Dynamics of the Heavy-Light Spread in the N. American Oil Marketjparsons/Presentations/Heavy...
Transcript of Dynamics of the Heavy-Light Spread in the N. American Oil Marketjparsons/Presentations/Heavy...
Dynamics of the Heavy-Light Spread
in the N. American Oil MarketRomain H. Lacombe and John E. Parsons
MIT Energy and Environmental Policy Workshop6-7 December 2007
2
The Issue
North American crude oil marketsLight sweet crude: global light marketHeavy sour crude: Mexican and Venezuelan oilNew entrant: heavy products from Canadian oil sands
Question: how do heavy and light crude prices relate?Is there a reliable long run equilibrium?• Fixed percent spreads? • Fixed differentials?• Other?
What about the dynamics of the market?• Short-run responses to shocks?• Long-run shifts?
3
The Data
Focus on three key marker crudes: WTI, LLB and MayaWest Texas Intermediate Blend global light crude marketLloydminster Blend Canadian heavy crude market (benchmark for Diluted Bitumen from the Athabasca oil sands)Maya Blend Central and South Am. heavy crude market
Data: weekly prices for the 1998 - 2007WTI: NYMEX front month contract for delivery at Cushing, OKLLB: spot contract for delivery at Hardisty, Alb.Maya: sold CIF to USGC based on Pemex marked price
4
Historical Evolution of Prices
02
04
06
08
01
00
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
WTI 1st Month NYMEX Maya BlendLloydminster Blend
1998/01:WTI: 16.63Maya: 11.12LLB: 6.70
2007/11:WTI: 95.10Maya: 81.98LLB: 67.02
Global rise in prices
Volatility periods:
KatrinaIrak9/11
5
Absolute Spreads: WTI-Maya and WTI-LLB
01
02
03
04
0
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
WTI-LLB Differential Fitted valuesWTI-Maya Differential Fitted values
01
02
03
04
0
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
WTI-LLB Differential Fitted valuesWTI-Maya Differential Fitted values
Shell shutdowns
Suncor shutdowns
Hurricane Wilma
1998/01:LLB: -9.93Maya: -5.51
2007/11:LLB: -28.08Maya: -13.12
6
20
40
60
80
100
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
Fitted values Fitted valuesMaya/WTI Spread (%) LLB/WTI Spread (%)
Percent Spreads: Maya/WTI and LLB/WTI
1998/01:LLB: 40.9%Maya: 64.6%
2007/11:LLB: 63.5%Maya: 82.8%
9/11Hurricane Keith?
Hurricanes Ivan & Jeanne
Hurricane Katrina
7
Early Conclusions
No simple long run equilibrium relationshipFixed price differentials exhibit heteroskedasticityFixed percent spreads are shifting with time
Differential shocks impact all marketsGlobal shocks have differentiated local effectsLocal shocks have repercussions on other markets
Need for thorough time series analysis
Time Series Analysis
9
Estimating a Model of Price DynamicsProblem in inference on time trended time series…
very easy to erroneously find a relationship between 2 series if they are not stationaryE.g. oil prices went up while steel price went up too: Causality? Correlation?
10
Estimating a Model of Price DynamicsProblem in inference on time trended time series…
very easy to erroneously find a relationship between 2 series
One solution is to first detrend the series, e.g., by taking first differences
this works sometimes, but the underlying problem is sometimes more subtle and undermines the validity of this simple solutionE.g. for oil and steel -- if energy prices impact steel price, the following structure may prevail:
In that case, differencing ignores long run equilibrium between the variables due to the shared stochastic trend
εε
βα
OilEnergyOil
SteelEnergySteel
PPPP
+=+=
11
Estimating a Model of Price DynamicsProblem in inference on time trended time series…
Very easy to erroneously find a relationship between 2 series
One solution is to first detrend the series, e.g., by taking first differences
This works sometimes, but the underlying problem is sometimes more subtle and undermines the validity of this simple solution
Resolution: cointegration analysisSearch for the cointegration vector… a more robust search through a broader universe of possible stationary linear combinations of the non-stationary variablesIf variables cointegrated…
stationary reversal to a long run equilibriumPP OilSteel βα
−
12
Traditional Diagnostics
Standard VAR(p) model:
Test lag order p
Standard estimation method: VAR(p) modelWorks for stationary variablesStandard form assumes no contemporaneous effect of variables on each other Structural form (informed by standard form) can allow contemporaneous effects
ε t
p
i itit PAP +∑==
−1
Structural VAR(p) model:
ε t
p
i ititt PABPP +∑+==
−1
Structural assumptions
13
Traditional Diagnostics Estimation method for non-stationary variables: VECM model
First differences of VAR(p) model in standard formImplies linear combination of lagged price levels is stationaryHence need to choose a constraint on rank Johansen test
VECM(p,r) model:
ε tt
p
i itit PPP ++∑= ΠΔΠΔ −
−
=− 1
1
1
Test lag order p
Johansen test for rank r Stationary linear combination
CECM(p,r) model:
ε tt
p
i itit PPP ++∑= ΠΔΠΔ −
−
=− 1
1
0Structural
assumptions
14
Overview of the Path for Estimation
Standard VAR(p) model:
Test lag order p
ε t
p
i itit PAP +∑==
−1
Structural VAR(p) model:
ε t
p
i ititt PABPP +∑+==
−1
VECM(p,r) model:
ε tt
p
i itit PPP ++∑= ΠΔΠΔ −
−
=− 1
1
1
Test lag order p
Johansen test for rank r
CECM(p,r) model:
ε tt
p
i itit PPP ++∑= ΠΔΠΔ −
−
=− 1
1
0
Unit root test
Stationary VAR Integrated VECM
15
Traditional Diagnostics: Unit Root Tests
Philipps-Perron unit root testNull hypothesis: price variables exhibit a unit root
First differences are found stationary by the same test
Conclusion: price variables are integrated of order 1 they behave like random walks
Therefore… need for co-integration analysis!VECM to reveal long run equilibrium and link with short run dynamicsCECM if specific structure is found
79.29%67.94%90.25%
P-value for null hypothesis
log Mayalog LLBlog WTIVariable
Variables exhibit unit roots
16
Bottom-line: Cointegration of Crude PricesPart #1. Long-run equilibrium relationship: co-integration
framework between WTI, Maya and LLBDiagnostics: lag order 4, rank 2Reveals long run equilibrium
Part #2. Linking short-run to long-run dynamics: Vector Error Correction Model (VECM)
Highlights relationship between long run equilibrium and short rum dynamicsReveals underlying asymmetry between WTI and the other variables
Part #3. Imposing structure on short run dynamics: Conditional Error Correction Model (CECM)
WTI is assumed exogenousWe study its contemporaneous and long-run effect on heavy crudes prices
17
Part #1Bottom Line: Long-run Equilibrium
Long run equilibrium between LLB and WTI:log LLB = (- 1.0613) + (1.115015) log WTI
Predicted ‘equilibrium’ in price levels:
0
20
40
60
80
100
30 40 50 60 70 80 90 100
WTI price
LLB
pric
e
LLBWTI
@$30: 51% spread to
WTI
@$100: 59% spread to
WTI
18
Part #1Bottom Line: Long-run Equilibrium (cont.)
Historical pricesActual and predicted prices Departure from equilibrium
02
04
06
08
0
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
Lloydminster Blend LLB (LR)
-10
-50
51
0
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1w_time
Disequilibrium (LLB to LLB LR) Reference
19
0
20
40
60
80
100
30 40 50 60 70 80 90 100
WTI price
LLB
pric
e
MayaWTI
Part #1Bottom Line: Long-run Equilibrium (cont.)
Long run equilibrium between Maya and WTI:log Maya = (- .2773277) + (1.02387) log WTI
Predicted ‘equilibrium’ in price levels:
@$30: 82% spread to
WTI
@$100: 85% spread to
WTI
20
Part #1Bottom Line: Long-run Equilibrium (cont.)
Historical Maya pricesActual and predicted prices
02
04
06
08
0
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1Time (weekly)
Maya Blend Maya (LR)
Departure from equilibrium
-10
-50
5
1998w1 2000w1 2002w1 2004w1 2006w1 2008w1w_time
Disequilibrium (Maya to Maya LR) Reference
21
Part #2Bottom Line: Short-run Dynamics
Shocks to WTI Affect LLB and Maya in the short runImpose a strong drag to equilibrium on both heavy crudes
Shocks to LLB and MayaAffect WTI in the short runBut drag to equilibrium is not significant: WTI is weakly exogenous
Shocks to LLB Affect Maya in the short runImbalance between LLB and WTI affects Maya in the long run
Shocks to Maya Affect LLB in the short runImbalance between Maya and WTI does not affect LLB
22
Part #2Bottom Line: Short-run Dynamics (cont.)
Shocks to WTI cause short run shocks to Maya and LLBOnce WTI is stabilized, shocks are persistent and impact long run prices of Maya & LLB
Convergence to long-run equilibrium takes over after 9 weeks
-0.10
-0.05
0.00
0.05
0.10
0 5 10 15 20 25 30 35 40 45 50
Time (weeks)
Shoc
k to
log
varia
bles
DeltaWTIDeltaMayaDeltaLLB
23
Part #2Bottom Line: Short-run Dynamics (cont.)
30
35
40
45
50
55
60
65
70
0 5 10 15 20 25 30 35 40 45
WTI
LR Maya
LR LLB
Maya
LLB
WTI price shockLong term
persistence: 57.1% of
initial shock
Long run pass-through to Maya: 75% of persistent shockLong run pass through to LLB: 54% of persistent shock
Initial adaptation of prices
Long run convergence to
equilibrium
24
Part #2Bottom Line: Short-run Dynamics (cont.)
Shocks to LLB cause short term shocks to other variables Once other variables have stabilized, LLB has limited further impact on long-run prices
Convergence to long-run equilibrium takes over after 5 weeks
-0.10
-0.05
0.00
0.05
0.10
0 5 10 15 20 25 30 35 40 45 50
Time (weeks)
Shoc
k to
log
varia
bles
DeltaWTIDeltaMayaDeltaLLB
25
Long run pass-through to WTI: 43% of initial shockLong run pass through to Maya: 42% of initial shock
Part #2Bottom Line: Short-run Dynamics (cont.)
30
35
40
45
50
55
60
65
0 5 10 15 20 25 30 35 40 45
WTI
LR Maya
LR LLB
Maya
LLBLLB price shock
Initial adaptation of prices
Long run convergence to
equilibrium
Long term persistence:
31.2% of initial shock
26
Part #2Bottom Line: Short-run Dynamics (cont.)
Shocks to Maya cause short term shocks to other variables Once other variables have stabilized, Maya has no further impact on long-run prices
Convergence to long-run equilibrium takes over after 6 weeks
-0.10
-0.05
0.00
0.05
0.10
0 5 10 15 20 25 30 35 40 45 50
Time (weeks)
Shoc
k to
log
varia
bles
DeltaWTIDeltaMayaDeltaLLB
27
Part #2Bottom Line: Short-run Dynamics (cont.)
30
35
40
45
50
55
60
65
0 5 10 15 20 25 30 35 40 45
WTI
LR Maya
LR LLB
Maya
LLB
Maya price shock
Long run pass-through to WTI: 16.0% of initial shockLong run pass through to LLB: 13.6% of initial shock
Initial adaptation of prices
Long run convergence to
equilibrium
Long term persistence:
19.1% of initial shock
28
Part #3Bottom-line: Exogenous impact of WTI
Implications of the VECM:Short run and long run movements of heavy oil prices are linked to WTI price through different channelsHowever, the model misses the contemporaneous effect of WTI on other variables
New model: Conditional Error Correction Model (CECM)WTI is assumed exogenous with a contemporaneous effect on heavy crudesResult: fit is much better! (R2 = 12% 52% for LLB, 8% 59% for Maya)But we loose information on the feedback from heavy crudes to WTI
29
Part #3Bottom-line: Exogenous impact of WTI (Cont.)
CECM estimates the following short run dynamics:
-0.10
-0.05
0.00
0.05
0.10
0 5 10 15 20 25 30 35 40 45
Time (weeks)
Shoc
k to
log
varia
bles
DeltaWTIDeltaMayaDeltaLLB
Differential hedging ration between long run and short runHedging ratios are dependent on the level of prices
WTI price shock
Short term response
Long term response
Implications
31
Optimal Hedging StrategyFor a natural long with heavy oil to sell:
There is no futures contract on heavy oilCan one hedge with the NYMEX WTI front month contract? CECM
Naïve hedging strategySingle, unconditional hedge ratio with NYMEX WTI 1st month to 1 year swapBMO (formerly Bank of Montreal): 78.1%
32
Optimal Hedging Strategy
33
Optimal Hedging StrategyFor a natural long with heavy oil to sell:
There is no futures contract on heavy oilCan one hedge with the NYMEX WTI front month contract? CECM
Naïve hedging strategySingle, unconditional hedge ratio with NYMEX WTI 1st month to 1 year swapBMO (formerly Bank of Montreal): 78.1%
Conditional long run strategyConditional hedge ratio for NYMEX WTI 1st month contractWTI @ $30/bbl ratio 51%, vs. WTI @ $100/bbl 59%
Short-run strategySingle hedge ratio for NYMEX WTI 1st month contract: 84.5%Position informed by reversal to long run equilibrium
The End