275
1 INTRODUCTION
Theaimofthispaperistoinvestigatetheimpactof
bunker prices in the compliance decisions related to
theMARPOLAnnexVIRegulations,andspecifically
of Regulation 14, which deals with the sulfur
emissions (SOx). The set limits apply to all vessels
that fly the flag of
a country, which has ratified the
convention.Additionally,itappliestovesselsthatfly
the flag of a non‐signatory state while operating
within waters under the jurisdiction of a country
adhering to MARPOL’s Annex VI Reg. 14. On
October 10
th
2008 more stringent amendments were
adoptedtoAnnexVI,whichenteredintoforceonJuly
1
st
2010. Regulation 14 specifies the sulfur limit in
marine fuel for global maritime trade. The highest
sulfur content allowed in ship fuel will reduce
globallyasof1January2012from4.5%to3.5%andas
of 1 January 2020 to 0.5%, depending on the
availabilityoflowsulfurfuel,
asthiswillbediscussed
andreviewedbytheIMOin2018.Evenstrictervalues
applyfortheEmissionControlAreas(ECA).TheECA
includes Baltic Sea, North Sea and the English
Channel,aswellastheoceaniccoastlineoftheUnited
StatesandCanada.WithintheECA’sthesulfurlimit
was1.5%untilJuly2010.AsofJuly2010theallowed
sulfurcontentwasreducedto1.0%.In2015thelimit
willfurtherbedecreasedto0.1%.
TheissueofcompliancewithRegulation14iswell
known in the related business and academic
literature.ThestudyofEMSA(2010)and
thereportof
Miola et al. (2010), as well as the studies on green
housegases(GHG)andairpollutionofBuhaugetal.
(2009)amongothers,havepavedthewayforfurther
research,regulatoryactionandvividdebates.Tothis
discussion, the authors have also contributed with
advanced methodologies estimating
the cost of
operatinginmixedECAandnon‐ECAareas(Schinas,
et. al. 2012a) as well as with policy support
documents,suchasinSchinasetal.(2012b).Inallthe
relatedstudies,thefocusiseithermicro‐economic,i.e.
a focus on the impact from the operation and
the
related expenses of burning fuels of diverse quality,
namely either High‐Sulfur (HS) or Low‐Sulfur (LS)
heavy fuel oil (HFO) or intermediated fuel oil (IFO)
The Cost of SOx Limits to Marine Operators; Results
from Exploring Marine Fuel Prices
O.Schinas
HamburgSchoolofBusinessAdministration,Hamburg,Germany
Ch.Stefanakos
TechnologicalandEducationalInstituteofAthens,Athens,Greece
ABSTRACT:Marineoperatorsareconfrontedwiththenewairemissionsregulationsthatdeterminethelimits
ofsulfurcontentinmarinefuels.Thelow‐sulfur(LS)marinefuelshaveahigherprice,andtheirfluctuationis
almostsimilartothefluctuationofhigh‐
sulfur(HS)fuels.ThepricedifferencebetweenHSandLSmightalso
determinethedecisionofoperatorsforalternativetechnicalmeans,suchasscrubbers,inordertocomplywith
thenewlimits.ThispaperaimstoprovideathoroughstatisticalanalysisofthecurrentlyavailableLSandHS
marinefuelstime
series,aswellastopresenttheanalysisofthedifferentialoftheHSandLSfuelprices.The
paperconcludeswithsuggestionsforfurtherresearch.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 7
Number 2
June 2013
DOI:10.12716/1001.07.02.15
276
marine diesel oil (MDO), or macro‐economic, i.e.
approaches dealing with the global fleet and total
effectofmarinefuelconsumptiontotheenvironment.
However, there is no study, to the best of the
knowledgeoftheauthors,examiningthetimeseries
ofthepricesoftherelatedfuels.Alizadeh
et.al.(2004)
examinedthecorrelationofRotterdam,Houstonand
Singapore bunker prices time‐series vis‐à‐vis the
prices of future contracts traded at the New York
MercantileExchange(NYMEX)andtheInternational
PetroleumExchange(IPE)inLondon;furtherworkon
this subject is not reported. Should the statistical
characteristicsofthetimeseriesbeknownthenbetter
forecasting and explanatory models can be derived.
This paper aims to offer some basic analysis of the
time series of HS‐HFO and LS‐HFO in Rotterdam,
discusstheresultsoftheanalysisandfinallydiscuss
some possible steps for further research.
It is out of
the scope of this paper to analyze the political and
economic factors that determine the prices, but
scientificallyanalyzetherelatedtime‐series.
The following section presents the value of
accurateoratleastreliableandtrustworthystatistical
attributesoffueloilpricetimeseriesfor
thesupport
oftechnical,operationalandfinancialdecisionsatthe
micro‐level. In the next session, the analyses of the
HS‐andLS‐HFOpricesinRotterdamaswellasofthe
derivedtimeseriesofthedifferencebetweenHS‐and
LS‐HFO are presented. In the last section the
conclusions
are summarized and suggestions for
furtherresearcharepresented.
2 THEIMPACTOFFUELPRICES
Although different ship types and propulsion plant
configurations suggest different cost structures, it is
commonknowledgethatfuelcostsdeterminealarge
percentage of the overall cost, and therefore the
financial performance of this asset.
The compliance
withreducedsulfurlimitscanbeachievedviavarious
instruments, however two major and most feasible
instruments are practically available to operators: a
dualfuelsystemandanexhaustgascleaningsystem
(EGCS),i.e.ascrubberorsimilartechnology.Thedual
fuelsystemimpliesa switchfromregularfuel
toLS
fueloilinECAs.Withtheinstallationofscrubbersitis
notnecessarytooperatewithmorethanonefuel:HS
fueloilscanbeused,theexhaustsarethencleanedby
thescrubbersinordertoreducetheemissionofsulfur
into the air. Other sulfur abatement
instruments are
also in discussion, but do not seem to be
implementable within the next couple of years, as
there are many unsolved questions regarding their
usage.
Oneofthosealternativesisthevesselpropulsion
via liquefied natural gas (LNG). Implementing this
technology makes costly retrofitting of the vessel
necessary.A
furtheraspect,whichhastobetakeninto
account,istheavailabilityofLNGbunkeringfacilities
inports.Recentpolicyinitiatives,includeaproposed
European Directive, where the Commission
introduces an obligation for all the major European
seaports to be equipped by 2020 with publicly
accessible LNG refueling points for
both maritime
andinlandwaterwaytransport(EC,2013).Addingto
thepolicycontroversyoverairemissionissues,ESPO
doubts, whether imposing LNG refueling
infrastructure in all the major European would be
appropriate,sincetheremaynotbeamarketforitin
allofthoseports,whereastherecouldbea
marketin
other,non‐coreports,aswellasalternativesolutions
to the development of LNG are and will become
increasinglyavailableinthenearfuture(ESPO,2013).
Generallyspeaking,intheshorttermtheavailability
of LNG bunkering facilities globally cannot be
guaranteed and thus this option is intentionally
neglectedinthisanalysis.
Thusthissectionwillcover
two alternatives that seem to be available in short
termin detail: the dualfuel system and the exhaust
gasscrubbertechnology.
Assuming the following data for a typical
containervessel:
1 OperatingSpeeds
ECOSpeed:16.5kn
DesignSpeed:18.5kn
MaximumSpeed:21kn
2 FuelTypesconsumed
IFO380
MDO
3 OperationalData
Steaming@ECOSpeed16.5kn(days/year):50
Steaming @DesignSpeed 18.5 kn(days/year):
140
Steaming@MaximumSpeed21kn(days/year):
20
Port/IdleTime(days/year): 150
Timenotusedforoperation(days/year):5
Estimated Annual Operation in ECAs
(days/year):80
4 MainEngine
MCR(kW):16,000
TypeofFuel:IFO380
FuelConsumption(t/yr):7,939
5 AuxiliaryEngine
No.ofMachineryofthisType:4
MCR(kW):1,600
TotalkWAuxiliaryEngine:6,400
SpecificFuelOilConsumption(g/kWh):180
TypeofFuel:MDO
FuelConsumption(t/yr):1,459
6 Boiler
No.ofMachineryofthisType:1
FuelOilConsumption(t/day):25
TypeofFuel:MDO
FuelConsumption(t/yr):1,664
Obviously the consumption data can be
thoroughly evaluated, and the calculations above
yieldtheproductofthedailyconsumptiontimesthe
timeofoperationatvarious‘expected’levelsofload.
Giventhat the consumptiondataaboveare accurate
enough to support further estimation of the cost,
operators face the following dilemma: they should
either install a scrubber (generally a EGCS) or to
install/useadualfuelsystem.Thecriticalparameters
that determine the decision are the price of HSHFO
and LSHFO, the time (days/year) of expected
operationinanECAandtherequiredinvestmentfor
EGCS. In
few words, operators have the option to
installanEGCSandconsumethecheaperHSHFOin
all operating cases or to use HSHFO and LSHFO
whereverpermittedorrequired.Thedifferenceofthe
discountedcoststreamswilldetermine,whichoption
277
is from a financial point of view more appropriate.
The method of net present value (NPV) could be
considered, although with some cautiousness as
manyassumptionscanjeopardizethevalidityof the
outcome. The NPV analysis pinpoints the economic
lifeoftheasset(theageoftheship),thebunker
prices
andthediscountrateasthecriticalparametersofthe
decision.Thisdoesnotcontradictwiththepraxis,and
impliesthattheonlyparameternotdeterminedbythe
ownerorthefleetstatusisthepriceofbunkers,which
drawsalsotheattentionasthemainriskparameter.
Having
saidthat,itisobviousthatitisofcritical
importance to consider rational scenarios of future
fuel price values, as the amortization of the
investmentwillbeconcludedinsomefutureyears.In
ordertodealwiththeseissues,atypicalspreadsheet
modelerwouldconsiderscenariosthatarestructured
as
follows:
CurrentFuelPriceUSD/t:610
PriceDifferencetoSulfurContentof1%USD/t:40
PriceDifferencetoSulfurContent of 0.5% USD/t:
140
PriceDifferencetoSulfurContent of 0.1% USD/t:
225
Considering the above data and an investment
horizonof25yearsfortheship,discountrates(forthe
DCF Analysis) of 1.0% (low), 5.0% (medium) and
10.0%(high), as wellas the assumptions of tables X
and Y below, the accumulated
consumption is
depictedinthefollowinggraphs:
Table1.CurrentPricesandPriceDifferences(Scenarios)
_______________________________________________
USD/tBase 50% 75%
_______________________________________________
CurrentFuelPrice610 610 610
PriceDifference1%406070
PriceDifference0.5% 140 210 245
PriceDifference0.1% 225 340 390
_______________________________________________
Table2.Growthrates(Scenarios)
_______________________________________________
ScenarioLow MediumHigh
_______________________________________________
EstimatedFuelPrice 1%5%10%
IncreasesperYearin%
EstimatedFuelPrice 1%10%15%
Increaseasof2015(ontop
oftheaboveincrease)
EstimatedFuelPrice 1%10%20%
Increaseafter2020/25(on
topoftheaboveincrease)
_______________________________________________
Figure1.Lifecyclecostofbunkersas perthescenarios
ThebunkeringhubofRotterdamistakenasabase
case,aspricesofallmajorbunkeringhubsfollowthe
samefluctuationpattern,asdepictedinFigure2(see
alsosection4aboutthisissue.)
Figure2.Pricesinvariousbunkeringhubs
Obviously the difference of the prices among HS
and LS fuels is determining at large the financial
exposure for bunkers. Having a closer look at the
results of this given and indicative example the
difference of 50% from the base scenario implies an
annualgrowthrateof1.9%vis‐à‐vis
1.5%forthelow
costscenariothatsuggestanincreaseofcloseto5.8%
(ca.600,000USD)oftheannualbunkercosts.Forthe
high‐scenario, the annual growth rate is 11.9% and
the annual increase close to 7% (close to 3.2millions
USD).Consideringthedataofthe75%scenario,then
thefiguresdramaticallychangeforthelowandhigh
scenarios,asanannualgrowthof2.0%and12.1%of
the expenses is envisaged, an increase of minimum
8.6%andmaximumof9.9%isexpectedintheaverage
annualbunkercost,implyinghigherexpensescloseto
minimum900kUSDandmaximum4.7mUSD.
In conclusion, operators can neither ignore the
evolutionoftheHSandLSprices,nordisregardthe
influence of the difference of the HS and LS prices
anditisimperativeforthemtodrafttheappropriate
scenarios,i.e.tohaveabetterinsightofthedynamics
of the statistical
attributes, in order to support their
decisions.
3 PROBABILITYANALYSIS
The data source used in the present study is the
weekly time series of the HS‐, LS‐HFO (380cSt) in
Rotterdam, as published by Shipping Intelligence
NetworkofClarksons.AlthoughdatafortheHS‐HFO
278
areavailablesince 1990, wedecided to useonly the
subsetofdatawherebothvalues(HS ‐ andLS‐HFO)
arepresent.Thus,theanalyzeddatacoversthetime
period from 2007 till 2012. The total amount of the
analyzed data (256 values) is not statistically
significant.However,itis
theonlyreliabledatasetup
tonow.
Along with the analysis of basic statistics,
probability analysis is also necessary. Probability
analysisisaversatiletoolfortheanalysisofhistorical
data. For each dry index, the empirical histogram is
calculatedasfollows(Spanos2003).
First,aparticularpartitionisdefined
oftheform
12 1
,,,,,
iI
(1)
in order to appropriately segment the range of
possiblevaluesoftheindex.Then,thetableofrelative
frequenciesofoccurrence(histogram)iscalculatedas
,1,2,,
i
i
k
iI
N
, (2)
where
1
# ' : , 1, 2,...,
inini
kofXsX n N
(3)
and N is the total number of observations
(measurements).
The selection of the appropriate partition is of
paramount importance for the probability analysis
andusuallyisatantalizingtask.Itrequiresmuchof
experimentation before concluding with the right
partitionthatrevealstherightformoftheunderlying
probability
mass. In the present analysis, the
partitionsgiveninTable3haveused.
Table3.Partitionsusedinprobabilityanalysis.
_______________________________________________
DataPartition
_______________________________________________
HS‐HFO[0160:20:820]
LS‐HFO[0160:20:820]
Difference[‐10:10:100]
_______________________________________________
3.1 HighSulfurHeavyFuelOilPricesinRotterdam
InTable4,thebasicstatisticsforthetimeseriesofHS‐
HFOaregiven.
Table4.BasicStatisticsfortheHS‐HFO
_______________________________________________
YearCount Mean MIN MAX St.Dev. Skewn. Kurt.
_______________________________________________
200722 412.95 337.50 495.00 50.15 0.06 1.55
200852 474.33 171.50 707.00 145.89 ‐ 0.61 2.54
200952 352.73 173.50 466.00 84.75 ‐0.40 1.71
201053 451.06 405.00 494.00 22.99 0.08 2.12
201152 619.51 511.50 669.50 39.33 ‐1.55 4.71
201225 671.06 553.00 720.00 48.16 ‐1.24 3.33
all 256 488.24
171.50 720.00 133.74 ‐ 0.22 2.40
_______________________________________________
In Figures 3‐9, the histograms of HS‐HFO are
presented first based on the whole amount of data
availableandthenoneachspecificyear.
InFigure3,forexample,thefollowingareaswith
concentrated probability mass can be distinguished.
Thereisa30%ofthevaluesfalling
intherange420‐
480$/ton,a22%in600‐660,anda9%in200‐300.
So,theseareasofHS‐HFOcanbeconsideredthat
theyhavegreaterprobabilitytooccurinthefuture.
Workingsimilarlywiththeannualresults,areasof
greaterprobabilitycanbedefined.
Figure3. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforallyears.
Figure4. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2007.
Figure5. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2008.
279
Figure6. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2009.
Figure7. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2010.
Figure8. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2011.
Figure9. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of HS‐
HFOforyear2012.
3.2 LowSulfurHeavyFuelOilPricesinRotterdam
InTable5,thebasicstatisticsforthetimeseriesofLS‐
HFOaregiven.
Table5:BasicStatisticsfortheLS‐HFO
_______________________________________________
YearCount Mean MIN MAX St.Dev. Skewn. Kurt.
_______________________________________________
200722 436.11 364.00 517.00 50.98 0.06 1.48
200852 515.55 212.50 760.00 141.54 ‐ 0.48 2.59
200952 368.36 197.50 484.50 84.86 ‐0.35 1.69
201053 471.63 415.00 510.50 20.33 ‐0.16 2.96
201152 655.38 516.50 722.00 50.36 ‐1.50 4.68
201225 712.94 593.50 781.50 53.10 ‐0.70 2.60
all 256517.41
197.50 781.50 140.30 ‐ 0.07 2.26
_______________________________________________
In Figures 10‐16, the histograms of LS‐HFO are
presented first based on the whole amount of data
availableandthenoneachspecificyear.
Working as in the previous section, areas of
greaterprobabilityaredefined.
InFigure10,forexample,thefollowingareaswith
concentrated probability
mass can be distinguished.
Thereisa 35.5%ofthevaluesfallingintherange440‐
520$/ton,a15%in640‐680,anda9%in240‐320.So,
these areas of LS ‐HFO can be considered that they
havegreaterprobabilitytooccurinthefuture.
By comparing the
probabilities of same areas in
HS‐andLS ‐HFOinterestingresultscanbeextracted:
for example, the area 420‐480 $/tn has a probability
25%inLS (30%inHS)andsimilarlytheresultyield
9%in600‐660(22%inHS)and8.5%in200‐300(9%in
HS).
Figure10. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
forallyears.
Figure11. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2007.
280
Figure12. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2008.
Figure13. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2009.
Figure14. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2010.
Figure15. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2011.
Figure16. Relative frequencies of occurrence (%) and
empiricalcumulativedistributionfunction(ecdf)ofLS‐HFO
foryear2012.
3.3 TheTimeSeriesoftheDifference
Thebasicstatisticsofthetimeseriesofthedifference
issummarizedinTable6.
Table6:BasicStatisticsfortheDifference(LS‐HS)
_______________________________________________
YearCount Mean Min MaxSt.Dev. Skewn. Kurt.
_______________________________________________
200722 23.16 16.00 50.00 8.121.82 6.35
200852 41.22 16.00 87.50 13.59 1.04 4.64
200952 15.63 ‐7.00 32.00 6.56‐0.75 5.91
201053 20.58 10.00 42.50 9.720.99 2.85
201152 35.87 ‐1.00 73.00 18.62 0.34 2.13
201225 41.88 8.00 71.50 18.29 ‐0.49 2.30
all 256 29.17
‐7.00 87.50 16.78 0.81 3.12
_______________________________________________
The results yield that it is very difficult and
scientificallynot justified to extract conclusions over
the mean value, which fluctuates substantially year
over year. The same result is also extracted for the
minimumandmaximumvalues.Theoverallanalysis,
i.e. the synthesis of all annual data, suggests that
a
mean of 30USD should be expected, yet with a
standarddeviationofalmost17USD,thussettingthe
range of expected fluctuation close to 35USD.
Moreover, it seems that the distribution is not
symmetric, ‘heavy’ tails should be expected as the
kurtosisis3.12,and the positive skewindicatesthat
the
tail on the right side is longer than the left side
andthebulkofthevalueslietotheleftofthemean.
Figure17. Relative frequencies of occurrence (%) and
empirical cumulative distribution function (ecdf) of the
DiffrenceLS‐HSforallyears.
Furthermore, in Figure 17, the histogram of the
Difference LS‐HS is presented based on the whole
amount of data available. The histograms based on
281
eachspecificyearareomittedduetospacelimitation.
Theyareavailableuponrequest.
By considering the overall probability analysis, it
isobviousthat the distributions arenotGaussianor
can even be easily approximated. Some of them
exhibit more than one crests dictating for a more a
sophisticatedprobability
modeling.
4 CONCLUSIONSANDFURTHERRESEARCH
Onthebasisoftheresultedstatisticsandgraphs,itis
palpable that operators cannot effortlessly draft
scenarios that are based or linked to the statistical
attributes of these time series. Although, forecasting
orscenariobuildingonthebasisofthesetimeseries
would
be desired and meaningful, it seems that
conventionalstatisticsdoprovidesolidfoundationfor
furtherunivariateforecasting.
Nonethelesstherelativelyfewobservationsofthe
LS time series imply that more advanced models,
suchasautoregressiveonesmightnotbe suitableas
well. In the literature there are many works dealing
with
theissueofsmallsamplepropertiesofestimates
of the parameters of autoregressive models. Taking
intoaccounttheformula:
y
t=α+βyt−1+σεt,t=1,2,...,T,εt∼iid(0,1). (4)
The majority of these works concentrates on
deriving either exact and/or approximate small
sampleresultsforthedistributionoftheestimatedα
T
and β
T of the Ordinary Least Squares (OLS)
estimatorsofαandβ,inthefirst‐orderautoregressive
(AR(1))model.Theestimationoftheseparametersis
tothedirectinterestofaforecaster,asbiasishiddenif
thedistributionsarenotaspertheorydescribes.Such
questions have attracted the interest of
many
researchers, as they melt down to the estimation of
the degrees of freedom ofσand [Var], and various
methodologieshavebeendeveloped,thatdemandnot
onlyempiricalanalysisbutalsotheoreticaltreatment.
In conclusion, researchers should develop new
approaches that could be useful to operators and
businesspeople.
Taking
into account the results of the analysis,
further research should be directed towards the
relationship (regression analysis) of the prices in
Rotterdamwiththatoftheotherhubs.Moreover,the
analysis of the basic statistics of all these bunkering
hubsshouldbe regularly updated andifpossible to
belinkedwith
timeseriesofwiderinterest,suchasof
oilpricesandglobalindustrialactivity.
REFERENCES
[1] Alizadeh, A, Kavoussanos, M.G., Menachof, D. (2004)
Hedging against bunker price fluctuations using petroleum
futures contracts; constant vs time varying hedge ratios,
AppliedEconomics,Vol.36,No12,pp.1137‐1353
[2] Buhaug,Ø.,CorbettJ.J,Endresen,O.,Eyring,V.,FaberJ.,
Hanayama,S.,Lee,D.,Lindstad,H.,Mjelde,A.,Palsson,
C.,Wanquing,W.,Winebrake,J.J.,Yoshida,K.,(2009)
Second IMO Greenhouse Gas Study. International
MaritimeOrganization,London.
[3] European Commission, Proposed Directive, COM(2013)
18/2
[4] European Maritime Safety Agency (2010) The 0.1%
Sulphur in fuel requirement as from 1 January 2015 in
SECAs; An assessment of available impact studies and
alternativemeansofcompliance.
[5] ESPOConcernedAboutObligationtoEquipCorePortswith
LNG Facilities, Thursday, 24 January 2013 11:41
(www.espo.be)
[6] Miola, A. Ciuffo, B. Giovine, E., Marra, M. (2010)
Regulatingairemissions from ships, Thestateofthe art on
methodologies, technologies and policy options, Joint
Research Centre, Institute for Environment and
Sustainability,ISBN978‐92‐79‐17733‐0.
[7] Schinas, O. Stefanakos, Ch. (2012a) Cost assessment of
environmental regulation and options for marine
operators, Transportation Research Part C: Emerging
Technologies, Vol. 25, pp.81‐99, doi:
10.1016/j.trc.2012.05.002
[8] Schinas, O. Bani, J. (2012b) The Impact of a Possible
Extension at EU Level of SECAs to the Entire European
Coastline,NotetotheEuropeanParliament,PE474.549
[9] Spanos, A. 2003. Probability Theory and Statistical
Inference. Econometric Modeling with Observational
Data.Cambridge:CambridgeUniversityPress.
