Speculators and time series momentum in commodity futures markets

Uhl, Bjö n
A icle — Published Ve sion
Specula o s and ime se ies momen um in commodi y
u u es ma ke s
Re iew o Financial Economics
P o ided in Coope a ion wi h:
John Wiley & Sons
Sugges ed Ci a ion: Uhl, Bjö n (2025) : Specula o s and ime se ies momen um in commodi y u u es
ma ke s, Re iew o Financial Economics, ISSN 1873-5924, Wiley, Hoboken, NJ, Vol. 43, Iss. 2, pp.
213-230,
h ps://doi.o g/10.1002/ e.1228
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/330168
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by/4.0/
Re Financ Econ. 2025;43:213–230. wileyonlinelib a y.com/jou nal/ e
|
213wileyonlinelib a y.com/jou nal/ e
1
|
INTRODUCTION
Commodi ies can o e dis inc di e si ica ion bene i s o in es o s as hey end no o be highly co ela ed o adi-
ional asse classes. Howe e , commodi ies may also exhibi ce ain ea u es such as a e y high ola ili y o a nega-
i e oll yield which need o be conside ed when cons uc ing a po olio ha includes an alloca ion o commodi ies
(see Fe nandez- Pe ez e al.,2016; Le ine e al.,2018; Skiadopoulos,2012). Speci ically, adding indi idual commodi y
ma ke s long- only o an exis ing po olio may ha e limi bene i . Fo ins ance, Vinzelbe g and Aue (2014) show ha
Recei ed: 23 July 2024
|
Re ised: 4 Decembe 2024
|
Accep ed: 27 Decembe 2024
DOI: 10.1002/ e.1228
ORIGINAL ARTICLE
Specula o s and ime se ies momen um in commodi y
u u es ma ke s
Bjö nUhl
Facul y o Business Adminis a ion,
Uni e si y o Hambu g, Hambu g,
Ge many
Co espondence
Bjö n Uhl, Facul y o Business
Adminis a ion, Uni e si y o
Hambu g, Moo weidens aße 18, 20148
Hambu g, Ge many.
Email: [email p o ec ed]
Abs ac
In his pape , we analyze he ela ionship be ween specula o s in commodi y u-
u es ma ke s and gene ic ime se ies momen um (TSMOM) ade s as well as
he impac o his ela ionship on he subsequen TSMOM s a egy pe o mance.
We ind s ong empi ical e idence ac oss all commodi y ma ke s ha specula-
o s in commodi y ma ke s end o ade a TSMOM s a egy, which con i ms he
esul s ound by Boos and G ob (Jou nal o Financial Ma ke s 64, 100774). On
he basis o his esul , we also asce ain whe he he deg ee o such alignmen
has an impac on he pe o mance o he TSMOM s a egy. We ind ha he e is
weak, bu s a is ically signi ican and obus e idence o sugges ha he highe
he deg ee o alignmen be ween specula o s and a gene ic TSMOM s a egy, he
lowe he ealized pe o mance o ading TSMOM in hese ma ke s. Albei we
ind li le e idence ha his can be exploi ed in a dynamic in es men s a egy,
his nega i e ela ionship sugges s ha i a Commodi y T ading Ad iso (CTA)
ades commodi y u u es ma ke s which a e less commonly aded by o he
CTAs, hese ma ke s may no only inc ease he in e nal di e si ica ion o hei
und bu hese ma ke s may also ha e a highe TSMOM Sha pe a io by hem-
sel es. Consequen ly, ou analysis p o ides aluable insigh s in o imp o ing he
po olio cons uc ion o CTAs.
KEYWORDS
commi men o ade s, commodi ies, momen um, specula i e c owding
JEL CLASSIFICATION
G11, Q02
This is an open access a icle unde he e ms o he C ea i e Commons A ibu ion License, which pe mi s use, dis ibu ion and ep oduc ion in any medium, p o ided
he o iginal wo k is p ope ly ci ed.
© 2025 The Au ho (s). Re iew o Financial Economics published by Wiley Pe iodicals LLC on behal o Uni e si y o New O leans.
214
|
UHL
adding c ude oil u u es does no imp o e key s a is ics o ypical s ocks and bonds po olios. The e o e, i can p o e
use ul o no in es ‘buy&hold’ in o commodi ies bu ade hem in an ac i e in es men s a egy such as al e na i e
isk p emia, see e.g. Fue es e al.(2015), Mi e(2016) and Ma kwa e al.(2020). One o he mos s udied al e na i e
isk p emia is ime se ies momen um (TSMOM) also known as end- ollowing (Moskowi z e al.,2012),1 which is yp-
ically aded by so- called Commodi y T ading Ad iso s (CTA). Essen ially, TSMOM dic a es o hold a long exposu e in
an asse i i s p ice has been inc easing and ice e sa o sho posi ions. E b and Ha ey(2006) a gue ha whils hold-
ing commodi ies long- only does no necessa ily gene a e equi y- like e u ns he e is e idence o sugges ha his o i-
cally long- sho in es men s a egies such as momen um we e p o i able. Geo gopoulou and Wang(2017) show ha
much o mu ual und pe o mance in equi ies and also in commodi ies can be explained by TSMOM. Cla e e al.(2014)
ind ha combining TSMOM wi h c oss- sec ional momen um o commodi y u u es yields e en highe isk- adjus ed
e u ns. The popula i y o momen um ading is owed o i s pe sis ence (e.g. Geczy & Samono ,2016) and esilience
(e.g. Fol ice & Lange ,2015). Consequen ly, when cons uc ing po olios, S ad mülle e al.(2024) a gue in a ou o
a co e po olio which is enhanced by a sa elli e in es men wi h a ixed alloca ion o commodi y u u es momen um.
The p esence o TSMOM ade s in commodi y u u es ma ke s has been in es iga ed ex ensi ely. Reg essing he e u ns
o wo b oad CTA indices on he e u ns o a gene ic mul i- asse TSMOM s a egy Hu s e al.(2017) ind signi ican ly
posi i e coe icien es ima es which sugges s ha CTAs indeed p edomina ely ade TSMOM. Using u u es ac oss all asse
classes Fan e al.(2020) ind ha specula i e p essu e as a ac o can help o explain he c oss- sec ion o e u ns a e con-
olling o a numbe o al e na i e isk p emia ac oss all sec o s bu ixed income. Lu zenbe ge (2014) shows ha in addi-
ion o TSMOM p edic abili y also o he exogenous a iables can help o ecas ing he e u ns o commodi y u u es. Mo e
ecen ly, he academic deba e has ocussed on he ela ionship be ween specula o s in commodi y u u es and TSMOM.
Fo ins ance, Bo ga ds and Czudaj(2022) show ha he changes in specula i e open in e es ha e o ecas ing powe o he
e u ns o he unde lying ma ke as well as hei ela ionship wi h momen um. In he same di ec ion Boos and G ob(2023)
hypo hesize ha specula o s in commodi y u u es a e end- ollowe s and show ha eg essions which explain he changes
in ne specula i e open in e es by gene ic TSMOM posi ions ha e subs an ial explana o y powe ac oss commodi y ma -
ke s. They also es ima e he a e age momen um il e weigh s aded by specula o s using a penalized eg ession.
Gi en he ample e idence ha specula o s in commodi y u u es ade a end- ollowing s a egy, we add ess he
ollowing ques ion: Does he deg ee o specula i e c owding among CTAs in a commodi y ma ke ha e an impac on
he pe o mance o a end- ollowing s a egy in ha ma ke ? Ra he han using gene ic measu es specula i e c owd-
ing e.g. he ac ion o specula o s in all ade s, we use a no el measu e which is ailo ed o he speci ic p oblem. We
conside how closely aligned he ne specula i e open in e es in any gi en commodi y ma ke is o he posi ions o
a gene ic end- ollowing s a egy. Consequen ly, ou wo k builds upon Boos and G ob(2023) who empi ically show
ha he changes in ne specula i e open in e es can be explained by he changes in a gene ic end- ollowing signal.
They epo high ou - o - sample
R2
o all ma ke s and es ima e he weigh s which he a e age CTA would ha e applied.
Ou wo k is closely ela ed o hei s bu we ocus p edomina ely on he consequences o such alignmen be ween
specula o s and TSMOM ade s. Essen ially, we use he deg ee o alignmen as measu e o specula i e c owding.
The mo e specula o s ade TSMOM he close hei agg ega e ne posi ioning should be o a gene ic end- ollowing
posi ion. In con as o s anda d me ics, such as e.g. he p opo ion o specula o s o all open in e es , his measu e is
mo e in o ma i e as i speci ically es how simila he posi ions o specula o s o he posi ions o a TSMOM s a egy.
I specula o s we e on agg ega e aligned wi h a speci ic ading s a egy such as momen um his could po en ially ha e
a mo e p o ound impac han me ely gene ic measu es o specula i e ac i i y because alignmen equi es di e en
specula o s o ade in he same di ec ion a he same ime. The unde lying hypo hesis is ha i oo many CTAs end-
ollow a commodi y ma ke in a simila ashion he expec ed e u ns o his s a egy diminish. The possible easons
could be he inc eased isk o e e sals and ail isks (e.g. Ba oso e al.,2022; B own e al.,2022).2 The consequences o
such s a egy alignmen could bo h pose a sys emic isk bu could also impac he pe o mance o ading ha s a egy.
The majo i y o he li e a u e ocusses on he sys emic isk aspec . Whe he specula o s in commodi y ma ke s cause
p ice impac has been amously alleged by Mas e s and Whi e(2008) whose hypo hesis ha long- only specula o s in
commodi y u u es we e he main d i e behind he 2007–2008 p ice spike has been es ed empi ically by I win and
Sande s(2012) who ound no e idence in a ou o his hypo hesis. Boyd e al.(2018) e iew he impac o he inancial-
iza ion o commodi y u u es and conclude ha specula o s ha e li le impac on p ice dis o ions bu ha hey p edom-
ina ely p o ide liquidi y o hedge s. B ooks e al.(2015) in es iga e ex eme p ice mo es in commodi y ma ke s and ind
li le, i any, e idence ha hese we e caused by specula i e bubbles. On he con a y, using in o ma ional e iciency as
a measu e o ma ke quali y, Bohl e al.(2021) ind empi ical e idence ha he highe he specula i e ac i i y in a com-
modi y ma ke , he lowe on a e age he quali y o ha ma ke . Decomposing p ice shocks o commodi y u u es in o a
|
215
UHL
pe manen and a ansi o y componen , Haase e al.(2019) ind no empi ical e idence ha specula o s cause long- e m
p ice impac . Haase e al.(2016) summa ize a la ge numbe o s udies on his subjec in a me a- analysis and ind ha
he e is no conclusi e e idence nei he o no agains he hypo hesis ha specula o s d i e commodi y ma ke p ices.
Howe e , Haase e al.(2017) ind some empi ical e idence ha specula o s in commodi y ma ke s can G ange - cause
ma ke ola ili y, whils Kim(2015) ind ha specula i e u u es ading does no des abilize commodi y ma ke s. In
ano he me a- analysis o a la ge numbe o published s udies Wimme e al.(2021) ind ha he null hypo hesis o non-
causali y canno be ejec ed.
Mo e owa d he di ec ion o ou wo k a e s udies ha ela e he deg ee o specula i e c owding o s a egy pe o -
mance. Fo ins ance, Bal as(2019) in es iga es he impac o c owding in isk p emia s a egies including momen um. His
esul s sugges ha he impac o c owding can be ei he bene icial o no depending on whe he he in es men s a egy is
con e gen (e.g. mean- e e sion ading) o di e gen (e.g. momen um ading). Focussing speci ically on end- ollowing
Bollen e al.(2021) in es iga e whe he hose CTAs ha pe o m mo e simila o hei pee s unde - o ou pe o m hem.
They ind ha simila i y o he pee g oup is associa ed on a e age wi h highe pe o mance. Building a he e ogenous
agen model, He and Li(2015) in es iga e he ela ionship be ween di e en ypes o ade s. They ind ha when mo-
men um ade s a e mo e ac i e in a ma ke , momen um s a egies wi h a sho ho izon end o s abilize he ma ke and
may be mo e p o i able. Ca e and Re o edo- Giha(2023) a ibu e a degene a ion in CTA e u ns in commodi y u u es
o he inancializa ion o such ma ke s. In con as o he a o emen ioned analyses, we ocus on he di e ences be ween
indi idual commodi y ma ke s wi h ega ds o he ela ionship be ween specula i e c owding and end- ollowing pe o -
mance. This can p o ide a use ul guide o in es men manage s o adjus hei weigh ing (o e en inclusion) o ma ke s
in hei s a egy.
In Figu e1, an example is gi en o illus a e he hypo hesized ela ionship be ween specula o s and CTAs. I shows
he ime se ies o he (no malized) ne specula i e posi ion o he A abica Co ee u u es oge he wi h he co espond-
ing (no malized) posi ions o a gene ic TSMOM s a egy. I can be seen ha hese appea o be closely ela ed.
In es iga ing he ading beha io in u u es con ac s using he CFTC commi men o ade s da a has been done in
inancial ma ke s, oo. Fo ins ance, using he S&P 500 index u u e Smales(2016) inds ha specula o s and small ad-
e s e eal some abili y o o ecas u u e e u ns.
The con ibu ion o his wo k can be summa ized as ollows. Fi s , a e eplica ing some o he key esul s o Boos and
G ob(2023), we in oduce he co esponding eg ession es ima es as measu e o alignmen among specula o s. This measu e
is speci ically ailo ed o he empi ical obse a ion ha he p edominan pa o specula o s in commodi y u u es pu sue
a end- ollowing s a egy. Second, we ind ha he pe o mance o end- ollowing a commodi y ma ke is nega i ely co -
ela ed o he deg ee o he es ima ed alignmen . Howe e , we also ind ha his e ec appea s o be p edomina ely con-
empo aneous hough and hus i is unlikely ha i can be exploi ed p o i ably in a dynamic ading s a egy. This nega i e
ela ionship sugges s ha i a CTA adds new commodi y ma ke s which a e less commonly aded by o he CTAs ela i e
o o he specula o s, such ma ke s do no only inc ease he in e nal di e si ica ion o he po olio bu may also end o
FIGURE 1 Example o simila i y be ween ne specula i e OI and end- ollowing posi ion. Time se ies o end- ollowing posi ions and
he % ne specula i e open in e es scaled by hei own ola ili y ( ixed) o A abica Co ee u u es.
216
|
UHL
indi idually pe o m supe io in a TSMOM s a egy. This inding may explain why mo e and mo e CTAs launch so- called
‘al e na i e ma ke s’ p og ams which ade mo e exo ic and less commonly aded commodi y u u es ma ke s.
This pape is s uc u ed as ollows. In he nex sec ion, we discuss he da a and me hodology ha we use o ou anal-
yses. In he hi d sec ion, some s a is ics a e p o ided ha allow o be e unde s and he ela ionship be ween he di e -
en ca ego ies o commodi y ade s. The ou h sec ion con ains he main empi ical esul s. The i h sec ion p o ides
some obus ness analyses o suppo he key indings. In he six h sec ion, we discuss whe he he indings can be used
p o i ably in an in es men s a egy. The inal sec ion concludes.
2
|
DATA AND EMPIRICAL STRATEGY
2.1
|
Da a
We collec da a o n = 26 commodi y u u es ma ke s. The sample pe iod is om Janua y 2006 o Decembe 2023 and is
limi ed by he his o ical a ailabili y o he disagg ega ed open in e es (OI) da a om he Commodi y Fu u es T ading
Commission (CFTC).3 All da a a e e ie ed om Bloombe g.
The con ac s wi h hei basic p ope ies a e shown in Table1. Mos ma ke s a e in he ag icul u als sec o . Wi h he
excep ions o lean hogs and eede ca le all con ac s a e physically se led.
TABLE 1 The able shows he con ac speci ica ion o he commodi y u u es se ies we conside .
Name Ticke Cash se led Exchange Sec o Sub- sec o
C ude CL1 N New Yo k Me can ile Exchange Ene gies Ene gies
Na Gas NG1 N New Yo k Me can ile Exchange Ene gies Ene gies
Hea ingOil / ULSD HO1 N New Yo k Me can ile Exchange Ene gies Ene gies
Gasoline XB1 N New Yo k Me can ile Exchange Ene gies Ene gies
Coppe HG1 N Commodi y Exchange, Inc. Me als Indus ials
Gold GC1 N Commodi y Exchange, Inc. Me als P ecious
Sil e SI1 N Commodi y Exchange, Inc. Me als P ecious
Pla inum PL1 N New Yo k Me can ile Exchange Me als P ecious
Palladium PA1 N New Yo k Me can ile Exchange Me als P ecious
Whea W 1 N Chicago Boa d o T ade Ags G ains and Oilseeds
Co n C 1 N Chicago Boa d o T ade Ags G ains and Oilseeds
Soybean S 1 N Chicago Boa d o T ade Ags G ains and Oilseeds
SoybeanMeal SM1 N Chicago Boa d o T ade Ags G ains and Oilseeds
SoybeanOil BO1 N Chicago Boa d o T ade Ags G ains and Oilseeds
Co ee KC1 N ICE Fu u es US So s Ags So s
Suga SB1 N ICE Fu u es US So s Ags So s
Co on CT1 N ICE Fu u es US So s Ags So s
Cocoa CC1 N ICE Fu u es US So s Ags So s
Lumbe LB1 N Chicago Me can ile Exchange Ags So s
Juice JO1 N ICE Fu u es US So s Ags So s
Li eCa le LC1 N Chicago Me can ile Exchange Ags Li es ock
LeanHogs LH1 Y Chicago Me can ile Exchange Ags Li es ock
Feede Ca le FC1 Y Chicago Me can ile Exchange Ags Li es ock
Ha dWin e Whea KW1 N Chicago Boa d o T ade Ags G ains and Oilseeds
Sp ingWhea MW1 N Minneapolis G ain Exchange Ags G ains and Oilseeds
Rough ice RR1 N Chicago Boa d o T ade Ags G ains and Oilseeds
Sou ce: Bloombe g.

|
217
UHL
Fo each o he u u es con ac s, we ob ain daily end- o - day closing p ices. As each u u es con ac expi es a a ixed
da e, we gene a e a con inuous se ies o p ices o each ma ke by olling e e y con ac ei he on i s las ading da e o
on he i s business day o he expi a ion mon h, whiche e is ea lie . Ra io adjus men s a e applied backwa d in his o y.
Wi h hese adjus men s, he e is no p ice- jump om olls bu he ca y e u n is ea ned (o paid) con inuously (Koijen
e al.,2018). Thus, using a io adjus men s o he olls be ween con ac s allows o compu e loga i hmic e u ns simply
as di e ence be ween he loga i hm o he adjus ed p ices.
Fu u es do no equi e a ull in es men bu a e aded on ma gin accoun s o which only an ini ial and a a ia ion
ma gin ha e o be pos ed. We assume ha he u u es posi ion is ully colla e alized, so ha he e u ns can be in e p e ed
as s anda d excess e u ns.
The OI da a is o iginally p o ided by he CFTC on a weekly basis in he ‘Disagg ega ed Commi men o T ade s
Repo ’. The epo con ains he agg ega e long and sho posi ioning o epo able ma ke pa icipan s in u u es
ma ke s which a e classi ied in ou ca ego ies: swap deale s, p oduce s, money manage s and o he s. In plain e ms,
‘swap deale s’ a e p ima ily in es men banks who ac as ma ke make s; ‘p oduce s’ a e all ma ke pa icipan s who
a e in ol ed wi h he physical unde lying o he u u es con ac , i.e. buye s o selle s o he ac ual commodi y; ‘money
manage s’ a e asse managemen i ms who hold posi ions o specula e on he u u e p ice de elopmen and hus a e
subsequen ly labeled ‘specula o s’; inally, ‘o he s’ a e any ade s who do no all in o any o he p e ious ca ego ies.
The OI da a is published on F idays o Tuesday's holdings. As we a e in e es ed in explaining alignmen we use he
Tuesday imes amps. We use he u u es- only da a and igno e op ions as CTAs end ade he o me . The o iginal
OI da a is sampled in lo s and is mapped o USD posi ions by mul iplying he numbe o lo s wi h he co esponding
con ac alue.4
Ne and g oss OI o each commodi y ma ke
i=1, …,n
and each ca ego y o ade ,
ca ∈{p od, swap, spec, o he }
,
a e de ined as
espec i ely and can be in e p e ed as he ne and g oss agg ega e posi ioning o specula i e ade s in a gi en ma ke . Fo
compa abili y we no malize bo h by he agg ega e g oss OI, i.e. o ne OI we ha e
In o de o ha e well- de ined s a is ics, bo h
OI%ne ,ca
i,
and
OI%g oss,ca
i,
a e no malized by he g oss OI.
2.2
|
Gene ic momen um signals and posi ions
The e a e wo main ypes o momen um s a egies: ime se ies momen um (TSMOM) which akes di ec ional ne expo-
su es, see Moskowi z e al.(2012), and c oss- sec ional momen um (XSMOM) which is gene ally ne cash o isk neu al,
see Jegadeesh and Ti man(1993). Albei a numbe o wo ks ha e discussed he la e (e.g. Shen e al.,2007) we ocus on
he o me because i is he p edominan s a egy employed by many CTAs (Hu s e al.,2013).
To compa e he ne specula i e posi ioning wi h he posi ions o a gene ic TSMOM s a egy we need o speci y a base-
line momen um model. We de ine he TSMOM signal
si
a ime
=1, …,T
o each ma ke
i=1, …,n
, in acco dance
wi h Le ine and Pede sen(2016) as he ma ke e u n o e he p eceding 260 business days no malized by he co e-
sponding es ima e o ola ili y, i.e.
Fan and Zhang(2024) highligh he impo ance o isk managing he posi ions in indi idual commodi y ma ke s in
isk p emia s a egies. Thus, he posi ions o a TSMOM s a egy a e ypically scaled in e sely p opo ional o ola ili y
(see e.g. Ha ey e al.,2021),
(1)
OIne ,ca
i, =OI
long,ca
i, −OI
sho ,ca
i,
and
OIg oss,ca
i,
=OIlong,ca
i,
+OIsho ,ca
i,
,
(2)
OI
%ne ,ca
i, =
OI
ne ,ca
i,
∑
ca ∈{p od,swap,spec,o he }
OIg oss,ca
i,
.
(3)
s
i =
1yea
i
𝜎 1yea
i
.
218
|
UHL
whe e we se
𝜎 a ge =10%
pe annum 5 and we also use he RiskMe ics(1996) s anda d o
𝜎 i
.
In he obus ness sec ion, we discuss he sensi i i y o he key esul s o hese speci ica ions. In pa icula , we es
he impac o applying a o ecas unc ion o he signal, e.g. he sign unc ion which is o en used in he li e a u e (e.g.
Moskowi z e al.,2012), and he impac o using a ‘slowe ’ posi ion scaling, which may be ele an in p ac ice.
2.3
|
Measu ing he alignmen be ween specula o s and momen um ade s
We hypo hesize ha specula o s in commodi y u u es ma ke s ade TSMOM as desc ibed by he gene ic s a egy in
he p e ious subsec ion. To es his hypo hesis empi ically, we ollow Boos and G ob(2023) and es ima e a eg es-
sion whe e he changes in he ne specula i e OI a e explained by changes in he posi ions o a gene ic TSMOM s a -
egy. While we base ou p ima y empi ical model on hei s, we de ia e in h ee ega ds, hough. Fi s , in his wo k,
he end- ollowing posi ion is ola ili y- scaled, which has been highligh ed by Kim e al.(2016) as a signi ican ly
con ibu ing ac o o TSMOM pe o mance and consequen ly is used in p ac ice by many CTAs, which is ele an
o his wo k. Second, we use weekly da a on bo h sides o he eg ession as we do no aim o back ou a e age il e
weigh s bu explain changes in he specula i e posi ioning. Las ly, we no malize he a iables o bo h sides o he e-
g ession by hei co esponding s anda d de ia ion. This ende s bo h sides o he eg essions dimensionless so ha
he coe icien es ima es can be in e p e ed in e ms o s anda d de ia ions and a e consequen ly compa able ac oss
ma ke s. Thus, he model speci ica ion eads
whe e
OI%ne ,spec
i
is de ined in (2) and he posi ions o he TSMOM s a egy
pi
a e de ined in (4). Va iables which a e
supe - sc ip ed by a ilde
⋅
deno e he co esponding a iable di ided by i s own ola ili y, i.e. o ins ance o he TSMOM
posi ions he no maliza ion is

p
i
=p
i
∕

s d
(
p
i )
.
In addi ion, we also use a speci ica ion in le els, which en ails he issue o high au oco ela ion in he eg esso bu
may s ill p o ide some addi ional in o ma ion. Thus, we also es ima e
I should be no ed ha he le els eg ession (6) uses highly au oco ela ed a iables which by cons uc ion a e s a-
iona y hough. The end- ollowing posi ions as de ined in (4) use a scaled e sion o he lagged 1- yea ma ke e u n.
The di e encing in (5) has he ad an age o emo ing he high au oco ela ion bu also emo es he le el in o ma ion
hough. Clea ly, bo h eg essions a e ela ed bu s ill ha e hei dis inc ad an ages and disad an ages so ha we use bo h
in he empi ical analysis.
A s a is ically signi ican es ima e o
𝛽Δ
1
o
𝛽1
, espec i ely, would sugges ha he agg ega e specula o s in ha ma -
ke beha e simila o a ime se ies momen um ade as de ined in sec ion2.2. We also compu e ou - o - sample
R2
OS
(Campbell & Thompson,2008), which compa es he mean squa ed e o (MSE) om he p edic ion o he eg ession o
he MSE o using he his o ical a e age as p edic ion, i.e. o ins ance o (5)
whe e
yi
=Δ

OI
%ne ,spec
i
o (5) and y
i
=
OI
%ne ,spec
i
o (6) and he p edic o s

y
and
y
use da a up o ime
−1
. As he p edic-
ion is ou - o - sample a minimum es ima ion window size
𝜏
o he eg ession needs o be speci ied which we se o he equi -
alen o 4 yea s o obse a ions. All s a is ical signi icance es s o hese eg essions a e based on he s a iona y boo s ap by
Poli is and Romano(1994).
(4)
pi =
𝜎 a ge
𝜎
i
si
.
(5)
Δ
OI
%ne ,spec
i
=𝛽Δ
0,i
+𝛽Δ
1,i
Δ
p
i
+𝜀
Δ
i
(6)

OI%ne ,spec
i
=𝛽
0,i
+𝛽
1,i

p
i
+𝜀
i .
(7)
R
2
i,OS =1−∑
T
=𝜏�yi −
yi �
2
∑
T
=𝜏�
y
i
−y
i �
2,i=1, …,
n
|
219
UHL
3
|
MARKET STRUCTURE
3.1
|
Who ades wi h whom?
We i s asce ain which ca ego ies o ade s in commodi y u u es ade wi h each o he . To his end, we co ela e he
con empo aneous weekly changes in he co esponding ne open in e es o each combina ion o ca ego ies. A nega i e
ela ionship sugges s ha as one ca ego y o ade s builds up a ne posi ion, he o he ca ego y educes i s posi ion and
hus on agg ega e can be in e p e ed as ading ac i i y be ween hese wo ca ego ies. Table2 shows he a e age co -
ela ions by sec o . In all sec o s, he mos p onounced co ela ion is ound be ween specula o s and p oduce s, which
economically sugges s ha specula o s end o be agains he hedges om p oduce s. In ag icul u als, his co ela ion
has by a he la ges magni ude whils in he o he wo sec o s all co ela ions a e clea ly nega i e which indica es specu-
la o s also ade wi h swap deale s and o he s ade s.
O e all, hese esul s a e in line wi h Boos and G ob(2023) who use a a iance decomposi ion and ind ha p oduce s
a e he dominan coun e pa y o TSMOM ade s.
3.2
|
Posi ioning and pe o mance by ca ego y o ade
Nex , we in es iga e he a e age posi ioning o each ca ego y o ade s and he associa ed pe o mance. Figu e2 shows
he box plo s o he agg ega e ne OI pe cen ages. On a e age p oduce s ha e held sho posi ions and hence ha e in
endency hedged hei exis ing physical long exposu es whils ma ke make s and specula o s ha e been p edomina ely
long and hus aken he opposi e side o he hedges.
We now es ima e he agg ega e pe o mance o each ca ego y o ade s. The ne OI o each ca ego y se es as p oxy
o he agg ega e posi ioning which we assume o be app oxima ely cons an o e he ollowing week due o he lack
o highe equency da a. We compa e he pe o mance o all ca ego ies oge he wi h he TSMOM s a egy. Fo a ai
compa ison we allow he la e o upda e i s posi ions also only once a week on Tuesdays. The posi ions o all ma ke s
a e scaled wi h an ex an e o ecas o ola ili y o a ge an annualized le el o 10%. The esul s a e shown in Table3.
All ealized ola ili ies a e well below 10% due o he di e si ica ion wi hin each o he po olios. O e all, he posi-
ions o he gene ic TSMOM s a egy ou pe o m all ca ego ies o ade s. Wi h he excep ion o ene gies, he same is
ue o all indi idual sec o s. In e es ingly, specula o s ac ually ealize a nega i e pe o mance o e all whils he o he
ca ego ies o ade s and he TSMOM s a egy ealize posi i e pe o mances o e he sample pe iod. The eason o he
poo pe o mance o specula o s is a posi i e bias in he specula o s' posi ions compa ed o he o he ca ego ies o ade s,
see Figu e2, and also compa ed o he TSMOM s a egy. An explana ion o his bias could be he p esence o long- only
commodi y specula o s who hold ai ly s a ic long exposu es o p o ide in es o s wi h p o ec ion agains commodi y
p ice in la ion by in es ing in o in alignmen wi h commodi y indices, o ins ance he S&P GSCI. By con as , TSMOM
models ha e held o e all sho posi ions on a e age o each ma ke . This is a sensible esul gi en ha mos commodi ies
end o ade in con ango, so ha he nega i e oll yield will cause a nega i e d ag in he signal de ini ion(3).6
4
|
SPECULATORS AND MOMENTUM TRADERS
4.1
|
A e specula o s momen um ade s?
We hypo hesize ha specula o s on agg ega e ade a s a egy ha is simila o he gene ic TSMOM s a egy p e-
sen ed in sec ion2.2, which we es in wo s eps. Fi s , we es o posi i e au oco ela ion in he changes o he ne
TABLE 2 A e age Pea son co ela ions o he weekly changes in ne open in e es among di e en ca ego ies o ade s.
Sec o P od/swap P od/spec P od/o he Swap/spec Swap/o he Spec/o he
All −0.04 −0.81 0.04 −0.21 −0.04 −0.38
Ags −0.20 −0.86 −0.02 −0.00 −0.09 −0.32
Ene gies 0.05 −0.66 0.14 −0.45 0.01 −0.58
Me als 0.44 −0.77 0.15 −0.74 0.05 −0.45
220
|
UHL
specula i e OI, which would imply ha he ade lows o specula o s in commodi y u u es a e ( o some ex end)
p edic able. This is a key ea u e o a end- ollowing sys em because TSMOM gene a es au oco ela ed signals and
posi ions by cons uc ion as i passes his ea u e on om he unde lying ma ke .7 Figu e3 shows he i s o de au-
oco ela ions o he weekly changes in he ne specula ion OI o each ma ke wi h 90%- con idence bands using he
s a iona y boo s ap o Poli is and Romano(1994). Fo all commodi y ma ke s, we ind ha he i s o de au oco -
ela ion is s a is ically signi ican ly posi i e. The a e age es ima e is lowes o ene gies and highes o ag icul u al
ma ke s.
As a second s ep, we es ima e he alignmen coe icien s in he eg essions (5) and (6), which desc ibe he mag-
ni ude o which he (changes in he) ne specula i e posi ions a y wi h he (changes in he) posi ions o a gene ic
end- ollowing sys em as de ined in sec ion2.2. Table4 summa izes he esul s by ma ke o bo h weekly changes
and le els. The sec o a e ages and an o e all a e age a e also shown a he bo om o he able. The es ima ed be as
FIGURE 2 A e age ne posi ioning o ade s by ca ego y. The igu es shows ‘box- and- whiske s’ plo s he dis ibu ion o he p opo ion
o ne open in e es (
OI%ne ,ca
) o each ca ego y o ade s
ca ∈{p od, swap, spec, o he }
. Each o he ‘boxes’ shows he median as well as he
i s and hi d qua iles. The ‘whiske s’ a e de ined by a dis ance o 1.5× he in e qua ile- ange om he nea es qua ile. Ou lie s, i any,
we e shown as indi idual do s ou side he whiske s. The as e isks ma k he means.
TABLE 3 The able shows he (cos - ee) pe o mance es ima es o he ca ego ies as well as o a gene ic TSMOM s a egy.
Sec o S a is ic P od Swap Spec O he TSMOM
All Re u ns (%) 0.11 0.41 −0.44 0.82 1.06
Vola ili y (%) 4.37 3.54 3.50 2.97 3.58
Sha pe a io 0.02 0.12 −0.12 0.28 0.30
Ag icul u als Re u ns (%) 0.41 0.03 −0.95 1.15 0.90
Vola ili y (%) 4.59 4.69 3.85 3.67 3.83
Sha pe a io 0.09 0.01 −0.25 0.31 0.24
Ene gies Re u ns (%) −0.31 0.22 1.80 0.50 1.74
Vola ili y (%) 6.99 4.83 6.12 6.11 7.42
Sha pe a io −0.04 0.05 0.29 0.08 0.23
Me als Re u ns (%) −0.80 1.84 −0.43 −0.02 1.06
Vola ili y (%) 7.65 5.19 6.94 5.45 6.61
Sha pe a io −0.10 0.35 −0.06 −0.00 0.16
No e: All ma ke s a e equally weigh ed and ola ili y scaled o an ex an e annualized ola ili y o 10%.
|
227
UHL
Se up and model/sec o All Ag icul u als Ene gies Me als
Simple eg ession ull sample −0.10 N/A N/A N/A
Simple eg ession annual sub- samples −0.63*** −0.70*** −0.39*** −0.63*
Fixed e ec s eg ession annual sub- samples −0.63*** −0.70*** −0.34 −0.66
Fixed and ime e ec s eg ession annual sub- samples −0.63*** −0.70*** −0.23 −0.14
Fo ecas : bina y, ola ili y scaling hal - li e: 260 days
Simple eg ession ull sample 0.15 N/A N/A N/A
Simple eg ession annual sub- samples −0.57*** −0.61*** −0.29*** −0.75***
Fixed e ec s eg ession annual sub- samples −0.59*** −0.65*** −0.32 −0.74*
Fixed and ime e ec s eg ession annual sub- samples −0.59*** −0.63*** −0.43 0.08
Fo ecas : oll o e , ola ili y scaling hal - li e: 11.2 days
Simple eg ession ull sample −0.07 N/A N/A N/A
Simple eg ession annual sub- samples −0.84*** −0.88*** −0.59*** −0.98***
Fixed e ec s eg ession annual sub- samples −0.89*** −0.93*** −0.52 −1.13**
Fixed and ime e ec s eg ession annual sub- samples −0.89*** −0.94*** −0.39 −0.65
Fo ecas : oll o e , ola ili y scaling hal - li e: 260 days
Simple eg ession ull sample 0.13 N/A N/A N/A
Simple eg ession annual sub- samples −0.84*** −0.86*** −0.62*** −0.98***
Fixed e ec s eg ession annual sub- samples −0.87*** −0.91*** −0.57** −1.04***
Fixed and ime e ec s eg ession annual sub- samples −0.87*** −0.91*** −0.73* −0.56*
No e: In Panel A he esul s o he
Δ
- alignmen be as a e epo ed and in Panel B he esul s o he le els alignmen be as a e epo ed. Only he slope
coe icien es ima es a e epo ed. S a is ical signi icance is deno ed by ‘*’ o p- alues < 10%, ‘**’ o p- alues < 5% and ‘***’ o p- alues < 1%.
TABLE 7 (Con inued)
TABLE 8 Back es esul s: We epo he Sha pe a ios o an equally weigh ed TSMOM s a egy using only high and only low
alignmen be a ma ke s.
TSMOM- sha pe TSMOM- sha pe
p- Value
TSMOM- sha pe TSMOM- sha pe
p- Value
Low
𝚫
- alignmen
be a
High
𝚫
- alignmen
be a
Low le els-
alignmen be a
High le els-
alignmen be a
Look- ahead
All .55 −.13 .00 .47 −.01 .01
Ags .48 −.14 .00 .48 −.17 .00
Ene gies .23 .24 .51 .16 .24 .64
Me als .17 .03 .31 .26 .17 .38
Ou - o - sample
All .32 .02 .07 .15 .29 .73
Ags .32 −.12 .05 −.00 .26 .83
Ene gies .20 .11 .38 .02 .29 .90
Me als .22 −.05 .21 −.09 .23 .86
No e: The op ows con ain he esul s o a look- ahead analysis whe e he be as a e es ima ed o e he same pe iod o e which he po olio is o med. The
bo om ows con ain he esul s o he ou - o - sample analysis whe e he alignmen be as a e es ima ed o e he p e ious yea and he po olio is o med o e
he nex yea . The le mos h ee columns use he alignmen be as om he eg ession in di e ences whils he igh mos h ee columns use he alignmen
be as om he le els eg ession. The p- alues a e compu ed o he null hypo hesis ha he high alignmen be a po olio has a Sha pe a io a leas as high as
he po olio wi h he low alignmen be as.

228
|
UHL
pe o mance also exis bu a e no s a is ically signi ican . Fo he be as om he le els eg ession, he e is no e idence o
sugges ha he e ec can be p o i ably exploi ed.
Consequen ly, he e ec ha he alignmen be a nega i ely co ela es wi h TSMOM pe o mance is p edomina ely a
con empo aneous e ec and has limi ed, i any, p edic i e powe . ag icul u als may be an excep ion hough. O e all, we
belie e ha he e ec canno be exploi ed in a dynamic in es men s a egy. Howe e , he s a ic na u e o he e ec could
possibly be ha es ed s a ically by including p edomina ely commodi y u u es ma ke s in a TSMOM po olio ha a e
less commonly aded by o he CTAs.
7
|
CONCLUSIONS
In his pape , we in es iga e he ela ionship be ween he agg ega e ne posi ions o specula o s and he posi ions o a
gene ic momen um s a egy ac oss a b oad ange o commodi y u u es ma ke s. We ind ha he ime se ies o bo h
ypes o posi ions a e closely aligned ac oss all in es iga ed commodi y ma ke s. The e idence p e ails o bo h he le els
eg ession and in di e ences. The high alignmen sugges s some le el o specula i e, s a egy- speci ic c owding in com-
modi y u u es ma ke s. Using he es ima ed alignmen coe icien s, we also documen a weak bu s a is ically signi ican
endency o ime se ies momen um pe o mance o degene a e when he specula o s in he unde lying ma ke a e mo e
aligned wi h momen um ade s. Albei he co ela ions a e nega i e ac oss all sec o s, he e idence is s onges in ag i-
cul u al commodi ies. In an in es men exe cise he s a is ical es s a e alida ed in a his o ical back es – bo h in- sample
and ou - o - sample. Al hough he e is also some e idence o suppo he la e , he e ec appea s o be p edomina ely
con empo aneous and hus i may no be possible o bene i om his e ec in a dynamic ading s a egy. Howe e , he
nega i e impac on momen um pe o mances sugges s ha ‘ha de - o- access’ o less commonly end- ollowed com-
modi y u u es ma ke s may no only p o ide di e si ica ion bene i s o a CTA p og am also be mo e p o i able on a
single- asse basis.
DATA AVAILABILITY STATEMENT
The da a in his pape a e downloaded om Bloombe g and a e hus subjec o licensing cons ain s.
ORCID
Bjö n Uhl h ps://o cid.o g/0000-0002-8018-4481
Endno es
1 Th oughou his pape we use he e ms ‘momen um’ and ‘ end- ollowing’ in e changeably. Essen ially, he e m ‘momen um’ co e s bo h
c oss- sec ional momen um and ime se ies momen um, which co esponds o end- ollowing. As we do no use he o me , he la e is he e also
simply e e ed o as momen um.
2 I should be no ed ha o c oss- sec ional equi y momen um Ba oso e al.(2022) ind ha c owding does no alone explain inc eased ail isk.
3 Some mino limi a ions apply o he his o ical da a, see CFTC(2022) o de ails.
4 Con ac alue is de ined as he quo ed p ice imes he co esponding con ac mul iplie .
5 The ac ual a ge
𝜎 a ge
does no ma e as we conside he agg ega e specula o s and no one speci ic CTA.
6 The end- ollowing signals in (3) use o al excess e u ns, which can be decomposed in o a spo and a ca y componen (see Koijen e al.,2018).
Thus, i he e is no secula end in he spo componen , he nega i e ca y o hose commodi y ma ke s ading in con ango will cause he end-
ollowing signals o be sho on a e age.
7 To see his, we use a gene alized de ini ion o he TSMOM signal in (3) which is he weigh ed sum o isk adjus ed e u ns,
s =∑swsR −s
,
R = ∕𝜎
and
ws≥0
. This de ini ion coincides wi h he one in Ha ey e al.(2021) i one assumes homoscedas ici y and and wi h he signal de i-
ni ion in (3) i one u he assumes equal weigh s
ws=1∕S
. Now we e- w i e he changes in his momen um signal as
Δ
s
=
∑s
w
s�
R
−s
−R
−1−s�
.
Using he baseline case o
ws=1∕S
. we ge
Δ
s
=
(
R
−R
−S−1)
∕
S
. Assuming simple au oco ela ion in e u ns,
co (
R
,R
−s)
=𝜌
s
, we
ge
co (
Δs
,Δs
−1)
=
(
2𝜌−𝜌
S
−𝜌
S+2)
∕S
2
≈2𝜌∕S
2
no ing ha
𝜌≫𝜌
S
. Wi h
a (
R
)
=
1
we ha e
a (
Δs
)
=2
(
1−𝜌
S+1)
∕S
2
and hus
co (
Δs
,Δs
−1)
≈
𝜌
.
8 These esul s a e no epo ed he e o b e i y and a e a ailable upon eques .
9 The ecommended smoo hing coe icien o daily da a is 0.94 which ansla es o a hal - li e o app oxima ely 11.2023 days.
10 In a ola ili y shock scena io a CTA will absolu ely educe i s posi ion in he espec i e ma ke . The sho e he hal - li e o he ola ili y o ecas ,
he mo e esponsi e he posi ion scaling and he la ge he ade size. Consequen ly, a CTA will limi i s own capaci y ce e is pa ibus i he ola-
ili y o ecas is as e .
|
229
UHL
REFERENCES
Bal as, N. (2019). The impac o c owding on al e na i e isk p emia in es ing. Financial Analys s Jou nal, 75, 89–104. h ps:// doi. o g/ 10. 1080/
00151 98X. 2019. 1600955
Ba oso, P., Edelen, R., & Ka ehnke, P. (2022). C owding and ail isk in momen um e u ns. Jou nal o Financial and Quan i a i e Analysis,
57, 1313–1342. h ps:// doi. o g/ 10. 1017/ S0022 10902 1000624
Bohl, M., Pü z, A., & Solewski, C. (2021). Specula ion and he in o ma ional e iciency o commodi y u u es ma ke s. Jou nal o Commodi y
Ma ke s, 23, 100159. h ps:// doi. o g/ 10. 1016/j. jcomm. 2020. 100159
Bollen, N. P. B., Hu chinson, M., & O'B ien, J. (2021). When i pays o ollow he c owd: S a egy con o mi y and c a pe o mance. Jou nal o
Fu u es Ma ke s, 41, 875–894. h ps:// doi. o g/ 10. 1002/ u . 22199
Boos, D., & G ob, L. (2023). T acking specula i e ading. Jou nal o Financial Ma ke s, 64, 100774. h ps:// doi. o g/ 10. 1016/j. inma . 2022. 100774
Bo ga ds, O., & Czudaj, R. (2022). Long- sho specula o sen imen in ag icul u al commodi y ma ke s. In e na ional Jou nal o Finance and
Economics, 28, 3511–3528. h ps:// doi. o g/ 10. 1002/ ij e. 2605
Boyd, N., Ha is, J., & Li, B. (2018). An upda e on specula ion and inancializa ion in commodi y ma ke s. Jou nal o Commodi y Ma ke s, 10,
91–104. h ps:// doi. o g/ 10. 1016/j. jcomm. 2018. 05. 005
B ooks, C., P okopczuk, M., & Wu, Y. (2015). Booms and bus s in commodi y ma ke s: Bubbles o undamen als? Jou nal o Fu u es Ma ke s,
35, 916–938. h ps:// doi. o g/ 10. 1002/ u . 21721
B own, G., Howa d, P., & Lundblad, C. (2022). C owded ades and ail isk. The Re iew o Financial S udies, 35, 3231–3271. h ps:// doi. o g/
10. 1093/ s/ hhab107
Campbell, J., & Thompson, S. (2008). P edic ing excess s ock e u ns ou o sample: Can any hing bea he his o ical a e age? Re iew o
Financial S udies, 21, 1509–1531. h ps:// doi. o g/ 10. 1093/ s/ hhm055
Ca e , C., & Re o edo- Giha, C. (2023). Financializa ion and specula o s isk p emia in commodi y u u es ma ke s. In e na ional Re iew o
Financial Analysis, 88, 1–10. h ps:// doi. o g/ 10. 1016/j. i a. 2023. 102691
CFTC. (2022). Disagg ega ed commi men s o ade s epo – explana o y no es. h ps:// www. c c. go / si es/ de au l / iles/ idc/ g oups/ publi c/@
commi men so ade s/ docum en s/ ile/ disag g ega edco expl ana o yno . pd
Cla e, A., Sea on, J., Smi h, P., & Thomas, S. (2014). T end- ollowing, isk- pa i y and momen um in commodi y u u es. In e na ional Re iew
o Financial Analysis, 31, 1–12. h ps:// doi. o g/ 10. 1016/j. i a. 2013. 10. 001
E b, C. B., & Ha ey, C. R. (2006). The s a egic and ac ical alue o commodi y u u es. Financial Analys s Jou nal, 62, 69–97. h ps:// doi. o g/
10. 2469/ aj. 62. n2. 4084
Fan, J., Fe nandez- Pe ez, A., Fue es, A.- M., & Mi e, J. (2020). Specula i e p essu e. Jou nal o Fu u es Ma ke s, 40, 575–597. h ps:// doi. o g/
10. 1002/ u . 22085
Fan, J., & Zhang, T. (2024). Commodi y isk p emia and isk managemen . Jou nal o Fu u es Ma ke s, 44, 1097–1116. h ps:// doi. o g/ 10. 1002/
u . 22507
Fe nandez- Pe ez, A., Fue es, A.- M., & Mi e, J. (2016). Commodi ies o he long un. In e na ional Re iew o Financial Analysis, 46, 219–226.
h ps:// doi. o g/ 10. 1016/j. i a. 2016. 06. 002
Fol ice, B., & Lange , T. (2015). P o i able momen um ading s a egies o indi idual in es o s. Financial Ma ke s and Po olio Managemen ,
29, 85–113. h ps:// doi. o g/ 10. 1007/ s1140 8- 015- 0246- 4
Fue es, A.- M., Mi e, J., & Fe nandez- Pe ez, A. (2015). Commodi y s a egies in based on momen um, e m s uc u e and idiosync a ic ola-
ili y. Jou nal o Fu u es Ma ke s, 35, 2530–2548. h ps:// doi. o g/ 10. 1002/ u . 21656
Geczy, C., & Samono , M. (2016). Two cen u ies o p ice- e u n momen um. Financial Analys s Jou nal, 72, 32–56. h ps:// doi. o g/ 10. 2469/
aj. 72. n5. 1
Geo gopoulou, A., & Wang, J. (2017). The end is you iend: Time- se ies momen um s a egies ac oss equi y and commodi y ma ke s. Re iew
o Finance, 21, 1557–1592. h ps:// doi. o g/ 10. 1093/ o / w048
Haase, M., Zimme mann, Y. S., & Zimme mann, H. (2016). The impac o specula ion on commodi y u u es ma ke s? a e iew o he indings
o 100 empi ical s udies. Jou nal o Commodi y Ma ke s, 3, 1–15. h ps:// doi. o g/ 10. 1016/j. jcomm. 2016. 07. 006
Haase, M., Zimme mann, Y. S., & Zimme mann, H. (2019). Pe manen and ansi o y p ice shocks in commodi y u u es ma ke s and hei
ela ion o specula ion. Empi ical Economics, 56, 1359–1382. h ps:// doi. o g/ 10. 1007/ s0018 1- 017- 1387- 2
Haase, M. M., Zimme mann, Y. S., & Zimme mann, H. (2017). Commodi y e u ns and hei ola ili y in ela ion o specula ion: A consis en
empi ical app oach. Jou nal o Al e na i e In es men , 20, 76–91. h ps:// doi. o g/ 10. 3905/ jai. 2017. 20.2. 076
Ha ey, C., Ra ay, S., & an Heme , O. (2021). S a egic isk managemen : Designing po olios and managing isk. Wiley.
He, X.- Z., & Li, K. (2015). P o i abili y o ime se ies momen um. Jou nal o Banking and Finance, 53, 140–157. h ps:// doi. o g/ 10. 1016/j. jbank
in. 2014. 12. 017
Hu s , B., Ooi, Y. H., & Pede sen, L. H. (2013). Demys i ying managed u u es. Jou nal o In es men Managemen , 11, 42–58.
Hu s , B., Ooi, Y. H., & Pede sen, L. H. (2017). A cen u y o e idence on end- ollowing in es ing. Jou nal o Po olio Managemen , 44, 2–16.
h ps:// doi. o g/ 10. 3905/ jpm. 2017. 44.1. 015
I win, S., & Sande s, D. (2012). Tes ing he mas e s hypo hesis in commodi y u u es ma ke s. Ene gy Economics, 34, 256–269. h ps:// doi. o g/
10. 1016/j. eneco. 2011. 10. 008
Jegadeesh, N., & Ti man, S. (1993). Re u ns o buying winne s and selling lose s: Implica ions o s ock ma ke e iciency. Jou nal o Finance,
48, 65–91. h ps:// doi. o g/ 10. 2307/ 2328882
230
|
UHL
Kim, A. (2015). Does u u es specula ion des abilize commodi y ma ke s? Jou nal o Fu u es Ma ke s, 35, 696–714. h ps:// doi. o g/ 10. 1002/
u . 21716
Kim, A., Tse, Y., & Wald, J. (2016). Time se ies momen um and ola ili y scaling. Jou nal o Financial Ma ke s, 30, 103–124. h ps:// doi. o g/ 10.
1016/j. inma . 2016. 05. 0036
Koijen, R., Moskowi z, T., & Pede sen, L. (2018). Ca y. Jou nal o Financial Economics, 127, 197–225. h ps:// doi. o g/ 10. 1016/j. j ine co. 2017.
11. 002
Kwon, K. Y., Kang, J., & Yun, J. (2020). Weekly momen um in he commodi y u u es ma ke . Finance Resea ch Le e s, 35, 101306. h ps:// doi.
o g/ 10. 1016/j. l. 2019. 101306
Le ine, A., Ooi, Y., Richa dson, M., & Sasse ille, C. (2018). Commodi ies o he long un. Financial Analys s Jou nal, 74(2), 55–68. h ps:// doi.
o g/ 10. 2469/ aj. 74. n2. 4
Le ine, A., & Pede sen, L. H. (2016). Which end is you iend? Financial Analys s Jou nal, 72, 51–66. h ps:// doi. o g/ 10. 2469/ aj. 72. n3. 3
Lim, B., Zoh en, S., & Robe s, S. (2019). Enhancing ime- se ies momen um s a egies using deep neu al ne wo ks. The Jou nal o Financial
Da a Science, 1(4), 19–38. h ps:// doi. o g/ 10. 3905/ j ds. 2019.1. 015
Lu zenbe ge , F. (2014). The p edic abili y o agg ega e e u ns on commodi y u u es. Re iew o Financial Economics, 23, 120–139. h ps:// doi.
o g/ 10. 1016/j. e. 2014. 02. 001
Ma kwa , T., Quis , J., & Zome dijk, C. (2020). Value in es ing o commodi ies. Jou nal o Al e na i e In es men , 23, 127–140. h ps:// doi. o g/
10. 3905/ jai. 2020.1. 107
Mas e s, W. M., & Whi e, A. (2008). The acciden al hun b o he s: How ins i u ional in es o s a e d i ing up ood and ene gy p ices. Technical
documen .
Mi e, J. (2016). Long- sho commodi y in es ing: a e iew o he li e a u e. Jou nal o Commodi y Ma ke s, 1, 3–13. h ps:// doi. o g/ 10. 1016/j.
jcomm. 2016. 01. 001
Moskowi z, T. J., Ooi, Y. H., & Pede sen, L. H. (2012). Time se ies momen um. Jou nal o Financial Economics, 104, 228–250. h ps:// doi. o g/
10. 1016/j. j ine co. 2011. 11. 003
Poli is, D., & Romano, J. (1994). The s a iona y boo s ap. Jou nal o he Ame ican S a is ical Associa ion, 89, 1303–1313. h ps:// doi. o g/ 10.
1080/ 01621 459. 1994. 10476870
RiskMe ics. (1996). RiskMe ics – Technical documen (4 h ed.). JP Mo gan/Reu e s.
Shen, Q., Szakma y, A., & Sha ma, S. (2007). An examina ion o momen um s a egies in commodi y u u es ma ke s. Jou nal o Fu u es
Ma ke s, 27, 227–256. h ps:// doi. o g/ 10. 1002/ u . 20252
Skiadopoulos, G. (2012). In es ing in commodi ies: Popula belie s and misconcep ions. Jou nal o Asse Managemen , 13, 77–83. h ps:// doi.
o g/ 10. 1057/ jam. 2011. 35
Smales, L. (2016). T ading beha io in s&p 500 index u u es. Re iew o Financial Economics, 28, 46–55. h ps:// doi. o g/ 10. 1016/j. e. 2015. 11.
001
S ad mülle , I., Aue , B., & Schuhmache , F. (2024). Co e- sa elli e in es ing wi h commodi y u u es momen um. Jou nal o Asse Managemen ,
25(3), 261–287. h ps:// doi. o g/ 10. 1057/ s4126 0- 024- 00352 - 5
Vinzelbe g, A., & Aue , B. (2014). Do c ude oil u u es s ill uel po olio pe o mance? Re iew o Financial Economics, 39, 402–423. h ps://
doi. o g/ 10. 1002/ e. 1127
Wimme , T., Geye - Klingebe g, J., Hü e , M., Schmid, F., & Ra hgebe , A. (2021). The impac o specula ion on commodi y p ices: A me a-
g ange analysis. Jou nal o Commodi y Ma ke s, 22, 100148. h ps:// doi. o g/ 10. 1016/j. jcomm. 2020. 100148
How o ci e his a icle: Uhl, B. (2025). Specula o s and ime se ies momen um in commodi y u u es ma ke s.
Re iew o Financial Economics, 43, 213–230. h ps://doi.o g/10.1002/ e.1228