Long run relationship between oil prices and aggregate oil investment: Empirical Evidence
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Long run relationship between oil prices and
aggregate oil investment: Empirical Evidence*
Sergio Guerra1
This version May 2007
Abstract
Using the aggregate number of oil rigs as a proxy of oil investment, I
evaluate the bidirectional relationship between oil prices and oil
investment in OPEC and Non-OPEC countries. We take advantage of
Bayesian estimation techniques and innovation accounting to
incorporate the long run dynamics of the oil market without imposing
strong restrictions on its structural form. Our results suggest that
aggregate oil investment reacts to oil price changes as predicted in the
traditional Hotteling model, but the response of oil price to shocks in
oil investment (when controlling for world oil demand and oil
production in both groups) is barely significant and usually transitory.
JEL classification: C11, C32, C53, L71, Q41, Q43
Key Words: Energy Prices, Oil Investments, Vector autoregression, Bayesian
methods.
*
The author would like to thank Osmel Manzano, Francisco Monaldi, Rodolfo Mendez, Giovanni
DiPlácido, Daniel Ortega, Pablo Acosta, José Pineda and Alejandro Puente for their helpful ideas and
comments. All remaining errors are mine.
1
Corporación Andina de Fomento (CAF) and Universidad Católica Andrés Bello. The ideas and
views expressed on this paper are the solely responsibility of the author and do not necessarily reflect
the ideas and views of the Corporacion Andina de Fomento. Comments are welcome at:
sguerra@caf.comI. INTRODUCTION
Investment theory, its consequences and determinants, will always be a relevant topic
in the economic literature. Recent contributions situate oil investment in a prominent
position on that literature. This observation may reflect the role of oil as the main
source of energy, its exhaustibility condition, the relevance of market failures, its
relationship with oil price volatility, and the non-competitive behavior of the market.
Our work seeks to group those characteristics that highlight the importance of oil
investment to study and measure both its aggregate response to shocks in oil prices,
and its influence on the long run behavior of oil prices. In order to do this, we used
the standard Hotelling (1931) approach distinguishing between OPEC and Non-
OPEC groups, and contemporary Bayesian techniques for estimation purposes.
Using the aggregate number of oil rigs as a proxy of oil investment, I evaluate the
bidirectional relationship between oil prices and oil investment. The estimation
follows those suggestion made by Sims (1980), Doan (2004), and partially Enders
(1995), for studying several endogenous variables when its long run comovements
face data restrictions. A Bayesian Vector Autoregressive analysis (BVAR) is then
proposed in order to estimate the model without imposing strong restrictions on the
structural form of the oil market.
One of the basic findings of Hotteling’s approach is that an unexpected temporary
change in oil prices today will not have an impact on investment decisions. This
conclusion allow us to work with an approximation of the long term evolution of oil
price, and as suggested by Pindick (1999), such and approach could be modeled
because prices seem to follow a mean reverting process2. Additionally, as suggested
by Christiano et al. (2006), VAR-based confidence intervals accurately reflect the
actual degree of sampling uncertainty associated with impulse response functions.
This complements one of the main advantages of VAR technique, which is to impose
the minimum amount of restrictions over a specific structural approximation. On the
other hand, Bayesian estimation allow us not just to improve the accuracy of forecast
(Sims et al. 1990), but to deal with the lack of long periods of data, as well as with
potential non-stationary problems.
This empirical approximation is relevant because: (1) we incorporate one of the basic
assumptions of the aggregate oil market, which is that investment decisions have an
impact on oil prices due to its influence in future oil supply and, (2) we examine a
bidirectional dynamic behavior rather than estimate an specific parameter, allowing
us to delimit the uncertainty level surrounding its magnitude.
To capture the actual behavior of oil investment, we used quarterly3 aggregate oil rig
data4, which is a good proxy for countries effort to discover and develop new and
2
Although the rate of mean reversion is slow.
3
Three month average
4
Available at www.bakerhughes.com
2current petroleum reserves levels5. Other variables used in the system where oil
production levels and OECD activity index as proxies for controlling both oil supply
and world oil demand respectively, and deflated Brent crude oil price.
Our results suggest that aggregate oil investment reacts to oil price changes as
predicted by the traditional Hotteling model, which states that only permanent or
expected increases in oil price can affect investment decisions, while the response of
oil price to shocks in oil investment (when controlling for world oil demand and oil
production in both groups) is barely significant and usually transitory. One possible
explanation for this result is that, on aggregate, oil investment is being directed to
less productive phases of the oil fields, limiting the potential increase in supply,
although we do not test for this hypothesis.
The rest of the study is organized as follows; section II briefly describes the
Hotteling model as well as recent theoretical and empirical developments of oil
investment theory. Section III describes the methodological approach suggested and
include the assumptions concerning with the prior used in the Bayesian estimation.
Section V presents the results obtained. Section VI concludes with brief comments.
The data set used it’s explained with detail in the annex A
II. THE THEORY OF OIL INVESTMENT
The purpose of this section is to briefly describe some of the standard contributions
in oil investment theory as well as recent empirical approaches. We begin describing
an extension of the Hotelling (1931) model that incorporate investment decision,
based on a producer maximizing the net present value of a set of cash flows in a
competitive environment. Then we present some considerations about the non-
competitive behavior of oil market and finally present recent empirical approaches.
The standard model of resource extraction was first introduced in Hotteling (1931).
In that approach, each producer has an amount (R) of an exhaustible resource and
faces the decision of extracting a quantity (q) of those reserves, and sells it at a fixed
price (p). The extraction of (q) involves a cost (c), and investment decision involves
cost i(d) for the development of new reserves (d). The amount of reserves evolves
positively with new discoveries and negatively with the amount extracted. This
scheme is presented in the next system of equations and follows Manzano & Guerra
(2006).
5
This data set only exclude the rig count of Russia and Kazakhstan
3The producer faces the following maximization problem
T . .
max ∫ [ p * q − c(q ) − i(d )]e − rt Subject to R = d − q (1) , D = d (2)
q ,d
0
Leading to the next FOCs:
( p − cq )e −rt = λ (1)
.
λ =0 (2)
id e −rt
= λ +ϕ (3)
.
ϕ =0 (4)
With transversality conditions H t = 0 , λt = 0 , Rt = 0 (5)
(1) Is the Hotelling’s main postulate which states that oil price will grow at a positive
real rate “r”, named the scarcity rent, (2) implicates that oil reserves do not
depreciates, (3) states that investment value will equal the shadow value of actual
and new reserves, (4) new reserves do not depreciates, and (5) imposes that when all
oil is extracted or if reserves are left, the shadow value of them is zero.
In this setting, we are assuming a constant price. In this case, an unexpected
temporary change in oil price today will not have an impact on investment. What will
happen is that oil production will increase today to the point where the FOC is met.
However, given that the permanent and final price will not change, the shadow value
of a reserve barrel will not change and because of that, investment will not change.
Therefore, the only way that prices will change investment is that for the firm to face
a permanent increase or an expected temporary increase that will induce investment
before the increase.
A non-competitive extension of this model is presented in Cremer y Salehi-Isfahani
(1991). This extension allows us to incorporate two important implications:
(a) The extraction pattern of the exhaustible resource will be biased towards
conservationism (Q non-competitive < Q competitive up to a certain point in which
the pattern will reverse).
(b) When the number of participants growth to infinite, the situation evolves to the
competitive case.
Aside from the different believes about any particular market situation, this extension
is important because geographic distribution of oil reserves is not homogeneous. Up
4to 78% of proven crude oil reserves belongs to 11 countries that groups OPEC6, and
among those countries, 7 are located in the Middle East. Additionally, as presented in
Lin (2006), if sequential investments are made in neighboring tracts located over a
common pool, firms might interact strategically because of the presence of
information and extraction externalities.
There is no academical consensus about the actual behavior of oil market7. The
discussion was left aside on early 90’s when oil prices went down. But, whatever be
the actual model, the empirical approach that we suggest is flexible enough to reflect
some of the insights concerning any particular market situation.
Recent empirical estimations focus only on the effect of oil prices in investment
decisions. Ringlund et al. (2004), estimates the elasticity of oil rig investment
demand to changes in oil prices in different countries and regions. Their empirical
approximation incorporates unexpected technological changes using a state-space
framework8. In particular, they argue that technological changes follow a stochastic
process similar to a random walk with drift.
Results suggest that oil rig investment respond positively to changes in oil prices,
although it varies in speed and magnitude around the region studied. According to
Ringlund et al. (2004), US will have the large price-elasticity to capital investment in
the oil sector in both the short and the long run. Our principal concern with this
work is that, however they consider regions that include OPEC countries, this result
could be biased due to the omission of a potential double causality between oil prices
and aggregate oil investment.
Another approximation is presented in Casassus et al. (2005) in which the authors
replicate the behavior of future and spot oil price in a DGE environment estimated
using Simulated Method of Moments. They modeled oil prices as a regime-switching
process in where investments are used as a proximity indicator.
In this model, there are two investment “regions”. In other words, investment will
not react to every price move. Only changes big enough will imply a “switch” of sate
that will affect price. Additionally, this work suggests the existence of a risk prime in
oil prices that depends on the distance between the decision to invest and the actual
realization of that investment. Our principal concern with this work is that the
investment cycle basically involves short run periods, and it can be argued that those
cycles are typically larger.
It is clear that oil investments are deeply conditioned on oil price behavior –i.e. the
price that receives a producer for his production. It determinates the future
profitability of those investments that are used in the oil production process, though,
higher prices should lead to larger investments.
6
http://www.opec.org/home/PowerPoint/Reserves/OPEC%20share.htm
7
See Cremer y Salehi-Isfahani (1991) for a literature review
8
Estimated through Kalman filter. See Durbin y Koopman (2001) for a detailed explanation.
5On the other hand, oil investment are in fact the previous phase in oil production –i.e.
it condition the future supply of the exhaustible resource9, which means that
aggregate oil investment (or that made by non-competitive actors) should affect oil
prices using the future supply of the resource as a transmition channel. In other
words, it is necessary to incorporate a bidirectional relation between oil prices and oil
investment when studying aggregate markets.
That’s why our work focuses on a multiequational approximation that allow us to
study and evaluate, qualitatively and quantitatively, the effects that may exist when
oil prices and aggregate oil investment are treated simultaneously. This will be
explicitly explained in the next section.
III. OUR METHODOLOGICAL APPROACH
As exposed previously, there are theoretical justifications to consider aggregate oil
investment as endogenous when studying oil price behavior. Sims (1980) proposes a
non-structural approach to analyze this kind of situation in order to avoid a large
number of restrictions. This is particularly useful when it is not whether variables are
exogenous, and a natural use for a vector autoregressive analysis (VAR).
Given that our work focuses on the study of a particular dynamic behavior, rather
than a parameter estimation, this approach results adequate and optimal.
Additionally, as suggested in Sims et al. (1990) and Mendez (2002), Bayesian
estimation allows to incorporate in a flexible and transparent way additional
information to the relatively scarce time series.
We are going to estimate the following VAR model presented on its reduced form:
5
xt = A0 + ∑ Ai xt −i + et (1)
i =1
Where xt is a (6x1) vector with observables variables such as: OPEC oil production,
Non-OPEC oil production, oil rigs in OPEC countries, oil rig in Non-OPEC
countries, OECD industrial activity index and Brent oil price in real terms. A0 is a
constant (6x1) vector, Ai is a coefficient matrix (6x6) for lagged values of xt , and et
is a vector of error terms. All variables are expressed in logs.
Lag selection follows Doan (2004a), who suggests the use of at least one year and an
additional period for quarterly data in VAR estimations. Regarding the stationarity
condition of the data and the decision to include them in logs, Sims (1980), Sims et
al. (1990), and Doan (2004), recommend against differentiating non-stationary
variables. They argue that the purpose of a VAR is to determine the interrelationship
among the variables, not to determine the parameter estimates. Besides, they suggest
9
Although this relation is not contemporary.
6that differentiating the series throws away relevant information concerning with the
common long run behavior of the data.
Having identifying preliminary transformations of the data and the number of lags in
the VAR system, we proceed to estimate it. As it is going to be described in the next
lines, we propose to do it following Bayesian methods. The discussion follows
Mendez (2002).
The main difference between the Bayesian and classic estimation is the use of prior
information. An illustrative example could be when we exclude a variable in a VAR
system based in own beliefs about the relevance of that variable for the estimation.
In that case, we are implicitly assuming prior information, -i.e. we are absolutely
certain that the parameter associated with that variable equal zero. On the other hand,
Bayesian approach incorporates available a priori information about the probability
distribution associated to that parameter within the system. Sims (1980, 1990),
Litterman, Sims & Doan (1986) and Doan (2004) are pioneers in the use of Bayesian
methods in VAR estimation.
Prior information in BVAR synthesizes all the knowledge within the time
series used in a standard presentation. Common priors include the following general
statistical ideas:
• Aggregate time series usually have a near to one root on its autoregressive
representation.
• Lagged coefficient have a larger probability to be close to zero when the lagged
period is longest (it could be associated with a decaying pattern)
• Own lagged terms of a particular variable tend to be more useful for forecasting
proposes than other explicative variables.
These assumptions on the data generating process of the variables are restrictions
that can be contradicted with the contribution of the data behavior. Standard
Litterman prior extensively incorporates this accepted information about economic
time series, and its variations are known called Minnesota priors. See Doan (2004a)
for a detailed explanation concerning the estimation.
In Litterman prior construction, we can identify three basic conclusions:
• The priors put on the deterministic variables in each equation are non-
informative (flat).
• It is centered on the hypothesis that variables follow a random walk process with
drift.
• It is independent about the number of parameter.
7Based on the next stationary test result, we construct our variation of the Litterman
prior10.
Table 1. Stationary condition of the variables11.
Integration order (ADF)
Variable Label
Level First Difference
Non-OPEC oil production LPNO I(1) I(0)
OPEC oil production LPO I(1) I(0)
Non-OPEC oil rigs LTNO I(1) I(0)
OPEC rigs LTGO I(0) I(0)
OECD Industrial Activity Index LIPIC I(1) I(0)
Brent Real Price LBR I(1) I(0)
• For each equation, first lag of the dependent variable has a prior mean of one,
except OPEC oil rigs cause it reject null-hypothesis of non-stationarity. We then
propose a fuzzy prior distribution with a coefficient near to 0.5.
• Uncertainty associated to the standard deviation of each coefficient, different
from that of first lagged value of dependent variable, was centered in 0.4 (instead
of its standard value 0.5) as suggested by Doan (2004) when VAR systems
contains more than five variables.
• Lag decaying pattern used was harmonic with a 2.0 value, as suggested in Doan
(2004a).
• Also as suggested in Doan (2004), global uncertainty parameter (Overall
tightness) was fixed at 0.2.
Table II. The Prior
Overall Tightness 0.2
Harmonic Lag decay 2.0
Type: Symmetric
S.D. LPO LPNO LTGO LTNO LIPIC LBR
LPO 1.0 0.4 0.4 0.4 0.4 0.4
LPNO 0.4 1.0 0.4 0.4 0.4 0.4
LTGO 0.4 0.4 1.0 0.4 0.4 0.4
LTNO 0.4 0.4 0.4 1.0 0.4 0.4
LIPIC 0.4 0.4 0.4 0.4 1.0 0.4
LBR 0.4 0.4 0.4 0.4 0.4 1.0
First own lag 1.0 1.0 0.5 1.0 1.0 1.0
Preceding the estimation, we also need to define an orthogonalization method for the
system in order to examinee shocks across variables,
To take advantage of innovation accounting12, we need error terms to be uncorrelated
both across time and equations. We used Choleski factorization methodology13 with
10
See Doan (2004) for a detailed explanation
11
See annex C for a detailed explanation.
12
Impulse-Response functions and Variance Decomposition
See Enders (1996). That imposes a number of ( n − n) / 2 restrictions to the reduced form of the
13 2
model
8the following order of the variables14: OPEC production, Non-OPEC production,
OPEC oil rigs, Non-OPEC oil rigs, OECD industrial activity index, and real oil price.
Synthesized as follows, we will present four principal ideas for the use of VAR
methodology, and three more for its Bayesian estimation.
Why a VAR analysis?
• We intend to determine the interrelationships among variables, not to determine
the parameter estimates.
• Oil price is needed to be treated endogenously when aggregate information about
supply and investment are considered.
• The remaining variables could be endogenous among themselves.
• It’s possible to identify and evaluate empirical effects as well as transmission
mechanism.
Why Bayesian Estimation?
• It permits to minimize the associated sacrifice in model estimation with both a
high number of parameters and scarce of data availability.
• Allow us not to differentiate and then to retain log run comovements of the data.
• Allow us to incorporate well accepted assumption about time series.
III. ESTIMATION RESULTS
In this section we will present the main findings of the estimation focusing on
impulse-response functions as well as in variance decomposition15 for the selected
variables.
Figure I Oil Investment response to shocks in oil price.
(OPEC-continuous line; Non-OPEC-dotted line)
0.05
0.04
0.03
Var.%
0.02
0.01
0.00
0 5 10 15 20 25 30 35
Note: Confidence interval suggest noisy behavior, however trends are clear.
14
Contemporary order of the variables where studied through or causality test which are presented in
annex B
15
See annex D for details.
9As evidenced in Figures I, aggregate oil investment for OPEC and Non-OPEC
groups respond positively to shocks in oil price. Both reach a maximum response
near to 6th to 7th quarter, however, Non-OPEC oil response at this point surpass
OPEC investment by 2,5 percentage points, basically doubling it. Those results are
completely related with that of Hotteling (1931). Additionally, variance
decomposition shows that in the long run (30 quarters) OPEC oil investment would
be explained by oil prices by 47,0%, and Non-OPEC oil investment by 39,7%.
Oil production variables for both groups positively respond to shocks in oil
investment, but again Non-OPEC behavior is higher in magnitude. Maximum
response occurs around the 10th quarter. Figure II shows that Non-OPEC average
response to Non-OPEC investment is around 0,6% (which represent 250 thousand
bbl/d of crude oil), meanwhile OPEC variation is around 0,4% (an additional
production of 190 thousand bb/d). Variance decomposition suggest a long term
behavior of OPEC oil production explained around 8% by its own investments, while
for Non-OPEC oil production, this estimate is near to 22,7%.
Figure II. Response of oil production to shocks in oil investment
(OPEP-continuous; No-OPEP-dotted)
0.006
0.005
0.004
Var.%
0.003
0.002
0.001
0.000
0 5 10 15 20 25 30 35
Note: Confidence interval suggest the same patter n as explained
On the other hand, oil production response to shocks in oil prices suggests an
interesting pattern. First of all, Figure III shows that OPEC has a larger capacity to
respond in the short term to a permanent oil price variation than Non-OPEC
producers as a group. Second, the latter has a slow growing but persistent pattern,
contrary to the OPEC group, which seems to react with an overshooting of oil
production in the short term (first 5 quarters) to then permanently reduce its
production level. This kind of response is consistent with a cartel behavior. Variance
decomposition suggest that long term OPEC oil production is explained in a 75% by
its own production capacity and only in a 4% by oil prices. Contrary, Non-OPEC
production seems to be explained in a 33% by its own capacity and a 13% by oil
prices.
10Figure III. Oil production response to shocks in oil price.
(OPEP-continuous; No-OPEP-dotted)
0.0050
0.0025
Var.%
0.0000
-0.0025
-0.0050
0 5 10 15 20 25 30 35
Note: Confidence interval suggest the same patter n. as explained.
The response of Industrialized Activity Index to shocks in oil price (Not presented),
initially seems to decrease up to the 6th quarter, then shows a recovery patter to initial
levels reaching it at the 12th quarter. This behavior, however, its lower in magnitude
than the others presented, and confidence intervals suggest the noisiest behavior. It
can be argued that the activity index of industrialized countries is rather inelastic to
oil price shocks. Variance decomposition suggest that this index is explained by self
variation in a 66,36% and by OPEC oil production in a 14,36%.
The same patter is followed by oil prices when an oil investment shock occurs, even
when considering short run or long run movements in both groups. This can be
appreciated in Figure IV and V. Variance decomposition suggest that long term oil
price its explained only by 8% to OPEC changes in oil investment and 12% in case
of Non-OPEC group.
Figure IV. Figure V.
Oil price response to shocks in Oil price response to shocks in
OPEC oil investments Non-OPEC oil investments
0.20
0.04
0.15
0.03
0.10 -2 SD -2 SD
LBR 0.02 LBR
+2 SD +2 SD
0.05
0.01
Var. %
Var. %
0.00
0.00
1 4 7 10 13 16 19 22 25 28 31 34 37 40
1 4 7 10 13 16 19 22 25 28 31 34 37 40
-0.05
Quarters -0.01
Quarters
-0.10
-0.02
-0.15 -0.03
-0.20 -0.04
11These results could be taking place for several reasons, and we will suggest two of them that could initiate a future research agenda. (1) As suggested in Hotteling (1931), the price evolution of an exhaustible resource is explained by both, technology and relative scarcity, not by investments. (2) Oil investments do not necessarily traduce in oil production. IV. CONCLUSIONS I have tested the theoretical relation between oil price and aggregate oil investment. In order to do this, we consider oil production and OECD industrial activity index as proxies of both, supply and demand behavior of the market as well as oil rigs used in exploratory and production phases of oil extraction as a proxy of oil investment. We argue that it is necessary to distinct OPEC and Non-OPEC countries in order to evaluate net effects of oil investment in future oil supply and then over oil price. The estimation follows a Bayesian vector autoregressive analysis with prior information suggested in Litterman (1986), Doan (2004) and stationarity test of the variables. Primary tools for evaluate this relation was impulse-response function and variance decomposition. Our findings follow the traditional Hotteling model in sense that oil investment seems to be explained by oil prices in the long term by 47% for OPEC group and 40% for Non-OPEC group. Their response to a permanent shock in oil price are at its highest level of 2,5% and 4,9% respectively, reached out around the 6th to 7th quarter. Additionally, in the long run OPEC crude oil production seems to be explained 8% by self investment decision and 73% by its own production capacity. For Non-OPEC countries, 22.3% of its production is explained by self investment decision, 30.2% by its own production capacity, 22,5% by OECD activity and 13.5% by real oil prices. However, contrary to common intuition, our result suggests that the oil price response to shocks in oil investment seems to be barely significant and usually transitory. This kind of response for oil price should arise in a model when a BVAR with 5 lags and 92 usable observations is run on the data from it. We do not claim about to discovered any particular robust response. This result is against any particular risk-prime on oil prices such as that suggested in Casassus et al. (2006). We suggest that this result could have two different explanations. First, as expressed in Hotelling (1931), long term determinants of an exhaustible resource price are technology and scarcity, not investment, which means that the theory of scarcity rent are in fact a relevant issue. Second, as frequently expressed by analysts16, more oil investment does not necessarily imply a greater oil production. Instead, a greater secondary extraction could be taking place suggesting that new oil findings aren’t in quantity and quality similar to those in the past. This mean that, on aggregate, oil investment is being directed to less productive phases of the oil fields, limiting the potential increase in supply 16 See for example http://www.iea.org/Textbase/press/pressdetail.asp?PRESS_REL_ID=159, and UBS (2007)
The framework presented here could be extended. One possible extension could be to
incorporate oil stock information, which up to day isn’t available for the period
studied. Additionally, a detailed research agenda about the quality of oil investment
in OPEC and Non-OPEC groups is required in order to completely understand the
former results.
REFERENCES
Abel, Andrew, Avinash K. Dixit, Janice C. Eberly and Robert S. Pyndick (1995).
“Options, the value of Capital, and Investment” NBER Working Paper 5227,
National Bureau of Economic Research. Cambridge MA.
Aune Finn Roar, Solveig Glomsrod, Lars Lindholt and Knut Einar Rosondahl.
(2005) “Are high oil prices profitable for OPEC in the long run?” Discussion Paper
No. 416. Statistics Norway, Research Department.
Casassus, Jaime, Pierre Collin-Dufresne y Brian R. Routledge. (2005)
“Equilibrium Commodity Prices with Irreversible Investment and Non-Linear
Technology” NBER Working Paper 11864, National Bureau of Economic
Research. Cambridge MA.
Cremer, J. y Salehi-Isfahani, D. (1991) Models of the oil market. Harwood
Academic Publishers.
Christiano, Lawrence J., Martin Eichenbaum, and Robert Vigfusson. (2006).
“Assessing Structural VARs.” NBER Working paper 12353,. Forthcoming NBER
Macroannual.
Dees, S., Pavlos Karadeloglou, Robert K. Kaufmann y Marcelo Sanchez. (2004)
“Does OPEC Matter? An Econometric Analysis of Oil Prices” Forthcoming in The
Energy Journal.
Dixit, Avinash K., y Pindyck, Robert S. (1994) Investment under uncertainty.
Princenton, NJ. Princeton University Press.
Doan, Thomas A. (2004) RATS User’s Guide Version and RATS Reference Manual,
Version 6.0. Estima. Evanston IL.
Elder, John y Peter E. Kennedy (2001), “Testing for Unit Roots: What Should
Students Be Taught?” Journal of Economic Education. Vol. 31, No. 2, p. 137-146.
Enders, Walter. (1995) Applied Econometric Time-Series. John Wiley and Sons,
New York. Second edition: 2004.
Enders, Walter (1996). RATS Handbook for Econometric Time Series. John Wiley
Sons Ltd. USA
13Farzin, Y.H. (2001) “The Impact of Oil Price on Additions to US Proven Reserves”
Resource and Energy Economics. Vol. 23, N. 3.
Guerra, Sergio (2006) “Interrelación dinámica entre el precio del petróleo y la
inversión petrolera mundial,” Tesis de grado para optar por el titulo de Economista.
Universidad Católica Andrés Bello.
Hotteling, H. (1931) “The economics of exhaustible resources” The Journal of
Political Economy. Vol. 39, No. 2, p. 137-175.
Leeper, Eric M., Sims, Christopher A. y Tao Zha (1996) “What Does Monetary
Policy Do?” Brookings Papers on Economic Activity. Vol. 1996, No. 2, pp. 1-78.
Lin, Cynthia (2006) “Do firms Interact Strategically? A Structural Model of the
Multi-Stage Investment Timing Game in Offshore Petroleum Production,” MIMEO
Litterman, R. (1986) “Forecasting with Bayesian Vector Autoregressions: Five Years
of Experience,” Journal of Business & Economic Statistics, Vol. 4, No. 1., pp. 25-38.
Litterman, R., Sims, C. y Thomas Doan (1986). “Forecasting and Conditional
Projection Using Realistic Prior Distributions”, Research Department Staff Report
93, Federal Reserve Bank of Minneapolis.
Manzano, Osmel y Sergio Guerra (2006) “Oil Investment Volatility and
Institutions”. MIMEO.
Méndez M. Rodolfo. (2002) “Modelos de Vectores Autoregresivos Bayesianos
(BVAR) para el pronóstico del Producto y los Precios en Venezuela”. Papel de
trabajo, Banco Central de Venezuela.
Pindyck, Robert. (1999) “The long-run evolution of energy prices,” The Energy
Journal, Vol. 20, Issue No. 2, pp. 1-27.
Ringlund, Guro, Rosendahl, Knut and Terje Skjerpen (2004) “Does oilrig activity
react to oil price changes? An empirical investigation”. Discussion Papers No. 372,
Statistics Norway, Research Department.
Sims, Christopher (1980). “Macroeconomics and Reality” Econometrica. Vol. 48,
No. 1, pp. 1-48.
Sims, Christopher A., Stock, James H. y Mark W. Watson (1990) “Inference in
Linear Time Series Models with some Unit Roots” Econometrica. Vol. 58, No. 1,
pp. 113-144.
Stuart, Jan (2007) “Q-Series®: Oil prices, what’s normal?”, UBS Investment
Research
14ANNEX A: THE DATA17
Variable Label Scale / Units Range Frequency Source
Energy Information
NON-OPEC oil Jan-73 to Quarterly
LPNO 000’ bbl/d Administration
production Jun-06 Average
(EIA)
Note: The series was seasonally adjusted using TRAMO-SEATS procedure. Quarterly averages were constructed
from monthly data. Source: http://www.eia.doe.gov/emeu/mer/inter.html
Energy Information
Jan-73 to Quarterly
OPEC oil production LPO 000’ bbl/d Administration
Mar-06 Average
(EIA)
Note: Quarterly averages were constructed from monthly data.
Source: http://www.eia.doe.gov/emeu/mer/inter.html
Baker Hughes
Number of Jan-82 to Quarterly
NON-OPEC rigs LTNO Incorporated
rigs Jun-06 Average
Oilfield Services
Note: The series was seasonally adjusted using TRAMO-SEATS procedure. Quarterly averages were constructed
from monthly data. Source: http://www.bakerhughes.com/investor/rig/index.htm
Baker Hughes
Number of Jan-82 to Quarterly
OPEC rigs LTGO Incorporated
rigs Jun-06 Average
Oilfield Services
Note: Quarterly averages were constructed from monthly data.
Source: http://www.bakerhughes.com/investor/rig/index.htm
OECD Industrial Index Jan-73 to
LIPIC Quarterly IMF
Activity Index 2000=100 Mar-06
Note: Seasonally adjusted from the source.
Source: http://www.imfstatistics.org
Constant Jan-73 to Quarterly
Brent Real price LBR IMF
1997 USD Mar-06 Average
Note: Quarterly averages were constructed from monthly BRENT data. Previously the series was deflated using
EE.UU. CPI 1997=100. Source: http://www.imfstatistics.org, http://www.bls.org
ANNEX B: BLOCK EXOGENEITY TEST
Multivariate generalization of Granger-Causality tests
Null: The Variables are exogenous to the remain variables
Label Chi-sq Significance
Level
ÅExogenousÆ
LPO 11,36 0.0034
LPNO 14.75 0.0006
LTGO 18,90 0.0001
LTGNO 21,03 0.0003
LIPIC 29,49 0.0000
LBR 59,19 0.0000
17
Capital L preceding labels suggest logarithmic transformation.
15ANNEX C: INTEGRATED ORDER OF THE VARIABLES
Producción Petróleo OPEP
35000 10.5
32500 10.4
10.3
30000
10.2
Thousand bbl/d
Thosand bbl/d
27500
10.1
25000
10.0
22500
9.9
20000
9.8
17500 9.7
15000 9.6
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
PO LPO
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notes
LPO T AIC=12 -3.43 -3.13 -1.73 6.34 5.39 2.10 La serie no es estacionaria.
BIC=0
Perron Unit Root and Structural Change Æ No puede rechazar la presencia de una raíz unitaria.
Log Producción Petróleo OPEP S.A.
Primeras diferencias
0.20
0.15
0.10
0.05
-0.00
Var%
-0.05
-0.10
-0.15
-0.20
-0.25
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
DLPO
ADF Joint Test
Label Det. Lags 5% 10% T-calc 5% 10% F-calc Notas
DLPO None AIC=2 -1.95 -1.62 -10.24 - - - Estacionaria
BIC=1
Producción Petróleo No-OPEP
Seasonally Adjusted
45000 10.7
42500
10.6
40000
Log Thosand bbl/d
37500 10.5
Thousand bbl/d
35000
10.4
32500
30000 10.3
27500
10.2
25000
22500 10.1
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
PNO LPNO
ADF Joint Test
Label Det. Lags 5% 10% T-calc 5% 10% F-calc Notas
LPNO Trend AIC=3 -3.43 -3.13 -2.38 6.34 5.39 3.54 Raíz unitaria con drift.
BIC=3
16Log Producción Petróleo No-OPEP S.A.
Primeras diferencias
0.03
0.02
0.01
Var%
0.00
-0.01
-0.02
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
DLPNO
ADF Joint Test
Label Det. Lags 5% 10% T-calc 5% 10% F-calc Notas
DLPNO None AIC=2 -1.95 -1.62 -2.67 - - - Estacionaria
BIC=2
Taladros OPEP
450 6.12
6.00
400
5.88
350 5.76
Log # taladros
# taladros
5.64
300
5.52
250 5.40
5.28
200
5.16
150 5.04
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
TGO LTGO
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
LTGO C AIC=18 -2.89 -2.58 -3.76 4.71 3.86 7.23 Estacionaria alrededor de
BIC=2 una media
Nota: Con el criterio de selección de rezagos AIC la conclusión se mantiene.
Log Taladros OPEP
Primeras diferencias
0.10
0.05
0.00
Var%
-0.05
-0.10
-0.15
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
DLTGO
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
DLTGO non AIC=11 -1.95 -1.61 -6.12 - - - Estacionaria
BIC=0
Nota: Con el criterio de selección de rezagos AIC la conclusión se mantiene.
17Taladros No-OPEP
Seasonally Adjusterd
5500 8.75
5000
8.50
4500
8.25
4000
Log # taladros
# taladros
3500 8.00
3000 7.75
2500
7.50
2000
7.25
1500
1000 7.00
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
TNO LTNO
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
LTNO T AIC=8 -3.45 -3.15 -2.40 6.49 5.47 3.69 La serie no es estacionaria
BIC=1
LTNO C AIC=8 -2.89 -2.58 -2.73 4.71 3.86 3.74 La serie no es estacionaria
BIC=1
Nota:
Æ(Componente deterministico: Trend) Con el criterio de selección de rezagos AIC la conclusión del test conjunto se modifica.
El valor F calculado es de 7.92, sugiriendo una comportamiento estacionario en tendencia.
Æ(Componente deterministico: Constant) Con el criterio de selección de rezagos AIC la conclusión del ADF se modifica. El
valor T calculado es de -2.57 sugiriendo un comportamiento no estacionario.
Log Taladros No-OPEP
Primeras diferencias
0.16
0.08
0.00
-0.08
Var%
-0.16
-0.24
-0.32
-0.40
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006
DLTNO
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
DLTNO non AIC=3 -1.95 -1.61 -5.33 - - - Estacionaria
BIC=3
Indice de Actividad Industrial de los países Industrializados
2000=100 S.A.
110 4.75
100
4.50
90
80
L o g In d e x
4.25
In d e x
70
60
4.00
50
40 3.75
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
IPIC LIPIC
ADF Joint Test
Label Det. Lags 5% 10% T-calc 5% 10% F-calc Notas
LIPIC Trend AIC=6 -3.43 -3.13 -2.82 6.34 5.39 4.44 Raíz unitaria con drift.
BIC=1
Nota: Utilizando el criterio de selección de rezagos AIC la conclusión sobre la estacionalidad de la serie no se modifica.
18Log Indice de Actividad Industrial de los países Industrializados
Primeras diferencias
0.06
0.04
0.02
0.00
Var%
-0.02
-0.04
-0.06
-0.08
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
DLIPIC
ADF Joint Test
Label Det. Lags 5% 10% T-calc 5% 10% F-calc Notas
DLIPIC None AIC=5 -1.95 -1.62 -7.77 - - - Estacionaria
BIC=0
Nota: Utilizando el criterio de selección de rezagos AIC la conclusión sobre la estacionalidad de la serie no se modifica.
Brent Real
1997 constant USD
90 4.50
80 4.25
4.00
70
3.75
Log 1997 USD
60
1997 USD
3.50
50
3.25
40
3.00
30
2.75
20 2.50
10 2.25
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
BR LBR
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
LBR T AIC=6 -3.43 -3.13 -2.60 6.34 5.39 3.40 La serie no es estacionaria
BIC=1
Perron Unit Root and Structural Change Æ No puede rechazar la presencia de una raíz unitaria.
Log Brent Real
Primeras diferencias
1.25
1.00
0.75
0.50
Var%
0.25
0.00
-0.25
-0.50
1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
DLBR
ADF Joint Test
Label Det Lags 5% 10% T-calc 5% 10% F-calc Notas
DLBR none AIC=2 -1.95 -1.62 -10.68 - - - Estacionaria
BIC=1
19ANNEXE D: VARIANCE DECOMPOSITION
LPO Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,040 100,000 0,000 0,000 0,000 0,000 0,000
5 0,066 96,293 0,186 0,362 0,818 0,989 1,352
10 0,076 89,619 0,356 2,015 3,973 2,604 1,434
20 0,086 80,410 0,534 5,410 6,309 4,645 2,692
30 0,092 75,756 0,919 7,231 6,362 5,960 3,771
40 0,096 73,025 1,391 8,344 6,479 6,570 4,192
LPNO Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,006 0,066 99,934 0,000 0,000 0,000 0,000
5 0,015 5,701 82,554 1,567 7,397 2,146 0,634
10 0,023 12,297 58,075 2,747 17,499 6,274 3,108
20 0,034 13,728 39,187 2,298 23,202 12,612 8,973
30 0,041 11,685 33,193 1,654 23,415 17,775 12,278
40 0,046 9,751 30,196 1,409 22,674 22,506 13,464
LTGO Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,043 2,101 0,043 97,856 0,000 0,000 0,000
5 0,083 5,303 0,493 76,147 6,661 0,524 10,873
10 0,108 5,312 0,852 47,926 9,398 1,755 34,757
20 0,124 5,206 0,780 36,711 7,540 2,696 47,067
30 0,126 5,237 1,167 35,886 7,384 3,372 46,954
40 0,127 5,172 1,453 35,393 7,422 4,246 46,315
LTNO Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,051 4,663 5,684 0,415 89,238 0,000 0,000
5 0,141 6,854 3,475 1,315 70,255 1,043 17,058
10 0,203 7,810 3,009 2,626 49,771 1,999 34,784
20 0,247 9,071 5,421 2,906 38,006 2,923 41,673
30 0,267 9,952 8,073 2,823 35,250 4,180 39,723
40 0,282 10,045 9,602 2,618 34,019 5,994 37,722
LPIC Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,010 0,026 5,520 0,672 10,978 82,804 0,000
5 0,024 1,102 6,603 1,479 7,452 83,019 0,345
10 0,034 3,917 5,823 6,006 5,389 78,254 0,612
20 0,049 9,706 3,802 11,105 3,824 71,204 0,358
30 0,060 14,367 2,663 13,608 2,700 66,366 0,297
40 0,068 17,952 2,046 15,230 2,078 62,462 0,233
LBR Variance Decomposition (%)
Quarters S. D. LPO LPNO LTGO LTNO LIPIC LBR
1 0,134 2,210 2,067 1,495 8,844 1,553 83,832
5 0,266 3,249 2,722 1,344 4,347 1,712 86,626
10 0,316 4,289 4,578 1,267 2,269 1,788 85,808
20 0,355 6,763 9,086 1,347 2,573 2,388 77,842
30 0,383 8,280 11,666 1,246 2,476 3,819 72,513
40 0,405 8,540 12,941 0,988 0,832 5,794 70,905
20You can also read
