predicting crude oil price trends using artificial … · predicting crude oil price trends using...

Predicting Crude Oil Price Trends Predicting Crude Oil Price Trends Using Artificial Neural Network Using Artificial Neural Network Modeling ApproachModeling Approach

Faisal ALAdwani Faisal ALAdwani LSU Craft & Hawkins Department of Petroleum LSU Craft & Hawkins Department of Petroleum EngineeringEngineeringWumi Iledare (Presenter)Wumi Iledare (Presenter)LSU Center for Energy StudiesLSU Center for Energy Studies

2005 USAEE 25TH North American Conference 2005 USAEE 25TH North American Conference Denver, CODenver, CO

22

OutlineOutline

IntroductionIntroductionMethodologyMethodologyData and Sources of DataData and Sources of DataCalculations and ResultsCalculations and ResultsConclusionConclusion


33

IntroductionIntroduction

WHAT DICTATES CURRENT TRENDS IN WORLD CRUDE OIL PRICES?– 1) TECHNICAL (Exploration & Production Activity)?

Plenty of activity worldwide therefore no effect?– 2) MARKET FUNDAMENTALS (Supply & Demand)?

No shortages foreseen in the near future , so little effect?– 3) SEASONAL ( Warm & Cold Climates)?

Warm weather for the near future but still little effect?– 4) PYSCHOLOGICAL (Rumors & False Reports)?

Plenty of rumors and false reporting and very large effect?– Speculative buying– Commercial traders raising prices


44


Accurate estimation of crude oil price trends Accurate estimation of crude oil price trends can optimize production strategies.can optimize production strategies.Accurate estimation of crude price trends is Accurate estimation of crude price trends is likely to lead to some stability in the world likely to lead to some stability in the world oil market.oil market.Prices are certainly affected by supply and Prices are certainly affected by supply and demand in addition to other factors.demand in addition to other factors.


55


Several authors have used ANN to model Several authors have used ANN to model complex noncomplex non--linear system, linear system, MoshiriMoshiri and and FooutanFooutan (2004).(2004).AgbonAgbon and and AraqueAraque (2003) used time series (2003) used time series analysis and Chaosanalysis and Chaos--Theory to add some Theory to add some complexity to ANN modeling framework.complexity to ANN modeling framework.The main objective of this study is to model The main objective of this study is to model price trends using ANN with supply and price trends using ANN with supply and demand levels as input variables. demand levels as input variables.


66

MethodologyMethodology

Generalized Regression Neural Network Generalized Regression Neural Network (GRNN) model is used in this study.(GRNN) model is used in this study.GRNN is a universal approximation for a GRNN is a universal approximation for a smooth function.smooth function.GRNN has two layers:GRNN has two layers:–– Radial basis transfer function.Radial basis transfer function.–– PurelinPurelin transfer functiontransfer function


77


88


Radial basis function has a maximum of 1 when its Radial basis function has a maximum of 1 when its input is 0. As the distance between w and p input is 0. As the distance between w and p decreases, the output increases. Thus, a radial decreases, the output increases. Thus, a radial basis neuron acts as a detector that produces 1 basis neuron acts as a detector that produces 1 whenever the input p is identical to its weight whenever the input p is identical to its weight vector w.vector w.The bias b allows the sensitivity of the The bias b allows the sensitivity of the radbasradbasneuron to be adjusted. For example, if a neuron neuron to be adjusted. For example, if a neuron had a bias of 0.1 it would output 0.5 for any input had a bias of 0.1 it would output 0.5 for any input vector p at vector distance of 8.326 (0.8326/b) vector p at vector distance of 8.326 (0.8326/b) from its weight vector w. from its weight vector w.


99


100

200

300

400

500

600

700

800

900

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

Radial Basis Function Spread Constant

Sum

Of S

quar

es E

rror

0.8

0.81

0.82

0.83

0.84

0.85

0.86

0.87

0.88

Correlation C

oeffecient

Spread R2 SSE0.93 0.8666 516.6035


1010


GRNN predictive power depends on input GRNN predictive power depends on input variables and how these variables can be variables and how these variables can be grouped / normalized.grouped / normalized.Several models were considered in this study.Several models were considered in this study.The best model was chosen based on the The best model was chosen based on the following statistical indicators following statistical indicators –– correlation coefficients (Rcorrelation coefficients (R22))–– Sum of Squares Error (SSE)Sum of Squares Error (SSE)–– Standard Deviation (St. Dev.)Standard Deviation (St. Dev.)


1111

Data and Sources of DataData and Sources of DataInput VariablesInput Variables–– OPEC ProductionOPEC Production–– World DemandWorld Demand–– US Domestic SupplyUS Domestic Supply–– OPEC ReservesOPEC Reserves–– OECD DemandOECD Demand–– OECD ConsumptionOECD Consumption–– OECD Refinery CapacityOECD Refinery Capacity–– OECD Refinery Throughput OECD Refinery Throughput

Target Predictor: WTI Spot Price ($/bbl)Target Predictor: WTI Spot Price ($/bbl)


1212

Data and Sources of DataData and Sources of Data

Data were collected from different Data were collected from different sources:sources:–– Energy Information Administration Energy Information Administration

website (EIA).website (EIA).–– The Annual Statistical Bulletin of OPEC The Annual Statistical Bulletin of OPEC

Website.Website.


1313

Data DefinitionsData Definitions

Production OPECSupply Demestic Total USINDEX_US

Production OPECDemand WorldTotalINDEX_WORLD

Production OPECT.) Ref. C., Ref.n,Consumptiomand,Average(DeINDEX_OECD

Production OPECReserves Proved OPECodPr/.sRe

=

=

=

=


1414

Data TrendsData Trends

0.00

5.00

10.00

15.00

20.00

25.00

30.00

35.00

40.00

45.00

Jan-

80Ja

n-81

Jan-

82Ja

n-83

Jan-

84Ja

n-85

Jan-

86Ja

n-87

Jan-

88Ja

n-89

Jan-

90Ja

n-91

Jan-

92Ja

n-93

Jan-

94Ja

n-95

Jan-

96Ja

n-97

Jan-

98Ja

n-99

Jan-

00Ja

n-01

Jan-

02Ja

n-03

Date

WTI

Spo

t Pric

e ($

/bbl

)

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

OPEC

Monthly Production (M

BO

PD)


1515


25,000

30,000

35,000

40,000

45,000

50,000

55,000

1/1/

80

1/1/

81

1/1/

82

1/1/

83

1/1/

84

1/1/

85

1/1/

86

1/1/

87

1/1/

88

1/1/

89

1/1/

90

1/1/

91

1/1/

92

1/1/

93

1/1/

94

1/1/

95

1/1/

96

1/1/

97

1/1/

98

1/1/

99

1/1/

00

1/1/

01

1/1/

02

1/1/

03

Date

Diff

eren

t Var

iabl

es (M

BO

PD)

0

5

10

15

20

25

30

35

40

45

WT

I Spot Price ($/bbl)

Refinary Capacity (MBOPD) Refinery throughputs (MBOPD) Average Demand (MBOPD)Consumption (MBOPD) WTI PRICE ($/bbl)


1616


5.00

10.00

15.00

20.00

25.00

30.00

35.00

40.00

45.00

Jan-80 Jan-82 Jan-84 Jan-86 Jan-88 Jan-90 Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02

Date

WTI

($/b

bl),

R/P

0

1

2

3

4

5

6

Index

W TI Price Res/Prod. OECD_INDEX W orld_INDEX US_INDEX


1717

OLS Estimation and ResultsOLS Estimation and Results

0

5

10

15

20

25

30

35

40

45

0 5 10 15 20 25 30 35 40 45

Predicted Price ($/bbl)

True

Pric

e ($

/bbl

)

R2 = 0.796SSE = 5747.07St.Dev.= 6.6764


1818

Model Estimation and ResultsModel Estimation and Results

OLS predictor target can be better OLS predictor target can be better enhanced using GRNN.enhanced using GRNN.


1919

GNNN Estimation & ResultsGNNN Estimation & Results

Training Dataset Correlation

0

5

10

15

20

25

30

35

40

45

50

0 5 10 15 20 25 30 35 40 45 50

Actual Price ($/bbl)

GR

NN

Cal

cula

ted

Pric

e ($

/bbl

)


2020


T estin g D a ta set C o rrela tio n

0

5

10

15

20

25

30

35

40

45

50

0 5 10 15 20 25 30 35 40 45 50

A ctua l Price ($ /bbl)

GR

NN

Cal

cula

ted

Pric

e ($

/bbl

)


2121

GRNN Estimation & ResultsGRNN Estimation & Results

Parameter Training TestingR2 0.8976 0.8666SSE 2678.2807 516.6035St.Dev. 3.3168 3.4659

GRNN Performance


2222

GNNN Estimation & ResultsGNNN Estimation & ResultsTraining Error=WTI_Price-WTI_Calc

-20

-15

-10

-5

0

5

10

15

20

80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03

Year

Tra

inin

g E

rror

($/b

bl)


2323


Testing Error=WTI_Price-WTI_Calc

-20

-15

-10

-5

0

5

10

15

20

80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03

Year

Testi

ng E

rror (

$/bb

l)


2424

ConclusionConclusion

GRNN can predict crude oil prices with a GRNN can predict crude oil prices with a reasonable degree of accuracy, taking into account reasonable degree of accuracy, taking into account different supply and demand levels in OECD different supply and demand levels in OECD Europe, WORLD and North America.Europe, WORLD and North America.The network captured with a better accuracy the The network captured with a better accuracy the correct behavior in comparison to the regular correct behavior in comparison to the regular regression analysis. Although the data plots show regression analysis. Although the data plots show good correlation between the input parameters and good correlation between the input parameters and price data.price data.Price driven indices were used in this study to Price driven indices were used in this study to show the effect of different factors on the behavior show the effect of different factors on the behavior of crude prices of crude prices

predicting crude oil price trends using artificial … · predicting crude oil price trends using...

Documents