUEH2017

canonlybedonebylookingintotheannualsurveysofthepastdecadescomputedbytheGoldFieldsMineralServicesLtdandavailableonlyinphysicalcopiesattheirofficesinL o n d o n

Non-Governmentphysicaldemandforgoldcanbebrokendownintothreedifferentcategories:

specificm odels, we p ro po s e workingw ith d iff e re nt pa ne l a p p

r o a c h e s a n d f o r m a l l y t e s t w h e t h e r o r n o t p o o l e d O r d i n

a r y L e a s t S q u a r e s ( O L S ) procedurescouldaccuratelyfitthedatawhiledeletingcountry-specificeffects

Thechoiceofcountryismadeinregardtothecountry’srelativeimportanceonboththeofferand/

orthedemandmarket ofgold.The following countriesare considered: Australia,C a n a d a , C h i n a , E g y p t , G e r m a n y , I n d i a ,Italy,Japan,M e x i c o , Russia,S a u d i A r a b i a , SouthKorea,Switzerland,Thailand,Turkey,theUnitedKingdomofGreatBritainandNorthernIreland,andfinally,theUnitedStatesofAmerica

Thispapercontributestothefieldbybeingthefirsttolookatphysicaldemandforg o l d , breakingdownthedemandintodifferenttypes.Weworkwithacleanandthoroughmethodologyandderiveinsightfulresultsintotheeffectofmacroeconomicvariablesonthephysicaldemandforgold.Therestofthispaperisorganisedasfollows:Section2offersabriefoverviewofther e l a t e d literature inordertodefendthechoice ofdata,Section3 presentsthemethodology,whileSection4outlinesanddiscussestheempiricalresults.Finally,Section5concludes

2 LiteratureReviewandDataPresentation

FERGALLITERATURE

TheannualGoldFieldsMineralServices(GFMS)surveyspublishedThomsonReutersprovidea n o v e r v i e w o f t h e a m o u n t o f g o l d s u p p l

i e d a n d d e m a n d e d a c r o s s v a r i o u s countriesoverthepastcalendaryear

Plottingthedemandforgoldandsilverrespectively(Figures1)indicatesashiftinthedema nd towardsarisingimportanceoftheinvestmentside,thegraphsarealsorevealingthatjewelleryconsumptionisthemostimportantfactorindemandforphysicalgold

ItshouldbenotedthatFigures1iscomputedtakingintoaccounttheglobaldemandforgold.However,theregressionresultsinthispaperarecomputedconsideringonlyasubsetofcountries,whichwerechosenbecauseoftheirrelativeimportanceoneitherthesupplyorthedemandsideofthegoldmarketrespectively.Thecountriesare:Australia,Canada,China,Egypt,Germany,India,Italy,Japan,Mexico,Russia,SaudiArabia,SouthKorea,Switzerland,Thailand,Turkey,theUnitedKingdomofGreatBritainandNorthernIreland,andfinally,theUnitedStatesofAmerica.WhiletheresearchofStarrandTran(2008)istheonlypaperfocusedonthedriverso f physicaldemandforgold,itisindeedtheonlysourcethatcanbeusedasasteppingstonewhendecidingwhatdatatoconsider.InlinewithStarrandTran(2008),t h e CPI,theGDPandtheexchangeratetotheUSDollarhavebeenconsidered

Figure1:GlobalDemandforGoldbyTypeinTonnes

Thelevelofthenationalequityindiceshavealsobeenconsidered,aswellasbothlongt e r m andshortterminterestratesinordertogetafeelingforthestateoftheunderlyingeconomy.H e r e , t h e s h o r t te rm i n t e r e s t r

a t e s c o n s i d e r e d a re t h e 3 M o n t h s In te rb a n k LendingRate,while10YearsGovernmentBondYieldsareusedasaproxyforlongterminterestrates.ThedatasetisalsoaugmentedwithnarrowmoneysupplyaswellastheEconomicUncertaintyIndexifsuchanindexisavailableforthecountryconsidered

s t i n g proceduret o detectt h e presenceo f possibleheteroscedasticityi n l i n e a r r e g r e s s i o n m o d e l s b y b u i l d i n g u p o n a c l a s s i c

a l r e g r e s s i o n modeloftheform:

whereasetofresidualsuˆcanbeobtained,whileanOrdinaryLeastSquar

esprocedurew o u l d constraintheirmeanvaluetobe0.Inthecasethatthisassumptionmightfail,thevarianceoftheresidualsmightbelinearlyrelatedtoindependentvariablesandthemodelcouldbeexaminedbyregressingthesquaredresidualsontheindependentvariables(B r o o k s (2014)):

H0:αis a2= =αis a p =0 (4)andthereforez t0αis a=αis a1sothatσ t2=

h(αis a1)=σ2isconstant

3.2 EvaluatingEstimatorConsistency

Hausman(1978)proposesatestthatevaluatestheknownconsistencyofanestimator

eoretically,

theprocedureisbasedontheexpectationthatforastandardregressionofthetype:

twoassumptionsaremade:first,thattheconditionalexpectationsofε givenxiszeroandthatεhaveasphericalcovariancematrix.Morespecifica

H=(c

)

e c

ˆ

whereβ cisthecoefficientvectorfromtheconsistentes

wherey itis thedependentvariableandαis a,β1,andβ2are1+K1+K2parameters(Drukker

(2003)).Xitis a(1∗K1

)vectoroftime-varyingcovariatesandZiis a(1∗K2

)vectoroftime-invariantcovariates,whileµ iistheindividualleveleffectanditistheidiosyncr

regression( D r u k k e r (2003)).Adiscussionontheestimatorsoftheco

efficientsofthecovariatesXitandZicanbefoundinWooldridge(2002)andBaltagi(2013)

Assumingthatthereisnoserialcorrelationintheidiosyncraticerrors,orassumingthat ]=0foralls6=t,Wooldridge(2002)relieson

theresidualsobtainedfromare gress ion infirst-differencesoftheform:

y it y it1

yit =(Xit it1 X it1)1it

=X it1it

(11)

where∆isthefirst-differenceoperator(Drukker(2003)).TheWooldridge(2002)p r o

c e d u r e estimatestheparametersβ1byregressing∆y iton∆Xitandobtainsthe

residualse ˆ it.Incasetheitarenotseriallycorrelated,then

5(Drukker(2003)).Wooldridge(2002)therefore

regressestheresidualseˆ itontheirlagsandteststhatthecoefficien

panelcorrelationintheregressiono f eˆ iton eˆ it−1byadjustingthevariance-covariancematrixforclusteringatthepanellevel( D r u k k e r (2003))

i ii Q(,

m e ff e c t m o d e l r e l i e s o n t h e followingestimation:

(y it y i )=(1)(xit x i ){(1

)v i(it i )}

(17)

whereθisafunctionofσ v2andσ ε2sothatˆ=1 (Stata

Corporation (2013)) It should be noted that incaseσ v2= 0, implying



it it i

it i

ComprehensiveapplicationexamplesofpaneldatamodelscanbefoundinNaugesandThomas(2003)onwaterconsumption,inGlaserandWeber(2009)ontheeffectofpaststockpric e re turn ontra dingv olum e , in

? on vola tilitydyna m ic s fo r the S& P500,in

leveleffects arecorrelatedwiththevaluesofthelaggedvariabley(Stat aCorporation(2013)).I n ordertotacklethisproblem,thefollowingsec

fpredeterminedande x o g e n o u s covariates.β1andβ2arerespective

(1988)onvectorautoregressioncoefficientsinpanel-data,ArellanoandB o n d (1991)buildtheirmethodologyupontheGeneralisedMethodofMoments(GMM)andproposeaproceduredesignedfordatasetswithmanypanelsbutfewobservation periods,withtheonlyrequirementthatnoautocorrelationispresentintheidiosyncra

ticer r o r s TheGMMestimatorαis aˆisbasedonthesamplemoments

N−1Z0v¯sothat:

ˆ=arg

min (v'Z)A N

(Zv)=

y1'ZA N Zyy1'

ZA N Zy1 (2

2)

stepestimatorαis aˆ2isobtainedbysettingA N=

whereX¯isastacked(T−2)N∗kmatrixofobservationsonx¯ itandthe

alternativechoicesofA Nwillproduceone-stepandtwo-stepestimators.TheprocedureproposedbyArellanoandBond(1991)assuresestimatorconsistencybyr e m o v i n g panel-leveleffectst h r o u g h first-differentiationa n d b y f o r m i n g m o m e n t c o n d i t i o n s derivedfromthefirst-differenceerrorsofEquation21

Duet o i t s a v a i l a b i l i t y i n d i ff e r e n t e c o n o m i c a n d s t a t i s t i

c a l s o f t w a r e p a c k a g e s , t h e procedureproposedbyArellanoandBond(1991)hasbeenwidelyappliedinfinancialresea rc h ExamplescanbefoundinPodreccaandCarmeci(2001)oneconomicgrowth,inCastellsandSol´e-Oll

´e(2005)onregionalallocationofinfrastructureinvestment,inL i u (2006)onfinancialstructure,corporatefinanceandgrowthofmanufacturingfirmsinTaiwan,inNaud

´eandKrugell(2007)ondeterminantsofforeigndirectinvestmentinAfrica,andfinally,inChangetal

(2011)ontherelationshipbetweenmilitaryexpenditureandeconomicgrowth

3.5.2 LinearDynamicPanelDataModelingwithAddition

alMomentCon-ditions

InthelightofpossiblemodellimitationshighlightedbyArellanoandBover(1995),BlundellandBond(1998)proposearelatedestimatortoArellanoandBond(1991)usingadditionalmomentconditionsinassu

ringestimatorconsistencyundertheonlyadditionalconditionthatE=[v i∆yi2]

=0holdsforalliinEquation21.Buildinguponaclassicaldynami cpa

nel-datamodelaspresentedinEquation21,BlundellandBond(1998)arguethatthelagged-

levelinstrumentintheArellanoandBond(1991)estimatorbecomesweak

effectsv itothevarianceoftheidiosyncraticerrori t b e c o m e stoolarge(StataCorporation(2013))

Buildingupont h e w o r k o f A r e l l a n o a n d B o v e r ( 1 9 9 5 ) , B l u n

d e l l a n d B o n d ( 1 9 9 8 ) proposetousemomentsconditionsthatuselaggeddifferencesasinstrumentsforthe levelequationinadditiontothemomentconditionsoflaggedlevelsasinstrumentsforthedifferencedequations-

h e n c e res ulting in a na dditiona l m o m e nt e s tima tor Econo

metrically, theirprocedureresultsinaGMMestimatorαis aˆofthefollowing

´as que z and Royue la ( 2016) on thede te rm ina nts of inte rnational footballsuccess.andfinally,inDaSilvaandCerqueira(2017)onhouseholdelectricitypricesint h e EU

3.5.3 ALinearDynamicPanelDataModelallowingforAutoco rrelationintheIdiosyncraticErrors

Asmentionedabove,ArellanoandBond(1991)proposeone-stepandtwo-stepGMMe s t i m a t o r u s i n g

m o m e n t c o n d i t i o n s r e l y i n g o n l a g g e d l e v e l s o f

t h e d e p e n d e n t

a n d predeterminedvariables.ThisprocedureisaugmentedbyBlundellandBond(1998)who

leveleffectvariancev tt ot h e varianceo f t h e idiosyncratice r r o ri t b e c

o m e stoolarge

Bothproceduresrequirethattherebenoautocorrelationintheidiosyncraticerrors,h e n c e l i m i t i n g t h e r e fi e l d o f a p p l i c a t i o n A l i

n e a r d y n a m i c p a n e l d a t a p r o c e d u r e c a n howeverberespecifiedassuch,thatthecorrelationintheidiosyncraticerrorsfollowsal o w -

exogenousvariablesinx it ,andk2asthe numberofpredermined

variablesinw it inordert o rewrite E qua tion21as asetof T iequationsfor

ˆ

whereZdiis

thematrixoftheGMM-typeinstrumentforthedifferenceequationandZLiisthem a t r i x o f t h e G

SquaresDummyVariable(LSDV)procedureinitiallyproposedbyNickell(1

Formally,anLSDVproceduretransformsEquation21intoamatrixformat:

BuildingupontheexogenousselectionprocedureofBunandKiviet(2003)appliedtounbalancedpanels,Bruno(2005a)proposesamoregeneralapproximationprocedureforthecoefficients inEquation21valid forth

eobservationinterval[0,T].Bruno(2005a)definesaselectionindicato

rr itsuch thatr it= 1if(y it ,x it )isobserved,andr it=0otherwisei n ordertodefine

adynamicselectionrules(r it ,r i,t−1)thatonlyincludesobservationsforwhichboththecurrentvalueandthelaggedvalueareobservable.Formally:

SquaresDummyVariableprocedurecanbefoundinK i v i e t (1995),inKao(1999),andfinally,inBunandCarree(2005)

4 EmpiricalResults

Whilealargeamountofresearchexistsontheimplicationsandtheeffectsofcertainmacroeconomicvariablesonthepriceofgold,onlyoneformalinvestigationexistsonthedriversofphysicalcountrydemandforgold(StarrandTran(2008)).Inordertoprovide

mored e t a i l e d r e s u l t s , t h e d e m a n d f o r g o l d i s b r o k e n d o w n i n

t o d i ff e r e n t c a t e g o r i e s , dependingonthefinalusageofthepreciousmetal

4.1 TotalDemand

Thet o t a l d e m a n d i s t h e a g g r e g a t e d s u m o f g o l d d e m a

n d r e q u i r e d f o r j e w e l l e r y production,investmentpurposes,andindustrialproduction,mainlyinelectronics

Animportantquestionto raiseinthelightofthedataon handistoseehowitshouldb e modelled,theLagrangeMultipliertestbyBreuschandPagan(1979)isusedtotestforlinearmisspecificationinthemodel

ThetestresultsdisplayedinTable1advicetorejectthenullhypothesisandsuggestt h a t t h e varianceo f t h e u n o b s e r v e d fi x e d e ff e

c t s i s differentt h a n 0

-a pooledOLSregressio nmightthereforenotbethe-appropri-atemodeltouse

Ino r d e r t o b u i l d a g o o d m o d e l t h a t fi t s t h e p h y s i c a l g o l d d

e m a n d d a t a o f t h e 17countriesinthesystem,anessentialquestionistounderstandifthedatashouldbefittedinarandomeffect orafixedeffect model,relyingontheHausmanSpecificationTest (Hausman(1978))

Table2

HausmanSpecificationTest:TotalDemandforGold

(b) Fix ed

(B) Rando m

(b-B) Differen ce

(b) Fixed

(B) Rando m

(b-B) Differen ce

p h y s i c a l d e m a n d f o r g o l d a c r o s s countries

Inafirststep,alinearpaneldatamodelisruninwhichthecoefficientsareapproximatedbyafixedeffectsestimator

0.43509 0.12102 -3.60 0.000 -0.6739 -0.19628

-1.5966

0.5160 0

2.53 0.012 0.2603

6

2.11297 lnexchang

e

0.3301

-0.1834 8

-2.11 0.037

-0.1261

-0.00413 syield 0.0452

3

0.0198 1

2.28 0.024 0.0061

4

0.08431 lnequity 0.3714

0

0.1006 6

6

0.57003 lnuncertai

nty

0.3491 6

0.0799 5

2.62 0.010 2.9446

9

21.0091 2