Sie sind auf Seite 1von 6

US

Crude Reserve Modeling


Theoretically,petroleuminventorylevelsareameasureofthebalanceorimbalancebetween petroleumsupply(domesticproductionandimports)anddemand,andthusprovideagoodmarket barometerforcrudeoilprice(e.g.spot,futures,etc.)changes. TheUSEnergyInformationAdministration(EIA)PetroleumStatusReportprovidesinformationonthe weeklychangeinpetroleuminventoriesintheU.S.,whetherproducedlocallyorabroad.Thisweekly reportgivesanoverviewofthelevelofcrudereservesheldandproducedbytheU.S.bothdomestically andabroad.Itisanindicatorofcurrentoilprices.

Why should we care?


Thelevelofinventorieshelpsinvestorstoestimatethepricesforpetroleumproducts(e.g.gasoline, heatingoil,diesel,etc.).Justlikeanyothergoodsandservices,pricesforpetroleumproductsare determinedbysupplyanddemand. Furthermore,crudeoilisanimportantcommodityintheglobalmarketandplaysacrucialpartofthe economy,sofluctuationsinitspricehaveadirectinfluenceonconsumerprices.Assuch,iflowreserves arereported,crudeoilpricesaremostlikelytoincreaseandthereforedriveupconsumerpricesaswell. Inthispaper,wewillmodelthedynamicsinthecrudeoilstockpile(includingSPR)levelovertime,and projectaforecastforthenext12months.

Background
Thegeneraldemandforpetroleumproductsishighlyseasonalandisgreatestduringthewintermonths, whencountriesintheNorthernHemisphereincreasetheiruseofdistillatedheatingoilsandresidual fuels.Supplyofcrudeoil,includingbothproductionandnetimports,alsoshowsasimilarseasonal variationbutwithasmallermagnitude. Duringthesummermonths,supplyexceedsdemandandpetroleuminventoriesnormallybuild;whereas duringthewinter,demandexceedssupplyandinventoriesaredrawndown.Asaresult,inventoriesalso demonstrateseasonality.

Analysis
Foroursampledata,wellusetheEIAdatafromtheweeklypetroleumstatusreportbetweenJanuary 1991(PostGulfWar)andAugust17th,2012.TheEIAsourcekey(i.e.designation)forthistimeseriesis WCRSTUS1.

USCrudeReserveModeling

SpiderFinancialCorp,2012

Millions

1150 1100 1050 1000 950 900 850 800

USOilReserve (inc.SPR)

Jan91

Jan92

Jan93

Jan94

Jan95

Jan96

Jan97

Jan98

Jan99

Jan00

Jan01

Jan02

Jan03

Jan04

Jan05

Jan06

Jan07

Jan08

Jan09

Jan10

Jan11

Jan12

Atfirstglance,thedatadoesnotexhibitthe12monthseasonalitythatwewereanticipating,solets lookcloser. Now,tobetterdistributethevalues,wellcomputethelogarithmictransformationofthetimeseries anduseitprimarilyinouranalysis. Usingsummarystatistics,thetimeseriesisseriallycorrelatedanditsprobabilitydistributionexhibitsfat tails.

Furthermore,thecorrelogramanalysis(i.e.ACFandPACFplots)uncoveranintegrationissuewithalag orderofone:

USCrudeReserveModeling

SpiderFinancialCorp,2012

Also,lookclosertothepartialautocorrelationnearlagorder5255.Theleapyearphenomenonmakes itdifficulttopinpointaseasonlengthinnumberofweeks. Toovercomethisproblem,wellusethemonthlystocklevelandmodelitinstead.TheEIApublishesthe monthlycrudereservestockpilelevel,buttheselevelsarepublishedonceamonthandaregenerallyfor thepreviousthreemonthperiod. Formonthlytimeseries,wellinterpolatethereservelevelonthelastdayofeachmonth(similartothe EIAmonthlyreport)usingtheweeklytimeseries. Letsrerunthecorrelogramanalysisonthelogtransformedmonthlytimeseries:

USCrudeReserveModeling

SpiderFinancialCorp,2012

ThecorrelogramplotsuggestsaseasonalARIMA(SARIMA)modelwithaseasonlengthof12months. WellproposeaspecialcaseofSARIMAtheAirlineModel.TheAirlineModelcannotonlycapturethe seasonality,butitrequiresonlythree(3)parameterstoestimate,soitistypicallyagoodstartingpoint. UsingtheNumXLAirlinewizards,wespecifieda12monthAirlinemodelandcalibrateditsparameter valuesusingthelogmonthlylevels.

Thecalibratedparametersvaluesyieldastablemodelandsatisfytheunderlyingmodelassumptions.In short,theAirlinemodelisareasonablecandidatemodelforthemonthlyloglevelprocess. Interpretation:

(1 L)(1 L12 ) xt zt (1 L)(1 L12 )at

Themodelissayingthattheyearoveryearofthemonthlychange( zt )isbasicallyaspecialmoving averageprocessoforder13. Letsgoaheadnowandusethemodelforourforecast:

USCrudeReserveModeling

SpiderFinancialCorp,2012

TheAirlinemodelsfunctionscomputethemeanandconfidenceintervalofthemonthlyloglevel.To movetheforecastbackintobarrelsunit,wetooktheexponentialtransformandadjustedthemean forecastasfollows: Where:

E[YT n ] e

xT n

2 T n

T2 n istheforecasterrorattimeT+n
1250 1200 1150 1100 1050 1000 950 900 850 800

Millions

Inthegraphabove,thedottedlinerepresentstheAirlinemeanforecastandtheshadedareaisthe95% confidenceinterval. EIA Weekly Reports and forecast Toforecastthecrudeoilreservelevelforanyarbitraryfuturedate(otherthantheendofthemonth), wesimplyinterpolatethevalueofthisdateusingthemonthlyforecast.Thesamecanbesaidforthe confidenceintervallimits. ForthecurrentmonthwhereEIAissuedoneormoreofitsweeklyreports,weusethelatestreport valueandtheforecastedendofmonthvaluetointerpolatethevalueforanintermediatedate.

Conclusion
Inthispaper,weanalyzedtheweeklytimeseriesofcrudeoilreservelevelsanddemonstrateditskey statisticalproperties:integrationandseasonality.Next,usingtheweeklytimeseries,weinterpolated themonthlytimeseriesandconstructedanAirline(12)model.Thecalibratedmodelsvaluessatisfiedall assumptions.

USCrudeReserveModeling

SpiderFinancialCorp,2012

UsingAirline(12)weprojectedan18monthforecastforthecrudereservelevelandconstructeda95% confidenceinterval.

USCrudeReserveModeling

SpiderFinancialCorp,2012

Das könnte Ihnen auch gefallen