_.

@
e
m
. .
II
Society of Petroleum Engineers
.
SPE 49032
Probabilistic Reserves Lead to More Accurate Assessments
J. R. Schuyler, SPE, Oil & Gas Consultants International
tipyright 1990,Society of Pdrdeum Engineers, Inc.
This paper was preparsd for presentation at the 1998 WE Annual Technkal Conference and
Exhibtion held in Nw OrIeans. LouMana, 2T~ *~*[ 1~ ---
This paper was sslected for presentation by an S= Program Cmmittes fdlmng revisw of
information contained in an abtract suhiM by the author(s) Gntents d the paper, as
presented, have not been reviewed by the Mety of ~deum Engin=~ and are su~~ ~.
mrrection by the author(s), The material, as presented, does not necssariiy reflect any
postion of the Society of Petroleum Engineers, its o~ers. or msrnbem. Papsrs prssented at
SPE meetings are su~ec.t b publication ravisw by Edtirial Committees of the Scciety of
Petroleum EngineeF. -onb reprodudon, distrihtiors, or storsge of any part of this pawr
fur wmemal pup without h written mnsent of the Socisly of Petroleum Engineers is
pmhibtsd, Perrnksion to rep~uce in print is rsstrided to an abstract & nd more than 3)0
wrds, illustrations may not be ~isd. The abstrad must contsin cons~cuous
acimdedgment of where and by Mm the Pper was presented. Write Librarian, SPE, PO.
Box -, Richard*n, TX 1--, U.SA., fas 01-972-952-943S.
Abstract
Property evaluations are key inputs for decision-making.
Reserve category definitions-guidelines to petroleum
property evaluation practi~intentionally produce conserva-
tive reserve valuations. Additional biases, actually errors, are
introduced inadvertently, Many evaluation engineers are
unaware of fundamental statistical processes that apply.
This paper describes two issues arising with the applica-
tion of probabilistic techniques in reserve estimations. The
first is aggregation of properties. This topic is identild
though without guidance, in the new SPE/WPT reserve
definitions. h we obtain proved reserves from a pofiolio of
probable reserves? Does a propertys acquisition tiect its
reserves? Can we get proved reserves by aggregating probable
or possible reserves? Professionals asked about these issues
have admitted mixed levels of understanding and had diverse
opinions as to best practice.
A won~ companion issue is that monetary values are
more important for decision-making than reserve volumes.
Therefore, prices and other economic assumptions are critical.
In probabilistic reserve assessments, probability distributions
express judgments about uncertainties. Forecast calculated
with probabilities lead to more complete and more accurate
evaluations.
New SPWPC Reserve Definitions
Words have long been used to express judgments about
uncertainty. Descriptions such as:
We have good geologic control.
Drilling this location has a high probability of success.
describe risks. However the interpretation is highly variable
depending upon the receivers perception of the words. For
example, one survey by Behnl produced the following results
from people asked to assign probabilities to word descriptors:
Responses of Persons Asked to Convefl
Words Into Probabilities
Inter-Quartile
Statement
I
Range Bounds
It is probable that 20-9870 61173180%
,, .
I 70174178%
r
)70
There is a possibility that 52-92%
There is a chance that 1-70V0 26135/50
It is very improbable that 1-8070
I
5I9I1OVO
The interqtiiletiunds divide ranked data into four goups.
Previously, engineers asked what probability suppo~s a
provedn classification would return answers typically ranging
from 50 to 98Y0.Now we have consistent numeric cotildenee
levels for proved and other reserve categories. I view the new
standards as a wonderfi.d advancement in the practice of
petroleum pro~rty evaluation. Assigning a probability
cotildence does not make the judgment more aeeurate.
However, the probabilities are unambiguous and usefil in
calculations.
Reserves definitions have a long history in the literature.
Some of the best, including those by Cronquis~ and Garb?4
have found their way into JPT. SPE Transactions Reprint
Series #3, Oil and Gas Property E\aIuation and Reserve
Estimates, is the classic historical reference (the series was
discontinued in 1994).
A group of engineers was discussing the 1991 proposed
reserves definition changes5 in late 1991. Someone passed
around a sheet asking us to vote. Who mnong the group is
genemlly in favor of the proposed definition changes? The
suxveyresults were:
G 20 Nos,
G 1 Yes, and
G My position that the then-proposed changes do not go far
enough !
Most aversion seemed attributable to a perception of greater
professional liability if the evaluation engineer is accountable
to a more precise numeric standard. Im unaware of any cases
where this has occurred. Property evaluation is still a
judgment business; the new definitions can be viewed as a
higher standard for documenting evaluation assumptions.
Given the emotional opposition at the time, I was pleased
that the new definitions by SPE and WPC, approved in March
1997Ycame about so quickly and so progressively. There are
179
2 J. R. SCHUYLER SPE 49032
still much dissection and work ahead. This paper hightights
several remaining problem areas and offers corresponding
recommendations.
Calculation MethodYou Have Choices
The new reserves definitions are a wonderful compromise
among constituencies. Most people want to standardize on
only one of the two calculation approaches: probabilistic or
deterministic (i.e., with or without the use of probability
distributions and calculations). Fundamentally, the two
methods are not in conflict and should produce comparable
estimations. The key definition embellishment is that the SPE
and WPCSstandards-setting committms and boards agree that
provedat least for probabilistic reserves-means 90Y0
cotildent. There are some other constraints discussed below.
The language is equivocal with respect to whether
deterministic and probabilistic methods should meet the same
test. I suspect that achieving a compromise agreement required
considerable wordsmithing. Nonetheless, the tone of the
document su~ests to me that the 90% target for proved
applies also to the deterministic method. Thus, if an engineer
is using a deterministic calculation the proved volumes are
90% Cotildent. The probabilistic (I prefer the word stochastic)
method provides more detail from which to pick the p,90 point
(90% cotildent of this volume or better). The engineer using
conventional deterministic calculations can judge what
volume meets the requirement. Another engineer evaluating
the same property with a probabilistic approach should arrive
at similar reserves.
Probabilistic methods have an important advantage,
reproved accuracy. This results from:
G More complete expressions and applications of the
professionals judgments. I have witnessed substantial
adjustments to project values when people with
deterministic models start to consider the inputs as
uncertti Judgments about risks and uncertainties are
expressed as probabilities and probability distributions.
Substituting mean values into the deterministic models
often causes dramatic corrections. What I call the base
case is the deterministic solution when the input
variables are set at their expected values (means).
Considering the possibilities as a distribution rather than
a single-point estimate often has profound effects on the
single-point inputs and deterministic solution.
Note that the best single-point values are not medians
(p.50s) or modes (most likely points). The objective
single-point forecast from a probabilistic forecast is the
mean (expected value).
G Probability distributions often interact in counter-
intuitive ways as they propagate through the evaluation
model calculations. The calculation correctio~78
stochastic variance, is the expected value forecast
minus the base case value. Surprisingly, this
phenomenon is little known. Solving a deterministic
model with ex~ted value @v) inputs does not provide
an EV solution result. Stochastic variance (though not
then named) is the topic of what Pad Newendorp
considers his most important paper.g
I named the correction te~ stochastic variance, with
regard to variance analysis as performed by cost
accountants (not to b confbsed with the varian~
statistic). The varianm to be explained is the differen=
between forecast and actual. Stochastic variance is a
component of the total varianm.
The values of options (decision points) are ofien key
contributors to stmhastic variance. For example, a
marginal property has little value shown by a
conventional, deterministic analysis. A probabilistic
model cotid and should incorporate decision logic to
represent the option to (a) continue or accelerate
production if oil prices increase or (b) shut-in or plug-
and-abandon if prims fall. Much greater value will be
realized from the marginal property if prices rise, yet
have little downside if prices fall.
There are principally two methods for probabilistic
calculations: Monte Carlo simulation and decision tree
analysis, Both we quite straightforward once you know how
they work. They are very different in their calculation method.
Intluence diagrams, a third calculation technique, has
characteristics very similar to decision trees. All of these
calculation tools produce essentially the same answer. One
technique or combination can better serve different situations.
Monte Carlo simulation is usually the best method for
evaluating discovered resources.
Some people are critical of the Monte Carlo method
because of needing to run many, typically hundreds, of trials.
This is merely a characteristic of the method. The simple, yet
elegant, mndom sampling technique is, in effect, propagating
probability distributions through the model calculations.
Integration to solve for expected value is otherwise impossible
for most real-world problems.
Is deterministic ever preferred over a probabilistic
approach? Apparently, yes, as evidenced by many
practitioners steadfastly using deterministic reserves. If there
is little uncertainty the deterministic solution produces
essentially the same result with slightly less work. This is
especially true for petiormance-based reserves. However, if
one considers price uncertainty (Recommendation 8), then the
deterministic calculation falls short. Deterministic evaluations
for mature properties will certainly continue to be popular,
especially if probabilistic prices never come into reserve
reporting.
Recommendation 1. Do not be partitiarly concerned
whether deterministic or probabilistic calcuhtions are used.
The goal is consistent evaluations by way of clear definitions.
I believe the rese~es definitions should clearly state that
deterministic reserves aspire to the same probability levels of
cotildence as defined for stochastic reserves.
180
SPE 49032 PROBABILISTIC RESERVES LEAD TO MORE ACCURATE ASSESSMENTS 3
Recommendation 2. Encourage the use of probabilistic tools
because they (1) better capture expert judgments; (2) better
characterize uncertainty; and (3) provide a more accurate
caltiation restit.
Toward More-Credible Evaluations
Most engineers are employed to provide information for
others to use in making decisions. The hallmark of profession-
alism is credibility. I defined a credible evaluation as one that
warrants the cotildence of the decision maker as being suited
to purpose. Expanding this, the attributes of a credible analysis
include:
G
Objectivity [meaning free from bias in measuring value
that matches the organizations objective(s)]
G Integrity of the model and calculations (that it represents
the stated assumptions)
G A decision model suited to purpose, including Ievcl of
detail -and-
G Adequate disclosure and commtication.
Since inceptiow reserve assessments have been purposely
conservative. The main reasons are (a) a lender relying on the
report is not compensated for assuming inordinate risks, and
(b) a potential investor relying on the report ought not be
defrauded. Better performance is okay, but woe to the
engineer who overstates reserves. Buyers and bankers
maintain that they want conservative estimates, yet are they
really being served? In capital investment analysis we know
that conservative estimates ofien lead to worthy projects being
bypassed. Which is better for the user of a reserves report:
(a) a conservative forecast? -or-
(b) an objective foreeast that also characterizes the
uncertainty of the estimate?
I submit tit the latter is more valuable in decision-making.
The report user is poorly served when information for decision
making is biased, whether conservative or optimistic. In
negotiation or bidding the most important piece of information
is the value of the asset.
A deeision analysis can determine whether compensation
is appropriate for the risks assumed. A conservative risk
policy ean be incorporated into the decision modeI in a
straightforward way using a utility finction to represent the
companys risk attitude.
Once, I heard a mmpanys chief reservoir engineer boast
that his groups reserves were more conservative than their
outside appraisal constitant. Were even more conservative
than (widely respected engineering firm). I suggested that-if
the objective is to internally report lower reserves-his
company could save a lot of money by tiring much of its
engineering sW. Any idiot could can up with lower reserve
numbers. A conservative report does not do owners or bankers
any favors; it just means that the bank engineer deducts less
for risk. Isnt a good forecast the goal of the evaluation?
A person or companys forecasts can be tested for
objectivity. My preference is one or more graphs showing the
distribution of the ratio: forecast estimates divided by actuals.
Alternatively, a cross-plot of forecast versus actuaIs shows the
same information. The actuals for comparing to reserve
reports can be the revised estimates of ultimate recovery
(adjusting out for price changes, new wells in existing
reservoi~ and such). We want evaluations on average to be
neither too high nor too low. Random errors will be observed,
though the average per-project error from many evaluations
should be approximately zero (i.e., mndom errors should be
diluted), The foremost evaluation goal is to be able to look
back on a series of evaluations and * that the mean of
actuals/estimates is nearly one. This should be true in subsets
of the properties as well as the aggregate.
With earefil reeonciliatiow we can test whether reserve
evaluations are hitting the target confidence levels. For proved
resewes, for example, about 90/0of the reserve adjustments
for reservoir performance should be upward, Examining
properties across several years will eliminate some noiw and
provide a better indication. This test applies whether
deterministic or probabilistic methods are used.
Objectivity is listed as the f~st criterion for a erdlble
Evaluation, Also important is a fair degree of precision in the
estimate. While not critical for internal deeisions, preeision is
very important in competitive bidding situations. Common
risk-adjustment factors applied to lower-cotildence reserve
categories bear this out. The width of the distribution of
actuals/estimates is a good way to characterize estimation
precision.
Recommendation 3. Provide re~rt users with expected value
volumes and monetary values. These are the most usefil value
indicators from the analysis. Characterize uncertainty as
distribution cmes (especially needed for competitive bidding
analyses). Use expected values for most decision-making.
Recommendation 4. Recognize that consemative reserve
volumes and monetary amounts may not be suitable for all
report uses. Provide cumulative probability graphs to charac-
terize the uncertainty in volumes and monetary values. Then
the report user can pick off whatever level of cotildence is
appropriate for the purpose. We might eventually be able to
di~nse with proved, probable and possible classifications
altogether.
Recommendation S. Evaluation professionals and firms can
benefit from tracking outcomes versus forecast estimates. The
reconciliation analyses are ofien called post-audits, post-
evaluations, and look-back evaluations. This performance
feedback is perhaps the most important way to improve the
quality of a companys evaluations. Performance fwdback is
most useful, psychologically, when provided to individuals so
that any need for bias-corrections is felt personally.
Reconciling a new versus prior reserves report provides
important fwdback. I recommend showing, in step fashiom
how reserves changed from the prior report: changes due to
production, price changes, same-property revisions, acquisi-
tions, and new work.
181
4 J. R. SCHUYLER SPE 49032
Aggregating Resewes
Oneof the most talked-about issues in reserve evaluations is
aggregating (i.e., summing for consolidation) reserves. The
new definitions only provide a warning statement:
Because of potential differences in uncertainty,
caution should be exercised when aggregating
reserves of different classifications.
The same can be said for adding reserves of the same
classification! Adding volume or monetary values for reserve
categories does not work except for expected values. It is
statistically invalid to add, for example, proved+probable
reserves in mtitiple properties where each is determined at the
507. (p.50) confidence level. ~is caution applies to net
prexent values ~Vs) as well.
It is easier for the reservoir engineer to cIassi& reserves at
the level he or she is evaluating. This classification is done at
the completio~ reservoir, well or field level-whatever is
most meanin@l and convenient. Most ofien the evaluated
volumes are wholly attributed to one category. This saves
some work. Increasingly, different classifications are assi@ed
to dmerence performance scenarios at the same evaluation
unit. For example, a well being analyzed using decline curve
analysis might have three forecasts:
G proved, for a conservative decline (highest decline rate)
G proved+probable, for an intermediate case
G proved+probable+possible, assigned to the volume using
an optimistic decline (lowest decline rate).
With deterministic dculations the pammeters could be
judged so as to approximate the p.90, p.50 and p. 10 volume
cotildence levels. Or, in a probabilistic analysis, the volumes
muld be picked off of a cumdative frequency distribution
generated with Monte Carlo simulation,
Example 1. Assume we have two reservoirs, where the gas
properties volumes are judged as
Reservoir A: lognormal distribution with
P=1O, 0=3.5x 106m3 (~=353, 0=124 Bcf)
Reservoir B: Iognorrnal distribution with
~= 12, - 5.0x 106m3 (w=424, 6=177 Bc~
There are usually common factors between zones. For illus-
tration I assumed a (rank) correlation coefficient p=O.50.
The combined statistics are obtained with Monte CarIo
simtiatio~ using 5,000 trials with Latin hypercube sampling.
Fig. 1 illustrates the distributions for the individual reservoirs
and their sum.
Mean. These always add, regardless of distribution shape
or dependence. The expected value (mean) combined reserves
are 10+12 =22 x 106m3.
Standard Deviation. Variances will add for independent
normal distributions. Had these been independent normal
distributions the standard deviation would be
2 2 1/2
cr=(3.5 +5 ) = 6.10 x 106m3.
This is a usefi benchmark for comparison.
Combining the correlated lognormal distributions yielded a
distribution with standard deviation 6 ~ s = 7.35 x 106 m3.
~is higher value illustrates that positive correlation increases
risk in a portfolio.
Confidence Points. The following table shows standard
cotildence levels for the individual reservoirs and their sums.
The amounts are gas volumes in 106m3.
p.90 p.so p.lo
Reservoir A 6.10 9.44 14.59
Reservoir B 6.63 11.08 18.50
Erroneous SUW 12.73 20.52 33.09
Adding the Above
I Actual Sum I 13.72 I 20.83 I 31.65 I
from Simulation
Zero Correlation Case,
from Simulation
15.10 21.08 29.86
In this simple example proved reserves are almost 8V0higher
using the p.90 cutoff. With a large number of properties, p.90
can be achieved at much lower confidences of the component
properties.
Increasing Proved Reserves by Addition. Conceptually,
having enough probable (p.50) or even possible (p. 10)
reserves can sum to a total that is 90/0 cotildent. This is
because the properdes unique risks are diluted with
diversification.
Aggregating properties will increase proved resemes.
Thus, reserves are improved when properties are purchased by
a larger company. Is this creating value? Most of any real
valueif rmy-in consolidation is through efficiencies and
synergies, not in the aggregation mathematics.
Risks divers~lcation further exacerbates the error of add-
ing reserve categories at the property or other lower level.
With a broad mix of properties, tbe otiy significant correla-
tions will be product prices and, perhaps, technology and
political risks. The relative independence, even negative
correlation, to the broader capital markets should make
upmream petroleum investments especially attractive to
investors.
In my conversations with engineers, many have never
given reserves aggregation much thought. The laws of
probability and statistics to apply, and the method of
aggregating reserves deserves serious attention.
Recommendation 6. Aggregate reserve categories so as to
mmt the target cotildence tests at the report (usually
company) level. That is, given the economic assumptions
182
SPE 49032 PROBABILISTIC RESERVES LEAD TO MORE ACCURATE ASSESSMENTS 5
there is a 90% likelihood that the proved volumes or more will
be produced.
Typically, reserve reports detail the reserve components.
What cofildence points at evaluation component levels will
combine to give proved (p. 90) reserves? Distribution curves
for the evacuation components (completions, reservoirs or
properties) provide the necessary data. Simply seek a common
level of cotildence such that adding the respective quantities
yields the target cofildence level.
In the two reservoir example, p.90 gas reserves are realized
by summing the p.86 levels (judged 86% confident of
ex~dng those volumes). I inte~lated between 50/O-point
eumdative frequency distributions.
Other ways may be suitable strategies to allocate volumes
by property (or by reservoir or completion) and by reserve
category. For example, an easier but less meaningful approach
wotid be to allocate component reserves by means or by p.50
values.
Classifying Reserves by Data Type Rather Than by
Data Quality
The definitions require that probabilistic proved reserves
satis& the more-than-90%-cotildence test. Further, proved
reserves must be delineated by drilling and defined by fluid
contacts or... definitive data. Therefore, presently, the type of
data is also a constraint. Reserves in undeveloped locations
must be within one normal well spacing distance or when the
engineer is reasombly certain of lateral continuity between
wellbores. By the current definition, compelling geostatistics
is not sufficient.
Example 2. Consider a play consisting of 20 pinnacle reefs
identtiled and welldelineated by seismic. One of the reefs has
been drilled and found productive. This proved the play
concept and common factors among the prospects, such as
source and migration.
The productivity of the remaining 19 reefs depnd ordy
upon having adequate porosity and permeability, and these
reservoir characteristics are believed independent between
prospects. Our explomtion expert judges that each prospect
has a .40 probability of success. Discovery oil volumes are
judged to be independent and Iognormally distributed with
p=2.0, c=2.5 106m3 (12.6, 15.7 MMbbl, respectively).
What are the proved reserves? The definitions limit us to
(1) the prospect that has a wellbore in contact with the fornla-
tion and (2) crediting undrilled locations only one appropriate
spacing away from existing wells (if at all).
Is this the best we can do? Consider the statistics. Inde-
pendent events (prospects) with identical chances of success
can be deseribed by the binomial distribution. The probability
of 4 or more additional discoveries is 937. as shown in Flg 2.
Monte Carlo sirmdation allows us to add-in discovery size
distributions. The results, shown as Fig.3 shows a 90% chance
of adding 6.5 x 106 m3 (41 MMbbl) or more in the 19
remaining prospects.
Thus the issue: We can estimate future volumes to be
recovered with reasonable certainty yet not meet all the
proved classification tests. In several ~ormal surveys during
the past months, Ive asked groups about what eviden~ is
stilcient to classifi reserves as proved, Notwithstanding the
definitions, the groups were about evenly split between
whether a wellbore should be required to call the resemes
proved. Admittedly, most of these people were engineers and
geoscientists predisposed toward probabilistic methods.
Suppose, in our example, we have now drilled 10 of 20
alternate prospects across the extent of the play and have
found four fields. Field sizes average 2.2 x 106m3 of oil (13.8
MMbbl). In the remaining 10 prospects of the same quality
there is a 95% chance of at least one additional discovery.
Shodd we book any proved resewes for the remaining
prospects? What if gwscience provided us without a wellbore
a means to detect the presence and movement of oil?
Recommendation 7. Simpl& the reserves catego~ defini-
tions by requiring only the cotildence tests. Do not substitute
additiomd safeguard criteria for professional judgment.
Deterministic calctitions are still acceptable, though they
shodd be performed so as to meet the same cotildence tests.
Rely on professional judgment and insist on communicating
and documendng the risks. Despite the subjectivity involved,
industrys n-s have been met by the competen~ ethical
reserve evaluator. (T. Scott HickmaniO)
Rethinking the Economic Assumptions
Most report users are more interested in mone~ value than
volumes. I am pleased that the new definitions allow latitude
for escalated economics. For new tieIds, prices compare with
reserves in importance, Prices gain in importance as the
properties mature. Then, reserves are usually the major
uncertainty near end of life.
The purpose of reserve definitions has always been to
provide a consistent guideline so as to allow comparisons.
Accountants have a culture where calculation consistency is
more important than objectivity (to my accountant frien@ I
mean this in the best possible context), We engineers tend to
be-foremost-pragmatic. At least this is our belief and quest.
AgaiL the credible evaluation warmnts the cotildence of the
decision maker as being suited to purpose.
People buying or holding investment positions in explora-
tion and production are (or should be) more optimistic than the
broader public about petroleum prices. I am amused when
people selling prospects or similar deals prepare their
economics using conservative eeonomic assumptions. People
optimistic about petroleum prices are going to value properties
the highest and are the likely purchasers.
Recommendation 8. Have an industry standards group agr=
at calendar year-end upon a consensus set of reference
economic assumptions for petroleum property evaluations. Oil
and gas fitures contractsand Delphi polls can help establish a
mainstream forecast for prices and escalations. Allow for
carefilly documented adjustments for location, quality,
183
6 J. R. SCHUYLER SPE 49032
currency exchange rates, and special circumstances. The price
forecasts shodd also characterize uncertainty. Perhaps the
easiest way would be to have hi~ middle (perhaps mean) and
low cases with probabilities for each price scenario.
Discount rate. I cannot resist bringing up a con~oversial
related topic: the present value (PV) discount rate. This is
obviously a very key assumption in property ap raisals. Some
1
finance professionals (e.g., Brealey and Myers ) believe we
should use a risk-free discount rate with Monte Carlo
simulation. I have satisfied myself in modeling shareholder
valuel 2 that this is approximately true. Most companies are
using a risk-adjusted discount rate, the traditional finance
approach. These companies are double-risking projects and
assets in their probabilistic evaluations. The application of
probabilities or conservative judgments adjusts sepamtely for
risk, and we do not want to also recognize risk in the PV
diswunt rate. Of course, comparable reserve reporting
requires agreeing upon a consistent PV discount rate. me
choice dramatically affects the long-lived property assess-
ments.
Recommendation 9. Not all users of a reserve report will
agree on the engineering and economic assumptions. Rather
than presenting alternate scenarios, I recommend spider
diagrams or equivalent sensitivity charts to show how the
report values are changed as the major variables change.
Expected value reserves and expected monetary value (EMV,
ex~cted value net present value, EV ~V) are the key
decision criteria. Sensitivity graphs can show EMV sensitivity
by reserve category by:
G
Average price escalation or reference year price (e.g., the
reference year might be at or near the production half-life
of the properties);
G
Success rates for undrilled statistical reserves; -
G
Initial production rate, decline rate or simil~, and
G
Ultimate recoveries (this obviously applies only to NPV
sensitivity).
As always, other material risks and uncertainties should be
communicated clearly. Report users needing to adjust mtitiple
variables may need to have special ad hoc cases run. For
example, they have forecasts of production and prices much
different than used in the resewes valuation report.
Recommendation 10. When next reexamining the guidelines,
lets open the discussions a bit further. Why not allow
forecasts about technology improvements? We are cotildent
of continuing improvements in drilling, reservoir modeling,
wismic prussing, and other technologies of the business. If
material to the evaluatio~ why not also include a forecast
technology improvement? I recommend that everything
important to the evaluation be included in the forecast model.
More Work for the Near Term
Some readers no doubt find my recommendations somewhat
mdical. Dont fix it if it aint broke. Other readers may be
interested injoining the evolution track.
Petroleum appmisal work is now typically 90% petroleum
engineering and only 10O/. evaluation economics. If you
implement most of the recommended changes, more work will
be required initially to perform reserves assessments. Perhaps
the mix. needs to & 70V0engineering and 30% economics.
Assume the same amount of engineering. More attention is
needed on the economics. To implement the recommenda-
tions, above, my calculations suggest that we initially need
about 3.8 times as much time devoted to the economic evalua-
tion and re~rves consolidation. This is a 38% temporary
increase in overall assessment effort. Doubtless, better tools
will evolve to ~e the extra human work involved.
Extended probabilistic reserves assessments do not require
more time-and may actually save time-once the education
and tools are in place.
Nomenclature
EMV =
EV =
NPv
pv .
p.xx =
s=
p.
~.
References
1.
2.
3.
4.
5.
6.
7.
8.
expected monetary value = EV NPV, also
catled risk-weighted NPV, same units as
NPv
expected vaIue, probability-weighted aver-
age, synonymous with mean
net present value, same as net present
worth; in $ or other currency units
present value, same as present wo~ and
usually the same as NPV (some people use
net to distinguish that the investment has
been deducted); in $ or other currency units
cotildence with XXV. probability of
realizing the reference quantity or more
sample set standard deviation
mean of a probability density distribution
or population
standard deviation of a probability density
distribution or population
Behn, R.D., and Vaupel,J.W.: QuickAnalysisfor Busv Decision
Makers, B&ic Books:NewYor~(1982). - - -
Cronquis4 C.: Preserves and Probabilities-Synergism or
Anachronism?,JPT (1991) 1258.
Garb, F.A.:AssessingRisk in EstimatingHydrocarbonReserves
and in EvaluatingHydrocarbon-ProducingProperties,JPT (Jun
1998)765.
Garb, F.A.: Oiland GasR~es Classification,Estimatio~and
Evaluation,JPT (Mar 1985) 373.
SPE: SPEConsidering Changes for Reserve Definitions, JPT
(1991)708.
Society of Petroleum Engineers and World Petroleum
Congresses Petroleum Reserves Definition (March 1997).
Available from SPE Book Orders and posted ordine at
http;llwww.spe, org.
Schuyler, J.R: Decision Anrdysis in Projects: Stochastic
Variance,PA4Nemork (July 1995) 11.
Schuyler. J.R: Decision Analvsis
tigernent Institute, UP Darby,
Chapter 11.
in Projects, Project
Pennsylvania (1997)
184
SPE 49032 PROBABILISTIC RESERVES LEAD TO MORE ACCURATE ASSESSMENTS 7
9. Newendoq, P.D., and @lck, A.N.: The Need for More Reliable
kision Crit~ for Drilling Prospect Analyses, paw SPE
16318 prewted at the 1987 SPE Hydrocarbon Economics and
Evrduation Symposium, Wflas, March 2-3.
10. Hit- T. S.: The Evolution of Economic Forecasts and Risk
Adjustments in Property Evaluation in tie U. S., JPT (Feb. 1991)
223.
1I. Brdey, R.A. and Myers, S.C.: Pnnciples of Co~omte FinOnce,
5th d,Mctiw-Hill(1996).
12. Schuyler, John R.: Rational Is Practical: Better Evaluations
through the Logic of Shareholder Value; paper SPE 030066
Figures

o 10 20 30 40
Recoverable Gas 106 m3
Fig. lExample resetvoir distributions and their sum,
Distributions A and B are Iognomal with a 0.50 rank
correlation coefficient.
1.00+ .
0. 90-
G
0. s0 -
0.70-
0.60-
0.60-
0.40-
0.30-
presented at the 1995 SPE Hydrocarbon Economics and
Evaluation Symposium, Dallas, March 27-28.
Customary Units Conversion Factors
m3x6.2897 E+OO= bbl
. .
msx 3.5315 E+O1 = fi~
:
..
0510152025303540
Oil Reserves 106 m3
Fig 3-Reverse cumulative distribution showing the
reserves distribution of n=l 9 prospects, each with
p=O.40 chance of success and success cases having an
. .
-ail volume distribution of lognomal(2,2.5) x 10 ms.
0.20-
0.1o- 4
0.00 +
b
I I ----
0 2 4 6 8 10 12 14 16
No. Discoveries
Fig 2Reverse cumulative distribution showing the
pr~bability of the x-axis remaining discoveries 6r greater.
This is a binomial distribution with n=19 events, each
with a p=O.40 chance of success on any trial.
185