1	2005 CEFER Conference On the Predictive Power of Crude-Oil Forward Prices Houston, February 18, 2005 Vincent Kaminski Rice University
2	The Challenge Marking-to-market long-term positions in trading portfolios Committing capital to long-term projects Requires assumptions about future price levels The rationale for using forward price curves in trading and investment decisions: The triumph of optimism over experience
3	EXPERIENCE
4	Construction of Forward Price Curves in Practice Forward price curves used in the practice of the energy markets are constructed, in most cases, in a very eclectic way, especially for longer tenors The front segment of a forward price curve is typically taken from a futures contract traded on an organized exchange, possibly adjusted for basis For longer tenors, the industry uses information from the OTC markets coming in the form of calendar, mostly year-on-year, spreads A trader has to apply seasonality coefficients to the annual or quarterly spreads The back of the curve typically comes from a fundamental model This is a highly stylized description of the current practice. Every market has its own conventions and standards.
5	Construction of a Forward Curve
6	Examples: Natural Gas (1) Hockey stick natural gas forward price curve in the early 1990s The long term prices (beyond three years) were driven by the cogen industry The long term prices reflected unique regulatory framework (PURPA) and market structure (captive long-term buyers, shrewd speculators) The long-term prices collapsed when deregulation undermined the cogen business
7	Examples: Natural Gas (2) The underlying volumes in the back of the curve were small Example: Hedging a purchase of a large natural gas field could take a year of accumulating trading positions Lessons learned: The forward price curves contain information for specific volumes Market structure matters Keynes correctly pointed out that the balance between natural longs and natural shorts matters
8	Examples: Power (1) “Dash to gas” in the late 1990s and early 2000s lead to considerable excess of generation capacity in many regions of the US The investment decisions were based on the “forward price curves” based on two approaches Fundamental models of supply and demand Steady state equilibrium concept
9	Examples: Power (2) Steady state equilibrium Power plants will be constructed in the future, therefore power prices will support the business of producing power Fundamental models require multiple assumptions about: Fuel prices Future levels of economic activity Generation and transmission capacity Market design and customer behavior
10	Examples: Power (3) Lessons learned Forecasts fail because people act (or do not act) on forecasts Small changes in assumptions to the fundamental models produce huge changes to the end results that accumulate over time. A wide range of forecasts can be generated through manipulation of inputs. Compensation mechanisms and organizational pressures influence the choice of models. Moral hazard leads to the creation of a market for forecasts.
11	Examples: Manipulation Small volumes of transactions for long tenors allow to create the illusion of a true market In the best case, the long-term forward prices reveal what a few traders think (or do not think) about long-term prices. There is only limited information discovery and aggregation.
12	THEORY
13	Empirical Tests of Predictive Power of Futures Prices Most tests use short term futures prices Practically all tests use regression or cointegration analysis If you saw a few, you have seen all of them In general, the evidence is mixed and varies from commodity to commodity and time period to time period Recommendations vary from one extreme to another (full reliance on the futures prices to contrarian actions) Two examples: Menzie Chinn, Michael Le Blanc, Olivier Coibion, NBER, 2005 Benjamin MirandaTabak, 2003
14	Chinn and Others Chinn and others used 1990 – 2004/October NYMEX data (except for the natural gas contract that opened in 1991) Futures are unbiased predictors of crude, gasoline, heating oil at the 3 months horizon (but not in the case of natural gas) Futures prices explain a small proportion of the variation in the underlying commodity prices ARMA models don’t offer superior forecasting performance compared to futures prices
15	Chinn and Others (2) Futures and spot price relationship is given by f(t,t+k) – s(t) = d(t,t+k) + Q(t,t+k) f(t,t+k) – futures price of the contract maturing at the time t+k, as of time t s(t) – spot price at time t d(t,t+k) - cost of carry Q(t,t+k) – adjustment for the MTM feature of the futures contract All variables used are represented by their logarithmic values Regression equation used by the authors s(t+k) – s(t) = b₀ + b₁ (f(t,t+k) – s(t))+ e_t b1 is equal to 1 if the basis is the optimal predictor of the change in the spot price It is assumed that the log spot price follows the a random walk with a drift and expectations are rational The idea goes back to Fama (1984) Limitations of the cost of carry models and random walk for commodities are well known
16	Chinn and Others: Reported Adjusted R² (3) Crude oil 3 months: 0.05 6 months: 0.06 12 months: 0.10 Natural gas 3 months: 0.17 6 months: 0.25 12 months: 0.35 Gasoline 3 months: 0.09 6 months: 0.18 12 months: 0.23 Heating Oil 3 months: 0.15 6 months: 0.13 12 months: 0.17
17	Tabak: Reported Adjusted R² Tabak investigated short term Brent IPE contracts, January 1990 to December 2000 Fama (1984) was used as in the case of the previous study for the logarithms of the prices Slope estimates b₁ and adjusted R² One - month contract: 0.81857, 15.57% Two - months contract: 0.95326, 19.58% Three – months contract: 1.01927, 19.02% The expectation hypothesis (futures prices predict realized spot prices) implies parameters restrictions for b₀=0 , b₁=1. Joint test was performed for H0: b₁=1, b₁=1. The hypothesis could not be rejected for all the three contract maturities included in the study. The hypothesis of unbiased predictive power of the futures oil prices is rejected if the full sample is divided into sub-samples corresponding to the periods of upward or downward trending prices. In any case, the predictive power of the futures prices is low.
18	Tabak: Cointegration Analysis (1) Engle-Granger approach was used for the futures and spot prices The null hypothesis of non-stationarity could not be rejected for the one- and two-month contracts Cointegration regressions of the spot prices on the futures prices (a, b, R²) One-months: 0.078525, 0.971888, 80.98% Two-months: 0.122782, 0.961548, 68.92% The residuals of the one-month contract equation are I(0) The results above were confirmed using Johansen (1988) approach Cointegrating vectors: One-months: 1, -0.9981 Two-months: 1, -1.0006 The conclusion: Information contained in one time-series contains the information that allows to predict another one. The model relating the two is relatively simple.
19	WHY USE FORWARD PRICES?
20	The Case for Futures Prices We should use forward prices because 1. It’s cool
21	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart
22	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart 3. My boss said so
23	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart 3. My boss said so 4. Our friendly accountant signed off on our forward price curve
24	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart 3. My boss said so 4. Our friendly accountant signed off on our forward price curve 5. It’s supported by the efficient market hypothesis
25	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart 3. My boss said so 4. Our friendly accountant signed off on our forward price curve 5. It’s supported by the efficient market hypothesis 6. Alan Greenspan is encouraged by the level of the oil futures prices of 2010 maturity
26	The Case for Futures Prices We should use forward prices because 1. It’s cool 2. The engineers don’t know what we are talking about and it makes us look smart 3. My boss said so 4. Our friendly accountant signed off on our forward price curve 5. It’s supported by the efficient market hypothesis 6. Alan Greenspan is encouraged by the level of the oil futures prices of 2010 maturity 7. We don’t know any better
27