1. In this article, the terms risk and uncertainty are used inthe same way as most of those in the oil and gas indus-try use them, according to the Aberdeen study discussedlater in this article. “Risk” means the chance, likelihoodor probability of something happening, and “uncertainty”refers to the range of possible values or sizes of thatsomething, if it happens. An alternative set of definitions,which is perhaps better and more rigorous, but not yet in common use in the industry, would include threeterms: chance, uncertainty and risk. “Chance” refers to the likelihood or probability of something occurring,“uncertainty” refers to the range of possible outcomes(given that something occurs at all), and “risk” refers to the threat of loss contained in a chancy businessventure with considerable uncertainty as to range ofpossible outcomes.

2. Simpson GS, Lamb FE, Finch JH and Dinnie NC: “TheApplication of Probabilistic and Qualitative Methods toAsset Management Decision Making,” paper SPE 59455,presented at the SPE Asia Pacific Conference onIntegrated Modeling for Asset Management, Yokohama,Japan, April 25–26, 2000.

20 Oilfield Review

Taking a Calculated Risk

William BaileyAberdeen, Scotland

Benoît CouëtRidgefield, Connecticut, USA

Fiona LambGraeme SimpsonUniversity of Aberdeen Aberdeen, Scotland

Peter RoseRose & Associates Austin, Texas, USA

For help in preparation of this article, thanks to Ben Ball,Massachusetts Institute of Technology, Cambridge, USA;Kent Burkholder, Merak, London, England; Keith Leslie,McKinsey & Co., London, England; Steve McColl, Conoco,Aberdeen, Scotland; Patrick McIntosh, Det Norske Veritas, Aberdeen, Scotland; David Morgan, UncertaintyManagement, Hertford Heath, Hertfordshire, England; Bill Pace, Imperial College of Science, Technology andMedicine, London, England; Sam Savage, StanfordUniversity, California, USA; Michael Walls, Colorado Schoolof Mines, Golden, Colorado, USA; and M.W. Whiteside,Indeva Energy Consultants, Henley-on-Thames, England.

Engineers, mathematicians and other experts have devised many tools to help us

understand uncertainty and to evaluate and mitigate risk. The oil industry is perme-

ated by uncertainty and encounters risk at every turn, yet many oilfield decision-

makers, perhaps most, give the new techniques a wide berth.

The oil industry is riddled with risk and uncertainty.Both loom so large at almost every stage of thebusiness—exploration, production and down-stream marketing—that the industry is regarded asa classic illustration of the need for sophisticatedapproaches to risk assessment. Yet the evidencesuggests that although many rigorous assessmenttools are available, they are underutilized. Even thelargest companies often use intuition and experi-ence rather than science when asked to appraiseinvestment opportunities and decide whether tocommit funds to particular projects.

Properly assessing risk and uncertainty con-fers a competitive advantage. Research done atthe University of Aberdeen, Scotland, into thedecision-making practices of 20 companies activein the North Sea has established a strong positivecorrelation between the degree of sophisticationin companies’ decision analysis and the successof their investment decisions. The research alsoshows that there are glaring gaps in the use ofavailable tools. Tools that can be, and are, used tocope with physical risk and uncertainty are practi-cally ignored when economic risks and uncertain-ties are in question.1 Tools involving probabilisticanalysis are used to capture the uncertaintyinvolved in, say, determining the recoverable

reserves in a field, but not to assess the eco-nomics of developing a field under conditions inwhich costs and oil prices vary.2

Many tools are available to help companiesmaintain a competitive advantage through prop-erly assessing risk and taking the appropriateamount of risk (see “Risk, or Chance of Success,Estimation,” page 22 ). These include, in roughlyascending order of sophistication: discountedcash flow, Monte Carlo analysis, and portfolio,options and preference theory. This article dis-cusses each technique in detail and presentscase studies to demonstrate their use in assess-ing risk in the oil and gas industry.

Autumn 2000 21

Intro Art to Fill this Page

The first step in any rational analysis of anopportunity is a subjective estimation of thechance of a least some minimal measure of suc-cess—for example, the chance of finding oil andgas, as opposed to drilling a dry hole. Chance ofsuccess is binary, and may be likened to an on-off switch. If the chance of something happen-ing is estimated to be X%, then the chance ofthat thing not happening is 100% minus X%.Generally, chances of success estimates are bro-ken into two categories: below-ground andabove-ground estimates. “Below-ground” chanceestimates in exploration and production (E&P)tend to be the concern of geoscientists andengineers considering geological evidence bear-ing on the likelihood of hydrocarbon presence,

reservoir and trap presence and other technicalinput. “Above-ground” chance estimates mayfocus on elements of politics, world economicsand technological developments, which are thenatural purview of experts in governmentalaffairs, finance and technology.

Characteristically, experts make all estimatesof chance, often working collectively, and con-sidering known facts, past experience and allpossible scenarios. Surprisingly, explorationiststend to be conservative when estimating chanceof success for “midrange” projects—thosethought to have a moderate 25% to 60% chanceof success. Such projects are often successfulabout 35% to 75% of the time.1 However, for“high-risk” projects—those thought to have less

than 20% chance of success—explorationistshave historically proven overly optimistic. Takenas whole, such ventures have found oil less than5% of the time.

Risk, or Chance of Success, Estimation

1. Alexander JA and Lohr JR: “Risk Analysis: LessonsLearned,” paper SPE 49039, presented at the SPE AnnualTechnical Conference and Exhibition, New Orleans,Louisiana, USA, September 27-30, 1998. Otis RM and Schneidermann N: “A Process forEvaluating Exploration Prospects,” AAPG Bulletin 81, no. 6 (July 1997): 1087-1109.McMaster GE and Carragher PD: “Risk Analysis andPortfolio Analysis: The Key to Exploration Success,”Proceedings of the 13th Petroleum Conference—Exploration, vol 2. Cairo, Egypt: The Egyptian GeneralPetroleum Corporation (1996): 415-423.

Discounted Cash FlowDiscounted cash flow (DCF) analysis, the mostwidely used investment-appraisal tool in the oilindustry, embodies a concept that is crucial to anindustry whose investment time scales are oftenmeasured in decades rather than years—namelythe time value of money. The time value ofmoney is based on the idea that an amount ofmoney received at some point in the future isworth less than the same amount receivedtoday. In the North Sea, there is an average gapof seven years between initial explorationexpenditure and the commitment to develop adiscovery. It takes another three or four years to begin production, and then fields normallyproduce for around 20 years before they are

abandoned. Most of the main costs, or cash out-flows, are incurred in the earlier exploration anddevelopment years, while the cash inflows, orrevenues, are spread over the active productivelifetime of the field.

Cash received later—in this case, the rev-enues received from the produced oil—is worthless than the same sum paid at an earlier datebecause it has not been available to earn interestin the intervening years.

DCF analysis is a way of determining thevalue today of money spent and—assumingsuccess—received in future years. The associ-ated concept of net present value (NPV) enablesthose who are evaluating potential investmentsto determine whether an investment should

proceed or not. The net present value is the sumof the discounted cash flows and represents thedifference between the present (discounted)values of the cash outflows over the projectedlife of the project and the present values of thecash inflows.

If the NPV is positive, the required rate ofreturn is likely to be earned, and the projectshould be considered. If it is negative, the projectshould be rejected. A key element in the calcula-tion of NPV is the discount rate applied. This canbe considered in several ways. For example,there is the risk-free rate of return that a bankwould offer for depositing money. If using thatrate in the calculation yields a negative NPV, thenit would be better to put the money in the bank.A positive NPV means investing the money in theproject is better than putting it in the bank. Analternative is to ask what it costs to borrow themoney, either from shareholders or the bank,then discount at that rate.

An example of discounted cash flow analysiscan be shown in tabular form (left). Using a 10% discount rate, the value of a $2,000 net cashflow ($2,500 of revenues less $500 of operatingexpenses) that is received in year 5, as a result ofinvesting $5,000 today, is worth $1,242. The totalNPV in this example (the sum of all the discounted

22 Oilfield Review

Year

012345Total

$5,000

$5,000

$2,000$2,000$2,000$2,000$2,000$5,000

$1,818$1,653$1,503$1,366$1,242$2,582

$1,667$1,389$1,157$965$804$982

$2,500$2,500$2,500$2,500$2,500$12,500

$500$500$500$500$500$2,500

Investment Revenue Net cashflow

10% discountednet cash flow

20% discountednet cash flow

Operatingexpense

–$5,000 –$5,000 –$5,000

> Discounted cash flow. This example shows the net present value (NPV) growth of $5,000 investedusing a 10% discount rate. [Adapted from Jones DR: “Some Basic Concepts,” in Steinmetz R (ed): The Business of Petroleum Exploration. Tulsa, Oklahoma, USA: American Association of PetroleumGeologists (1992): 9.]

Autumn 2000 23

net cash flows) is $2,582. In other words, the$5,000 is recovered, plus a 10% return, plus$2,582. If the $5,000 had been invested in a bankat 10% interest, the return would be $2,582 lessthan an investment in this project.

The usefulness of DCF is limited by its insen-sitivity to the changing circumstances and longtime scales in the oil industry. To surmount thisshortcoming, DCF is often used in conjunctionwith a technique known as sensitivity analysis, inwhich the consequences of possible changes tothe variables are examined. Changes to interestrates, cash flows and timing are fed into the cal-culation to determine the value of the project ifsuch changes actually occur. Used together withDCF, sensitivity analysis allows for a limited num-ber of “what if” scenarios, but the choice ofwhich variables to alter and how to alter them ishighly subjective.

While DCF combined with sensitivity analysismay give decision-makers a better idea of thepotential positive and negative outcomes of aninvestment, it makes no attempt to quantify the probability of any given outcome, informa-tion that would be extremely valuable to the decision-maker.

Monte Carlo SimulationMonte Carlo simulation brings risk and uncer-tainty center stage as integral parts of the calcu-lations rather than as afterthoughts. Mostimportantly, it brings probability into the picture.This statistical technique addresses the ques-tion, “If something happens, then what is therange of possible outcomes.” It yields probabilityversus value relationships for key parameters. Itcan be used to answer technical questions—What is the range of recoverable reserves ofhydrocarbons in this acreage?—and economicones—What is the probability that the NPV ofthis prospect will exceed the target of $X million?

It is easiest to see how Monte Carlo simulationworks by examining the comparatively straightfor-ward task of determining recoverable reservesfrom an underground prospect (above left).

If reservoirs were homogeneous, it would besimple to work out the recoverable reserves in aprospect using a unique value for each parame-ter. But, in practice, it usually is not possible toassign such single values to each parameter.Geologists and engineers have to estimate field-wide average values for properties like porosityand gross rock volume (GRV) on the basis ofincomplete information.

What they can do with the limited data theyhave, however, is to draw a distribution—a curvethat describes the probability of a particularvalue occurring—for each input variable. If, forexample, the range of possible porosities forsandstone is typically between 10 and 35%, thedistribution curve relating probability on the

vertical axis to the porosity value on the horizontalone would describe the probability of each poros-ity value occurring.

Similar distribution curves can be drawn forall the other inputs. In a Monte Carlo simulation,each of these inputs is then randomly sampledand the individual values multiplied together (aprocedure known as a “trial”). The result of a sin-gle trial provides one possible answer for recov-erable reserves. This random sampling of each ofthe input distributions is repeated many times—typically between 1000 and 100,000 dependingon the type of calculation being performed. Withso many trials, the simulation will sample themost likely outcomes of each distribution morethan the extremes because there are more exam-ples in that range. The end result is a new distri-bution curve—a range of possible recoverablereserve sizes and the probability of any particularvalue occurring.

In an ideal world, the individual distributioncurves would be based on many measurements.But, in practice, the data available are often min-imal. Discipline experts who bring their experi-ence to bear will suggest the shape of the curvethat is consistent with the small amount of dataavailable. For instance, geologists often drawanalogies between the porosity of the rocksbeing examined and the porosity of rocks from asimilar previously exploited area.

The shape of distributions can vary enormously(above right). A triangular distribution, for instance,might be chosen for porosity if the experts wereconfident that they knew the minimum, most likely,

Nr = recoverable reservesGRV = gross rock volumef = net pay/gross pay ratioφ = porositySh = hydrocarbon saturationεr = recovery efficiencyB = shrinkage or expansion factor

Recoverable reserves

Nr= (GRV) f φ Sh εr B

> Formula for estimating recoverable hydrocar-bons. Gross rock volume is the total volume ofthe “container” mapped out by the geoscientists.Net pay/gross pay is the proportion of the con-tainer that is reservoir rock (for example, sand)as opposed to nonreservoir rock (shale). Porosityis a measure of the fluid storage space, or pores,in the reservoir rock, as opposed to sand grains.Hydrocarbon saturation is the proportion of fluidin the pore spaces that is hydrocarbon as opposedto water. Recovery efficiency is the proportionof producible hydrocarbons in the reservoir. Theshrinkage or expansion factor accounts for thevolume shrinkage or expansion of hydrocarbonsas they travel to the surface. For hydrocarbonsthat are liquid in the reservoir, the release ofpressure resulting from transport to the surfaceallows gases to come out of solution, and thevolume of liquid shrinks. For gas, the situation isreversed; a reduction in pressure causes gasexpansion, so the volumes of gas at the surfaceexceed the reservoir volume.

Normal Triangular BinomialPoisson

Lognormal Uniform GeometricExponential

Weibull Beta CustomHypergeometric

> The various shapes of distributions. The best known is the ‘normal’ curve, whose shape was firstrecognized, in the 17th Century, by the English mathematician de Moivre. The curve is bell-shaped andsymmetrical. Its mean, mode and median are all in the center. The normal distribution describes manynatural phenomena, such as people’s IQs or heights. A triangular distribution describes a situation inwhich the minimum, maximum and most likely values to occur are all known. In a uniform distribution,the rectangular shape indicates that all values between the minimum and maximum are equally likelyto occur. The skill of the geologist or engineer lies in deciding which curve best describes the situationbeing examined, like the spread of possible porosities of a reservoir rock.

and maximum porosities. A lognormal distributionmight seem most appropriate for GRV, indicatingthat the experts think that the range is greater onthe high side than on the low side.

Monte Carlo simulation is widely used forestimating reserves, yet only a small number ofcompanies use it as an aid to economic decision-making, or to assess political and safety risks,though the principles are the same (see“Unconventional Risks,” next page). This suggestsan unusual perception of risk—that it exists and isimportant in the physical world, but is somehowabsent in the economic one. That clearly is notthe case as the gyrations of oil price, costs, inter-est rates and many other financial factors haveshown over the years.

The following example considers a hypo-thetical field with recoverable reserves of 150 million barrels [2.4 million m3] of oil (MBO).Yearly production immediately reaches a plateauof 12% of total reserves, that is, 18 MBO/yr [2.8 million m3/a] for 5 years, then declines at20% per year thereafter, until all 150 MBO havebeen produced. Five production wells areneeded, at a cost of $15 million per well over twoyears. Platform and pipeline costs are $765 mil-lion over three years. Operating costs are $75 million per year, and abandonment expendi-ture after last production is $375 million.Corporation tax is 30%, inflation is 3.5% through-out the period, and the discount rate is 10%. Oilprice is assumed to be $18 per barrel, rising atthe rate of inflation.

A simple deterministic net present valuecalculation gives a nominal net present value,discounting the cash flow at 10% per year(NPV10) of $125 million. This is a positive number,so the decision to proceed with developmentshould be straightforward.

But a probabilistic assessment of the samefield gives the decision-maker a broader picture toconsider. Assume the probabilistic assessmentuses the above figures as the most likely inputs(those falling at the midpoint of the range) but alsosuggests the following ranges of possible valuesfor inputs: drilling, capital and operating expendi-tures are assumed to be normally distributed witha standard deviation (SD) of 10% about the mean.Abandonment expenditure is normally distributed,with a SD of 20% of mean. Production volumesare also normally distributed, but with a positivecorrelation to operating expenditure.

24 Oilfield Review

0.024

0.018Pr

obab

ility Frequency

0.012

0.006

0.000-100 13 125 238 350

244

183

122

61

-100 13 125 238 350

0

Frequency

10,000

0

1.00

Prob

abili

ty

0.75

0.50

0.25

0

Forecast NPV, $ million

Forecast NPV, $ million

> Monte Carlo simulation results. A Monte Carlo simulation, named after thecasino in Monte Carlo where systems for winning at various games of chanceare often tried, shows the entire range of possible outcomes, such as netpresent values (NPV) of an asset shown on the X-axis and the likelihood, or probability, of achieving each of them (top) shown on the Y-axis. The frequency of each outcome for 10,000 trials is also shown on the Y-axis.The simulation does not give a single answer, but a range. It provides the deci-sion-maker with the “big picture.” Several examples taken from the forecastprobability distribution and frequency plot are shown in the table (middle).The reverse-cumulative probability distribution (bottom) shows the likelihoodof obtaining an NPV greater than a value on the X-axis.

01025507590

100

–1122771

122176223422

Percentile, % Value, $ million

In the oil industry, risk and uncertaintymodels usually deal with wellbores andreservoirs. But similar models also can be used to explore the potential impact of more unconventional risks—politicalrisks, terrorist threats, legal decision-making, environmental, health and safety regulations, and many others.

The approaches to modeling this sort of uncertainty use many of the mathematicaltechniques common to more traditionalfinancial or physical risk analysis. However,many additional intangibles have to bedefined before it is possible to properly framethe questions to be addressed by the riskmodel. Probabilities still have to be assigned,and, just as with physical and economic risks,a team of experts may be needed to developappropriate distributions.

For example, in evaluating the politicalstability of a country in which a companywanted to operate, the risk team could setup probability distributions for possible gov-ernmental vulnerability, possible internaldisorder, ethnic or religious problems, demo-graphic pressures, or even possible war. Witha proper combination and weighting of thevariables, a Monte Carlo simulation couldprovide a cumulative probability plot of, forexample, the total political risk in a country.This, in turn, could be compared with that ofother countries to help the corporation makethe appropriate strategic decision. A corre-sponding quantitative sensitivity analysiscould also shed some light on the relativeimportance associated with the diverse risks.

In modeling unconventional risks, themodeler is attempting to quantify humanactivities and emotions, so the model can be only a rough guide. However, such modelscan generate data essential to the overallprocess of making an enlightened decision.

Unconventional Risks

Autumn 2000 25

Future oil price over the period of interest isthought to be best described as lognormal, withSD of 10% in the first year of production, rising by2% per year, and reaching 34% by the last year ofproduction. This gives a roughly constant low priceat about $10/barrel, with the high price rising from$23/barrel to $37.5/barrel through field life.3

Results from 10,000 Monte Carlo trials showthe probability of a range of possible outcomes(previous page). The mean, or average, expectedvalue is $124 million. That means a statisticallysignificant number of identical opportunitieswould, on average, be worth $124 million each, in NPV terms.

But there is also a range of possible outcomesand a chance of very different results. For exam-ple, 10% of the cases run in the simulation gavevalues less than $27 million. Thus, the so-calledP10 value of the outcome, or the value at whichthere is a 10% chance of the outcome being lessthan (or 90% chance of exceeding) is $27 millionin this example. The lowest value given by any ofthe trials is –$112 million, and approximately 5%of the trials gave negative NPVs. On the otherhand, the P90 was $223 million, so 10% of the tri-als gave values greater than $223 million.

For this particular field, there is a smallchance, about 5%, of losing money, but a goodchance of making a significant amount of money(for example, a 16% chance of making more than$200 million). The decision would probably stillbe to go ahead, but the Monte Carlo analysis, byrevealing the broader picture, gives decision-makers greater comfort that their decision hastaken everything into account.

Monte Carlo analysis is a powerful tool, butmust be used with care (see “Monte CarloAnalysis of Interventions,” next page). An error inassessing just one input, such as the range in oilprice, can render the whole analysis defective.Thus, a Monte Carlo analysis of a North Sea fielddevelopment in the 1980s might give a pro-foundly misleading picture if it was predicated ona range of oil prices that varied at or around the$35/barrel price that prevailed at the start of thedecade. By the end of the 1980s, the price perbarrel was $15 and below.

Portfolio TheoryMost oil companies have many assets, such asfields, or shared interests in fields, and makeevery effort to assemble and hold the best possi-ble combination of such assets. Portfolio theoryshows how assets can be combined in such a way that risk is minimized for any given level of expected return. Alternatively, it may be seenas the study of the way in which the companycan achieve a maximum rate of return from aportfolio of assets, each of which has a givenlevel of risk attached to it.

The portfolio approach is based on the workof Harry Markowitz, who won the 1990 NobelPrize for Economics for his theories on the evalu-ation of risk and reward in stock markets.Markowitz sought to prove that a diversified port-folio of financial assets, one that mixed assets sothat return was maximized while risk was mini-mized, could be practical. Energy-risk analystssoon realized that there were parallels betweenthe stock market, in which paper assets—stocksand shares—are traded, and the oil business inwhich companies hold and trade portfolios ofreal assets by, for example, buying and sellingshares in joint ventures.

Portfolio theory can appear counterintuitive.4

Imagine being responsible for investing $10 mil-lion in exploration and production projects. Onlytwo projects are available, and each requires thefull $10 million for a 100% interest. One project isrelatively safe, and the other is relatively risky(above). The chances of success are independent.

Outcome NPV,$ million

Independentprobability, %

Dry holeSuccessDry holeSuccess

Safe

Risky

–1050

–1080

40606040

ENPVSafe = 60% x $50 + 40% x (–$10) = $26 million

ENPVRisky = 40% x $80 + 60% x (–$10) = $26 million

> Comparison of hypothetical safe and risky E&P ventures. (Adapted from Ball and Savage,reference 4.)

3. The discount rate chosen is again 10%.4. Ball BC and Savage SL: “Holistic vs. Hole–Istic E&P

Strategies,” Journal of Petroleum Technology 51, no. 9(September 1999): 74–84.

Oilfield Review

A coiled tubing intervention program was pro-posed for a well in a mature North Sea field toremove a plug, isolate a watered-out layer andperforate an additional new producing zone.Previous experience suggested that a six-day jobforecast for midwinter operations was overlyoptimistic because of the likelihood of badweather increasing nonproductive time (NPT).

A model was needed to establish whether theinitial projections provided were realistic and todetermine how long such a job could run beforeit became uneconomic. It was assumed that theviability of the job would be determined by bal-ancing the cost of doing it against the revenuegenerated from the additional oil produced,whether incremental (because access would begained to otherwise untapped reserves) oraccelerated (because accelerated productionwould provide a revenue stream earlier thanwould otherwise have been the case).

A model constructed to analyze the problemconsidered the following variables:• oil price and lifting costs • NPT due to weather and other operational-

related downtime. This adds to the costs.Fixed costs of products and services did not vary.

• expected additional oil production after a successful job

• possibility of not completing the jobsuccessfully

• probability of correct diagnosis of theproblem—including correct water influxlocation and mechanism

• discount factor. The model was used to compute the net value

of the intervention for 100 different job times.Each separate Monte Carlo simulation consistedof 5000 trials, resulting in 500,000 total separatetrials. Results indicate that if the time to com-plete the job was only 20 hours, then there is a50% likelihood (P50) that the net value to theclient would be £750,000 or more (the P90 isover £1 million). On the other hand, if the jobwere to take 100 hours, the model suggests thatthere would be only a 32% chance of profitability.

The analysis had several implications:1

• Reasonable working times for the job could bedefined up-front.

• Sensitivity to various parameters is obvious.• Prediction of additional oil had the greatest

impact.• NPT had the second greatest impact.

The analysis showed that the initial projec-tion of six days for the job was overly optimisticand that the job was likely to yield a net loss.The results were used to define an alternativewater shutoff proposal and a brief study to betterunderstand additional production potential.

Monte Carlo Analysis of Interventions

1. This model simplifies reality by assuming the indepen-dence of some of the variables. In more complex analy-ses, interdependences can be accommodated usingso-called Markov chain Monte Carlo (MCMC) methods.The Markov chain Monte Carlo analysis method properlyrepresents the interdependences of variables that wouldnormally be treated as independent or otherwise corre-lated to others using correlation coefficients duringMonte Carlo sampling. In MCMC calculations, the valueof one variable impacts the probability distributions of the other variables. There are classes of problems in the oil industry, such as evaluating electrical sub-mersible pump failures, that are thought to be solvableonly with MCMC methods.

26

The expected net present value (ENPV) ofeach, which is the NPV of the successful outcomemultiplied by the probability of that outcomeoccurring plus the NPV of the unsuccessful (dryhole) outcome and the probability of it occurring,is the same: $26 million.

Realistic complications now can be added. Ifmoney is lost, shareholder confidence tumbles.There is a 40% chance of losing shareholder con-fidence with the safe project and a 60% chancewith the risky one. The ENPV for both is $26 mil-lion, so there is no way of increasing that by

choosing the risky over the safe project. Underthe circumstances, the safe project is obviouslythe better choice.

Adding a further complication, suppose it is possible to split the investment evenlybetween the two projects. Intuitively it wouldseem a bad idea to take 50% out of the safeproject and put it into the risky one. But is intu-ition a good guide? There are now four possibleoutcomes (next page, top).

The expected NPV is still $26 million, but theonly way to lose money and thereby forfeit share-holder confidence is to hit two dry wells—scenario 4—for which the combined probabilityis 24% (combining 40% x 60%). That cuts the riskof forfeiting shareholder confidence by almost half,compared to investing 100% in the safe project.Moving money from a safe project to a risky oneactually reduces risk—a counterintuitive result,that represents the effect of diversification.

Diversification is clearly the thing to do; yetmany in the oil industry persist in doing some-thing else. They rank exploration projects by

Autumn 2000 27

expected present worth. While this method usescommon sense, it ignores the benefits of diversi-fication. In the above example, this would haveled to allocating all the funds to the safe project,with nearly twice the risk of the best portfolio.

The example makes one major assumption—that the projects are independent. Often they arenot. Their outcomes may be interrelated, more for-mally known as statistically dependent. For exam-ple, if both projects involve drilling wells in thesame hydrocarbon source area, a lack of hydrocar-bon generation in the area would doom bothendeavors. The simplest example of statisticaldependence is correlation, which may be eitherpositive or negative. The correlation is positivewhen a given outcome for one project increasesthe chances of an outcome in the same directionfor the other, thereby diminishing the effect ofdiversification. It is negative when a given outcomefor one project decreases the chance of an out-come in the same direction for the other, therebyenhancing the effect of diversification.

Applying this to the above example, a positivecorrelation on a 50-50 split between the safe andrisky alternatives would mean that if the safeoption succeeds, the risky one is also more likelyto succeed, and if the safe project fails, the riskyone is more likely to fail. There is still a 40%chance that the safe project will fail, but if itdoes, the chance that the risky one fails will begreater than 60%. So the probability of losingshareholder confidence is now more than 24%.

Following the same logic, if the correlation isnegative, the probability of forfeiting shareholderconfidence drops to below 24%.

The goal in portfolio management is to spreadthe investments across many opportunities whileseeking out negative correlations and avoidingpositive ones. Statistical dependence can comefrom a variety of sources, including, for example,place and price. The economic outcomes of twosites close to one another may be positively cor-related through geological similarities, such asdraining the same reservoir rock or relying on thesame hydrocarbon source or sealing rock. Twowidely separated sites, on the other hand, wouldhave little or no geological correlation and wouldtherefore be more diversified.

Crude-oil prices tend to be similar around theworld, so the economic outputs of oil projects arepositively correlated relative to fluctuations in

crude prices. By contrast, natural-gas prices indifferent locations tend to track neither crude-oilprices nor each other very well. This means aportfolio consisting of a gas project and an oilproject would tend to be less positively corre-lated and better diversified than a portfolio con-taining two oil projects.

A method for changing a suboptimal portfolioto a good one is explained in Markowitz’s theory,which is based on three precepts.5 First, the ratio-nal investor will choose more value over lessvalue given a constant level of risk, but will alsoprefer less risk to more risk given a constantvalue. Second, there is more than one optimalportfolio. Third, the portfolio as a whole is moreoptimal than its individual projects. Each projectmust be considered based on what it brings tothe portfolio as a whole.

Markowitz says that a portfolio is efficient ifthere isn’t another portfolio that has a greaterexpected return with no more risk, and there isn’tanother portfolio that has less risk with no lessexpected return. If one or both of these conditionsare false, the portfolio is inefficient. When allportfolios are plotted on a graph in which the ver-tical axis is value and the horizontal is risk, theefficient portfolios form a line called the efficientfrontier (below).

Moving up the frontier line results in anincrease in both risk and return. The portfolio rep-resented by Point A is inefficient because thereare portfolios with the same value but less risk—Point B—and portfolios with the same risk butmore value—Point C—as well as portfolios witha combination of these two conditions.

ENPV of portfolio = 24% x $65 + 36% x $20 + 16% x $35 + 24% x (–$10) = $26 million

Scenario

1

2

3

4

Success

Success

Dry hole

Dry hole

Success

Dry hole

Success

Dry hole

60 x 40 = 24

60 x 60 = 36

40 x 40 = 16

40 x 60 = 24

50% x $50 +50% x $80 = $65

50% x $50 +50% x (–$10) = $20

50% x (–$10) +50% x 80 = $35

50% x (–$10) +50% x (–$10) = –$10

Shareholders’confidenceretained

Shareholders’confidenceretained

Shareholders’confidenceretained

Shareholders’confidencelost

Safe Risky Probability, % Return, $ million Result

> Portfolio approach to hypothetical safe and risky ventures. Four scenarios show the possible outcomes from investing equally in two projects. (Adapted from Ball and Savage, reference 4.)

5. Markowitz HM: Portfolio Selections: Efficient Diversifica-tion of Investments, 2nd ed. Oxford, England: BlackwellPublishing Company, 1991.

Efficie

nt frontier

Risk

AB

C

Expe

cted

val

ue

> Making a portfolio efficient. The aim is toassemble and exploit the best possible collectionof assets. A portfolio is efficient if no other port-folio has a greater expected return with no morerisk, and no other portfolio has less risk with noless expected return. Portfolios, like that repre-sented by Point A, are inefficient. For the level ofrisk they involve, there is a possible combinationof assets that would result in a higher expectedvalue. (Adapted from McVean J: “Monte Carlo: An Alternative Approach to Efficient Frontier,”http://www.merak.com/news/documents/ef-0399.html.)

Actual constraints can be included in the opti-mization process so that the portfolios on theresulting efficient frontier represent realisticalternatives from which management canchoose, depending on the trade-off they wish tomake between higher risk with higher return andlower risk with lower return (see “A System forEvaluating Exploration Projects,” below ).

Options TheoryAn important aspect of decision-making is tim-ing, or deciding “when” to decide. Conditionsand input information can change over time,altering the outcome if decisions are made atdifferent points.

Many oil companies—three-quarters of theAberdeen sample—use decision trees as an aidto decision-making (next page). Decision treesillustrate the choices available, the uncertaintiesfaced by the decision-maker and the estimatedoutcomes of each possible decision. These treesclarify the choices, risks, objectives, monetarygains and information needs involved in invest-ment decisions.6 Putting in an estimated value foreach possible outcome and a judgment of theprobability of each of those outcomes occurringallows the calculation of an overall risked valueof the outcome of the decision.

Decision trees enable the decision-maker tochoose based on the financial outcome of theoptions. Options theory, more properly known asthe theory of real options, enables a value to beassigned to the option itself. Options theorybuilds on the idea that most projects consist notof “all or nothing” decisions but of a sequence ofchoices, many of which involve choosing amongoptions—for instance, between investing moneynow in a development or postponing the decisionon whether to invest in the development until fur-ther information becomes available.

The traditional way of evaluating investmentprojects in the oil industry, the discounted cashflow (DCF) analysis reviewed earlier, is based onthe unrealistic assumption that once an invest-ment is made, it runs its course without interven-tion. It also evaluates only the successfuloutcome. The possibility of abandoning it in theface of adverse circumstances or expanding it inresponse to unanticipated demand is not consid-ered. Options theory is more sophisticated thanDCF because it captures the flexibility inherent inmost projects. Options theory is both a tool, likeDCF, and a mind-set. As a tool, it helps peoplemake decisions. As a mind-set, it pushes people

into thinking about projects in a much moredynamic way, constantly looking for alternativesand better ways to run projects.

The theory of real options draws parallelsbetween the financial world of stocks and bondsand the world of real physical assets, whichmight be anything from factories to oil fields. Infinance it is possible, for a fee, to buy an option,which is the right (but not an obligation) to buy orsell a financial security such as a share at a spec-ified time in the future at a fixed price. An optionto buy is known as a “call” option and is usuallypurchased in the expectation that the price of thestock will rise. Thus a call option may allow itsholder to buy a share in Company ABC for $500on or before a set date. If the price of the stockrises above $500 on or before that date, theholder of the option can exercise it and pocketthe difference. A “put” option is bought in theexpectation of a falling price and protectsagainst such a fall.

Chevron has developed a process that enablesmanagement to compare a wide variety of globalexploration opportunities on a uniform and con-sistent basis.1 The process includes the integra-tion of geologic risk assessment, probabilisticdistribution of prospect hydrocarbon volumes,engineering development planning andprospect economics.

The process is based on the concept of playsand hydrocarbon systems. A play is a combina-tion of reservoir, source, seal and trap geometrythat has the potential to contain hydrocarbons.Geologic risk assessment, volumetric estimation,engineering support, economic evaluation andpostdrill feedback are all considered extensionsof the fundamental knowledge and understand-ing of the underlying geological, engineeringand fiscal constraints.

A foundation is set, describing the geologicframework and the prospect in terms of theplay—source, reservoir, trap and seal, and thetiming and dynamics of fluid migration. Theinformation from this description becomes thebasis for subsequent steps in the process. Riskassessment assigns a probability of success toeach of these four elements of the play, andmultiplication of these probabilities yields theprobability of geologic success. Chevron consid-ers a well a geologic success if a stabilized flowof hydrocarbons is obtained on test. Volumetricestimation expresses uncertainty in the form of a distribution of possible hydrocarbon vol-umes for the prospect. This is constructed fromranges of parameters obtained from informationspecific to the prospect, and data described by the parent play concept.

With this distribution, engineering supportprovides development scenarios for threecases—a pessimistic case (10%), the mean(50%), and an optimistic case (90%). Economicevaluation is run for each of the three cases,providing a range of economic consequences of the geological, engineering and fiscal frame-work. Commercial risk is based on the results of this evaluation, and overall probability of success is the multiplication of the proba-bility of geologic success and probability of commercial success. Postdrill feedbackdetermines whether predicted results are con-sistent with actual outcomes.

A System for Evaluating Exploration Projects

1. Otis RM and Schneidermann N: “A Process forEvaluating Exploration Prospects,” AAPG Bulletin 81, no. 6 (July 1997): 1087-1109.

6. Newendorp PD: Decision Analysis for PetroleumExploration. Tulsa, Oklahoma, USA: PennWell PublishingCompany, 1996.

7. Black F and Scholes M: “The Pricing of Options andCorporate Liabilities,”Journal of Political Economy 81(1973): 637-654.

28 Oilfield Review

Autumn 2000

Real options are analogous to financialoptions. For example, if the oil company decidesnot to develop an oil field today, it can do so in thefuture. By paying a fixed license fee to the govern-ment, the company buys a real option: the right torealize payoffs at any time during the license termby making a further investment to develop thefield, but with no obligation to do so—analogousto the exercise price of a call option.

The existence of alternative courses ofaction, such as to develop in the future ratherthan right away, is valuable. The flexibility givesthe project a value that cannot be reflected in astatic DCF analysis.

Because oil industry projects tend to consistof a sequence of discrete stages—seismic sur-veys, drilling, building the platform and layingpipelines, production, and ultimately, decommis-sioning—there are many decision points alongthe way. There may be several options to choosefrom and several opportunities to capitalize onthat flexibility (see “Framing the Problem,” right ).

In 1973, economists Fischer Black and MyronScholes published the so-called Black-Scholesformula for the valuation of financial options.7

Some theorists argue that adaptations of theBlack-Scholes formula and other more sophisti-cated formulas can be used to value realoptions—to make valuations that, unlike DCF,fully value flexibility. Using a valuation formula,the project can sometimes be shown to have avalue significantly greater than that shown byDCF analysis. Projects that would have beenrejected by managers using DCF analysisbecause they have a negative value can some-times be shown, by real options valuation, tohave a positive value that suggests the projectshould proceed.

Take, for example, an oil company that is try-ing to value its license in a block. Paying thelicense fee is equivalent to acquiring an option.The company now has the right to invest in the

Techniques like Monte Carlo simulation and options theory promise more accurateevaluation and better decision-making, butthey are useless if the ground has not beenproperly prepared. If decision-makers misssomething important in a contract or mis-understand an engineering point, then thewhole superstructure of sophisticated analy-sis may be built on a faulty foundation.

In Conoco, the first critical step to effec-tive decision-making is “framing.” Framinginvolves establishing a team of people who are versed in the disciplines needed to tackle the problem, then using a facilita-tor to draw from these people crucial piecesof information, such as:• what is known—facts and values• what is not known—risks and uncertainties• problems or difficult items• what decisions have already been made—

company policy.Framing enables the team to concentrate

on the key building blocks of the decisions to be made and the variables that will havethe greatest influence. Framing sessions areconducted as informally as possible. Theteam’s job is to make sense of the randomflow of information captured as notes duringthe more rigorous later phases.

Framing facilitators use a variety of tech-niques, like brainstorming, to stimulatediscussion. These generate decision hierar-chies, decision and risk timelines and strategytables. The final output of the framing ses-sion is an influence diagram, which formsthe basis of any economic or technical modelthat is used to further examine the problem.

The framing process is a model of decision-making itself: start by thinking as widely as possible, distill the information, siftthrough the options and make a decision.Framing tackles the first two elements andcan sometimes lead to the final decisionwithout further analysis.

Training sessions on framing establish a common language for Conoco employeeswhen talking about risk and help avoid themisinterpretations that may otherwise arise.

Framing the Problem

A

B

C

D

E

I J

F

G

H

Buyacreage

Don't buyacreage

Drill

Largefield

Dryhole

Marginalfield

Runseismicsurvey

Dropacreage

Drillsecondwildcat

Largefield

Marginalfield

Dry hole,drop

acreage

Dry hole,drop

acreage

Seismic surveyconfirms

lead

Seismic surveyshows nostructure

Largefield

Largefield

Marginalfield

Marginalfield

Dryhole

Dropacreage

Dropacreage

Drillsecondwildcat

Drill

Dropacreage

> Decision trees for decision-making in uncertain conditions. A decision tree sets out alternativecourses of action and the financial consequences of each alternative, and assigns a probability to thelikelihood of future events occurring. With this information, it is possible to determine the expectedvalue of each outcome. Decision-makers use decision trees to clarify the consequences that alternativecourses of action may have. Decisions are shown as branching points, or nodes, in the tree. Eachpossible outcome is shown as a branch. Decision trees can be simple with only a small number ofbranches and nodes, or complicated with many of them. (Adapted from Newendorp, reference 6: 117.)

29

block at the exercise price once uncertainty overthe value of the developed reserves—the stockprice—has been resolved.8

Assume that the company has the opportunityto acquire a five-year license and that the block isexpected to contain some 50 million barrels[8 million m3] of oil. The estimated present valueof oil from the field in which the block is locatedaverages $10 per barrel, and the cost of develop-ing the field (in present value terms) is $600 mil-lion. With static net present value calculations,the NPV will be:

$500 million – $600 million = –$100 million.The NPV is negative, so the company would

be unlikely to proceed. The NPV valuation ignoresthe fact that decisions can be made about theuncertainty, which, in this case, we consider to betwofold: there is uncertainty about the quantity ofthe oil in the block and about its price. It is possi-ble to make reasonable estimates of the quantityof the oil by analyzing geophysical and geologicaldata in similar areas, and there is also some his-torical data on the variability of oil prices.

Assume that these two sources of uncertaintyresult in a 30% standard deviation around thegrowth rate of the operating cash inflows.Assume also that holding the option obliges thecompany to incur the annual fixed costs of keep-ing the reserve active—say $15 million. This rep-resents a dividend-like payout of 3% (15/500) ofthe value of the asset.

Applying the Black-Scholes formula, but thistime valuing a real option rather than a stockoption, gives a real options value (ROV) of +$100million.9 The $200 million difference between theNPV valuation of negative $100 million and theROV valuation of positive $100 million representsthe value of the flexibility of being able to investif and when the uncertainties are resolved.

Calculations like this can have a powerfulinfluence on how corporate strategists regardtheir assets. One company collected a large port-folio of North Sea license blocks. When the NPVof blocks was positive, the company drilled anddeveloped them. When NPV suggested theywere uneconomic because costs were too high inrelation to revenues, development was notundertaken. The company eventually decided tosell the uneconomic blocks to companies thatfound them attractive.

Then the question arose whether the blockswere being valued in the right way. It was sug-gested that holding the license might be consid-ered an option to develop, if, in the future, newdrilling and production technologies wouldenable improved hydrocarbon recovery. A newfinancial model demonstrated how blocks couldbe priced at their option value over five years.The option value would recognize uncertainty inreserve size and oil prices, and would take intoaccount the flexibility of the situation.

The valuation exercise had a profound effecton the company’s management. They decided toretain the blocks that had a high option value andto sell or trade the rest at a figure that reflectedtheir revised worth.

Preference, or Utility, TheoryPeople may use computers or decision tools like discounted cash flow or Monte Carlo analy-sis to help them, but in the end the decision ismade by an individual or a group of people.Subjectivity complicates the decision-makingprocess, since the individual’s psychologicalmakeup can influence decisions. In the oil indus-try, risk plays an important part in the thinking ofexecutives. Understanding individual or grouppreferences and attitudes toward risk and risk-taking can be important.

In 1738, the mathematician Daniel Bernoullipublished an essay in which he noted awidespread preference for risk aversion.10 Nearly250 years later, Daniel Kahneman and AmosTversky gave a simple example of risk aversion.11

Imagine you are given a choice between twooptions. The first is a sure gain of $80, and thesecond a more risky prospect in which there is an85% chance of winning $100 and a 15% chanceof winning nothing. According to Kahneman andTversky, people prefer the certain gain to thegamble, despite the fact that the gamble has ahigher “monetary expectation”—the sum of theoutcomes weighted by their probabilities—thanthe certain outcome. With the certain outcome,one is assured of $80. With the riskier option, themonetary expectation would be $85 ($100 x 0.85plus $0 x 0.15). The choice is risk-averse if oneprefers the assured $80 to gambling on theriskier outcome (see “Risk Aversion,” page 32 ).

Mathematician John von Neumann andeconomist Oskar Morgenstern expanded prefer-ence theory with a number of axioms that areparaphrased in the following statement:

Decision-makers are generally risk-averseand dislike incurring a loss to a greater degreethan they enjoy making a profit of the sameamount. As a result, they will tend to accept agreater risk to avoid a loss than make a gain ofthe same amount. Also, they derive greater plea-sure from an increase in profit from a smallinvestment than they would from the same profitincrease from a larger investment.12

These statements can be expressed graphi-cally as a utility curve (next page, far right). Thisexample shows that the pleasure, or utility, asso-ciated with winning $4,000 is generally less thanthe displeasure of losing the same amount.People will take a greater chance to avoid a lossthan to make a gain of the same amount. Peoplealso tend to feel more pleasure about gaining$10 by going from $10 to $20, than they do aboutgaining $10 by increasing from $1,500 to $1,510.

Theoretically, it is possible to draw such a curvefor any individual or even a company. Curves withdifferent shapes would denote different types ofdecision-makers (next page, right). The shape ofthe curve in the lower left-hand quadrant describeshow the company feels about loss, and that in theupper right quadrant shows its attitude to risk andthe levels of profit associated with risk.

30 Oilfield Review

8. Leslie KJ and Michaels MP: “The Real Power of RealOptions,” McKinsey Quarterly no. 3 (1997): 4-22.

9. The value of a real option, P, is estimated by applyingthe Black-Scholes formula as follows:P = Se-δt x {N(d1)} - Xe-rt x {N(d2)},where d1 = {ln(S/X)+(r-δ+σ2/2)t}/(σ x √t),d2 = d1-σ x √t,and where S = stock price, X = exercise price, δ = dividends, r = risk-free interest rate, σ = uncertaintyof stock-price movements, t = time to expiration, andN(d) = cumulative normal distribution function.By analogy the value of a real option uses the sameformula, but this time S = present value of expected cash flows, X = present value of fixed costs, δ = thevalue lost over the duration of the option, r = risk-freeinterest rate, σ = uncertainty of expected cash flows,and t = time to expiration.Substituting the values in the example discussed in themain text givesROV = (500e-0.03 x 5) x {(0.58)}- (600e-0.05 x 5) x {(0.32)}= $251 million - $151 million = +$100 million.

10. Bernoulli D: “Specimen Theoriae Novae de MensuraSortis,” (Exposition of a New Theory on theMeasurement of Risk) 1738, Translated from the Latin bySommer L: Econometrica 22 (1954): 23-36.

11. Kahneman D and Tversky A: “The Psychology ofPreferences,” Scientific American 246, no. 1 (1982): 160-173.

12. Pace B: “Petroleum Economics Seminar,” lecture notes,Imperial College of Science, Technology and Medicine,London, England, 1998.

13. Simpson et al, reference 2.

Autumn 2000 31

It is possible, by analyzing the previous deci-sions of an individual or a company, to constructa utility curve that represents what the decision-maker thinks about risk, or rather how he or shereacts to its presence when making decisions.And such an instrument could, of course, be usedto help guide decision-makers farther down thehierarchy in making decisions that fit with thethinking of the top management or the companyas a whole.

In practice, few companies use preference the-ory to help in decision-making. Critics say thepractical problems are just too great. Within thesame organization, one manager may typicallychampion risky projects, while another in a similarposition may be more conservative. Perhaps,rather than being used as a tool to help individualsor companies make decisions, preference theorydoes have a more limited but still important role. Itcan graphically demonstrate to decision-makerswhat their style of decision-making implies.

The Value of Risk AssessmentIs it possible to quantify the value gained by usingthese tools? This was the objective of theAberdeen study mentioned earlier in this article.The Aberdeen study ranked the participating com-panies according to the level of sophistication oftheir decision-making (below). The levels includedthe risk-assessment tools described in this articleand others such as analysis, holistic view, risk anduncertainty definitions, and the combination ofqualitative and quantitative techniques.

“Analysis” refers to the use of some form ofcost-benefit analysis in investment appraisaldecision-making. All but one company used somestructured analysis. “Holistic” indicates whetheror not a company takes a holistic view of the totalcumulative net effect of the consequences of a

decision. For example, any upstream decisionmust include abandonment of facilities and thecost and timing implications of any environmentalprotection measures that need to be taken.

“Risk and uncertainty” indicates whether thecompany adopts rigorous definitions of risk and

uncertainty, and incorporates them in their anal-ysis. Risk, here, is defined as the probability of anevent occurring. Uncertainty is the range of pos-sible values for the size, cost and benefits of anevent, if it happens. The category “qualitativeand quantitative” indicates whether companieshave formal techniques to deal with qualitativeas well as quantitative input—things such ashabit, instinct and intuition.13

These criteria were arranged in ascendingorder of sophistication. Companies scored zero ifthey did not use a particular risk-assessmentmethod; they scored 1 if the method was partiallyimplemented; and scored 2 for full implemen-tation. The scores were summed to rank the level of sophistication of the companies. Theresearchers also ranked the companies by several business performance measures.

–$6,000 –$4,000 –$2,000

Plea

sure

Disp

leas

ure

Dollars lost Dollars gained

+$2,000 +$4,000 +$6,000A theoretical utility curve

> Building a utility curve. A typical curve mightpurport to describe how an individual felt aboutgaining or losing money—usually the pleasureassociated within winning a given amount is less than the displeasure associated with losingthe same amount. [Adapted from Rose PR:“Dealing with Risk and Uncertainty in Explo-ration: How Can We Improve?“ AAPG Bulletin71, no. 1 (1987): 1-16.]

Pref

eren

ce

ProfitLoss

+

–

> Utility curves representing different types of decision-makers. The utility curve for a risk-taker (top), for example, might be represented bya sharp rise in the upper right quadrant showingthat the attraction of big money overcomes thefact that there is obviously a disproportionaterisk. A different utility curve for a large company(bottom) that accepts losses with equanimity is shown by the straight line. However, the sharpdrop-off in the lower left quadrant makes it clearthat there is still a maximum allowable exposureto loss. (Adapted from Pace, reference 12.)

Pref

eren

ce

Maximum allowableloss on a single prospect

ProfitProfitLoss

+

–

Criteria

Company

A B C D E F G H I J K L M N O P Q R S T

Numerical analysis

DCF analysis

Holistic view

Monte Carlo

Risk/uncertainty

Portfolio theory

Options theory

Preference/utility

Qualitative/quantitative

> Ranking companies by their level of decision-making sophistication. Research done at the Universityof Aberdeen showed that the decision-making practices of 20 companies (labeled A to T) active in theNorth Sea correlate well with the success of their investment decisions. Companies scored highest (red)if they fully implemented the decision-making criteria shown in ascending order of sophistication in theleft column. If the criterion was partially implemented by the company, a green square is indicated in thechart. Uncolored squares indicate that the company did not use the particular risk-assessment method.

32 Oilfield Review

A recent study by the Colorado School of Mines(CSM) in Golden, Colorado, examined the risk-taking behavior of 40 of the top US-based oilcompanies over the 15-year period from 1981through 1995.1 The researchers’ starting pointwas the fact that when decision-makers arejudging a potential investment, they are awarenot just of the risks involved but of the amountof the company’s capital that is being exposedto the chance of loss. Economists usually haveassumed that the degree of risk aversiondecreases as wealth increases, and that as a company becomes wealthier, it is more pre-pared to take on larger, riskier projects. A smallcompany that is offered a risky project in whichthe upside is significant profit but the downsideis a loss that would swallow up a big chunk ofthe company’s capital may reject it. But a largercompany, for whom the downside might not rep-resent such a large proportion of its resources,might accept the project.

The CSM researchers used the concept of therisk tolerance ratio (RTR) to compare companies.The RTR is a ratio of the company’s observedrisk tolerance, RT, (a number derived by mathe-matics beyond the scope of this article) dividedby its predicted risk tolerance. An exampleshows how the RTR works (top right). For anyfirms i, the RTRi value is equal to RTi/RTi

’ whereRTi is the observed risk tolerance for firm i inperiod t and RTi

’ represents the predicted risktolerance of firm i as a function of size measuredby the standardized measure of discounted

future net cash flows (SMCF) for that same period. An RTR value greaterthan 1.0 implies a stronger propensityto take risk than do other firms ofequivalent size. AnRTR value less than1.0 implies a weakerpropensity to takerisk than firms ofequivalent size.

Risk toleranceratios were calcu-lated for the top 17 US oil companiesover the period 1983 to 1995 (below).Compare Exxon andShell in 1988. Exxonhad an RTR of 0.87 while Shell’s was 2.76. Thissuggests that Exxon was significantly less will-ing to take risks than firms of equivalent size,while Shell was an aggressive risk taker com-pared with other companies of similar size.

The study defined four categories of risktaking (above right). Firms in the high-risktolerance category (more than 2.5) show signifi-cantly higher return on assets (8.6%) than firmsthat are less willing to take risks. Firms in theother categories have exercised risk-averse

behavior and get much lower returns on theirassets. The CSM study suggests that, on average,companies in the exploration business tended to be too cautious with respect to risky projectsand as a result have realized lower return onassets than they might otherwise have done.

Risk Aversion

Operator 1 Operator 2 Operator 3

SMCFi (wealth) $1,000 million $100 million $10 million

RTi ' (predicted) $100 million $15 million $2 million

RTi (actual) $50 million $20 million $2 million

RTRi (RTi /RTi ') 0.50 1.33 1.0

> Estimating risk tolerance ratios (RTR). An RTR value greater than 1.0 implies a stronger propensity to take risk than firms of equivalent size. An RTR value lessthan 1.0 implies a weaker propensity to take risk than firms of equivalent size.

1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983

1.030.17NA

0.210.830.800.651.001.441.925.221.390.95NA

0.22NANA

Company

ExxonChevronTexacoAmocoMobilShellUSXArcoConocoPhillipsUnocalOccidentalAmeradaAnadarkoPennzoilKerr McGeeUniontex

0.830.311.810.491.061.330.464.632.411.37NA

2.612.181.570.502.98NA

1.580.461.470.711.741.150.423.132.821.811.922.582.791.290.67NANA

1.000.481.260.441.911.85

10.081.492.361.972.01NA

0.692.05NANANA

0.750.640.950.29NA

2.330.361.413.262.802.922.91NA

2.000.381.74NA

0.51NA

0.740.125.892.580.451.233.311.851.963.32NA

2.800.270.970.84

0.50NA

8.820.360.702.390.380.963.053.231.972.156.990.740.444.082.43

0.871.240.760.550.462.760.791.313.77NA

3.482.390.951.160.831.422.80

0.650.430.720.480.291.640.661.383.64NA

1.832.491.101.27NA

1.751.25

0.760.350.560.280.271.700.631.752.861.62NA

1.92NA

1.87NA

1.391.83

0.470.290.410.330.161.860.251.022.381.26NA

1.750.781.64NA

1.541.46

0.630.390.940.440.231.820.380.90NA

1.41NA

1.400.732.12NA

0.923.34

1.070.900.480.410.322.192.641.35NA

1.55NA

4.361.18NANA

2.144.41

1995 E&P assets, $ million

68,85227,91318,73415,24114,39311,97610,1099,1276,6494,8284,7194,5943,8732,2671,9921,7481,695

> Risk tolerance ratios for different companies between 1983 and 1995.

RTR group

RTRMaximumMinimumMeanSt. Dev.

High Moderate

1.5 to 2.5 24.2% –34.2% 5.2% 9.3%

Average

0.5 to 1.5 32.0% –37.0% 5.1% 8.7%

Low

< 0.5 20.9%–25.8% 5.6% 5.5%

> 2.5 28.1% –5.5% 8.6% 6.8%

> Performance analysis—return on E&P assets. Firms in the high-risk tolerancecategory show significantly higher return on assets (ROA) than firms that areless willing to take risks.

1. Walls M: “Corporate Risk Taking and Performance: A 15-Year Look at the Oil Industry, paper SPE 49181, pre-pared for presentation at the SPE Annual TechnicalConference and Exhibition, New Orleans, Louisiana, USA,September 27–30, 1998.

Autumn 2000 33

Five indicators of success were considered.First, market capitalization indicated the invest-ment community’s view of the future value of thecompany’s ability to make wise investment deci-sions. Second, the number of employees servedas a rough indication of both past success andanticipated future success in selecting and gain-ing access to the best investment opportunities.Third, the volume of booked reserves was used asa proxy for size and as an indicator of past success in investment decision-making. Fourth,return on equity indicated past successful deci-sion-making. Fifth, Wood Mackenzie’s estimate ofthe company’s total UK base value (value of com-mercial reserves + value of technical reserves +value of exploration) was used as an indicator ofsuccessful past investment decision-making.14

The outcome was a strong positive correla-tion between the ranking positions of companieson the decision-making scale and their positionsin the rankings for total base value, market capi-talization and proven reserves. The correlationwith number of employees was modest and thatbetween decision-making and return on equityweak. The return on equity figure did not surprisethe researchers. This measure relates strongly tohistorical decision-making—decisions made 15 or20 years ago—while in most companies the cur-rent practice in decision-making is recent, usuallyless than five years old. By contrast, the measuresof volume of booked reserves and, particularly,total base value captured the effects of recentdecision-making performance. The strong correla-tion between Wood Mackenzie’s total base valueand the level of sophistication rank list clearlydemonstrated a link between sophistication oftools used and business success (right).

In addition to these correlations, theresearchers found that while Monte Carlo analy-sis is used widely for prospect reserves—a clearrecognition of the importance of uncertainty atthis technical level—it is scarcely used at all forprospect economics.15 The researchers suggestthat this implies, in the cases of those who donot use it, an assumption of complete certainty incost, product price, fiscal terms and timingparameters, clearly an invalid assumption. Inaddition, it is mainly the larger companies thatuse portfolio theory. The researchers found that some smaller companies are less likely touse portfolio theory, because they feel they have insufficient properties to constitute a “port-folio” (though the theory applies equally to justtwo properties).

Since the Aberdeen findings were published,another group of risk experts sponsored by theNorwegian Petroleum Directorate and mostexploration and production (E&P) companiesactive in Norway has published a research paperwhich, among other things, suggests that theuse of probabilistic methods in the decision-making process is an important contributor tocompany performance.16

The researchers looked at different method-ologies to describe the maturing of projects andthe subsequent decision-making process. Theyfound that although most companies appear tobe technically capable of applying probabilisticmodels, only a few of them use these methodsroutinely in their decision-making process.

14. Simpson et al, reference 2.15. Simpson et al, reference 2.16. Jonkman RM, Bos CFM, Breunese JN, Morgan DTK,

Spencer JA and Søndenå: “Best Practices and Methodsin Hydrocarbon Resource Estimation, Production andEmissions Forecasting, Uncertainty Evaluation andDecision Making,” paper SPE 65114, presented at theSPE European Petroleum Conference, Paris, France,October 24–25, 2000.

Tota

l bas

e va

lue

rank

18

16

14

12

10

8

6

4

2

00 2 4 6 8 10 12 14 16 18

Level of sophistication rank

Correlation coefficient = 0.65

Without outliers = 0.85

Outlier

Outlier

Outlier

> The correlation between level of sophistication in the use of decision-making tools and total basevalue (TBV). TBV is a measure devised by Edinburgh-based energy analysts Wood Mackenzie, whichtakes into account proved, probable and possible reserves and makes an attempt to value explorationacreage. The Aberdeen researchers believe TBV is a particularly good measure since it captures theresults of decisions made in the recent past, and most sophisticated decision-making tools have beenin use for a comparatively short time.

Among the most important of the methodolo-gies studied was what the researchers calleddecision and risk analysis (D&RA), whichincludes elements of the various techniquesdescribed above. This was defined in theNorwegian study as a fully integrated, multi-disciplinary probabilistic approach based onranges for various parameters including fieldgeology, reservoir properties (such as porosity),steel costs, manpower costs, facilities downtimeand development scenarios. It also incorporatesthe propagation and aggregation of uncertaintythrough the various linked models and throughthe various decision levels.

Using an economic benchmarking study of themajor oil companies listed on the New York StockExchange, the researchers inferred a relationshipbetween company performance and workingpractices (above left). Companies that integratedtheir workflow and used D&RA saw their perfor-mance improve shortly after the introduction ofthis methodology.

The Norwegian study claims that a fully prob-abilistic, multidisciplinary workflow according toa D&RA process influences a company’s com-petitive position. There is also circumstantialevidence to support the contention that the morean E&P company integrates its workflow and the more probabilistic its approach in decision-making is, the better the company will perform.

34 Oilfield Review

StructuredQuantitativeDecisionAnalysis

Prospect description

Chance of success

Chance of achievingmeasure of success

Failure

Discounted cash flowMonte Carlo analysisPreference theoryPortfolio theoryOptions theory

Decision-making criteria

Cost of failure

Gut feel + rules of thumb

Individual habits

Experience Ambiguity

QualitativeDecisionAnalysis

QuantitativeDecisionAnalysis

> Understanding the different sides of decision-making. The lower half of the figure (blue) shows howquantitative means, like discounted cash flow and Monte Carlo analysis, are used to analyze risk andmake decisions. The upper half (pink) shows how qualitative means can be used for the same sort ofanalysis. There is often tension between the two (white arrows), as when executives follow theirintuition rather than the figures. Researchers are now investigating how decision-makers make theirdecisions, looking at the softer, qualitative aspects of the decision. Once that is understood, the nextstep will be to find the linkage between them.

17. Lamb FE, Simpson GS and Finch JH: “Methods forEvaluating the Worth of Reserves in the Upstream Oiland Gas Industry,” Geopolitics of Energy 22, no. 4 (April 1999): 2-7.

18. Capen EC: “The Difficulty of Assessing Uncertainty,”Journal of Petroleum Technology 28, no. 8 (1976): 843-850.

19. The 1997 Annual Report of the Norwegian PetroleumDirectorate. Stavanger, Norway: Norwegian PetroleumDirectorate Publications (1998).

20. Citron GP, Carragher PD, McMaster GM, Gardemal JMand Jacobsen D: “Post Appraisal and Archival: CriticalElements in Successful Exploration Risk Assessment,”paper presented at the first Landmark WorldwideTechnology Forum, Houston, Texas, USA, February12–14, 1997.

21. Smith P: “Managing Uncertainty in Oil FieldDevelopments: A Practical Guide to Making BetterDecisions,” paper presented at the Schlumberger Oiland Gas Decision and Risk Analysis Symposium, Austin,Texas, USA, November 20–21, 1997.

22. Simpson et al, reference 2.Carragher PD: ”Leveraging Learnings from ExplorationRisk,” AAPG 2000 Convention Abstracts, Annual Meetingand Exhibition of the AAPG, New Orleans, Louisiana,USA (April 16–19, 2000): A23–A24.Jonkman et al, reference 16.

19881

3

5

7

9

11

13

15

1990

Schr

oder

s Ba

nk ra

ting

Company A Company C

1992 1994 1996 1998

Year (5-year period ending in each of the years shown below)

> A benchmarking study of major oil companies listed on the New York StockExchange. Company performance, as rated by Schroders Bank, increasedafter decision and risk-analysis processes were introduced (white arrows).

Autumn 2000 35

The Roles of Intuition and Bias The processes described above do not constitutethe complete story. While structured quantitativeanalysis is part of the standard decision-makingprocess, qualitative intuition and judgment areextremely important (previous page, bottom). Thismodel represents a view of the complete deci-sion-making process, as seen by the Aberdeenresearchers.17 The interface between the quantita-tive and qualitative factors, and the relative pro-portions of each used in each decision aredescribed in terms of an analogy with a geologi-cal feature called an angular unconformity.

The vertical axis of the model represents thetype of decision being considered, with higherlevel decisions (for example, whether or not toenter a new basin or country, or to take overanother company) fitting in the top half of themodel, and more operational, routine decisionsfalling in the lower half.

The position along the horizontal axis relatesto company culture; some companies rely primar-ily on quantitative “hard” analysis with very little

“gut feel” input, and therefore would be posi-tioned on the left side of the model. Other com-panies are the reverse. This perhaps explainssome of the problems experienced when makingdecisions in partnerships and alliances when thevarious partners may occupy significantly differ-ent positions on this axis. This model of decision-making in the upstream oil and gas industry seemsto fit well with the experiences and practices ofthose working in the industry. However, no signifi-cant correlation has been found between a com-pany’s position in the model and its businesssuccess. It is a question of fit: each companyworks best where it is culturally most comfortable.

There is also a link between the structuredquantitative and the unstructured qualitativeinput: the decision-maker’s qualitative judgmentinfluences the numbers that go into the quantita-tive analysis. For any of the tools describedabove to be truly useful, the required geotechni-cal and financial inputs must be reliable. Butwhere do such numbers come from? To startwith, almost all of the geotechnical inputs and

many of the financial ones represent estimatesmade. Increasingly, the E&P industry is recogniz-ing the need to express such estimates as proba-bilistic ranges, rather than single “best-guess”values, but there is strong evidence of pervasiveand substantial bias in exploration and produc-tion project estimates:• The predictive ranges of key parameters are far

too narrow—uncertainty is underestimated.18

• Discovered fields typically contain only about40% of the oil and gas volumes that were pre-dicted prior to exploration drilling (left).19

• High-risk prospects fail about four times more often than predicted because risk isunderestimated.20

• Actual well costs often exceed forecast costsby 20 to 100%.

• Economic projections and yardsticks used tomeasure and rank ventures are often uncali-brated, being only infrequently compared withactual outcomes.

On worldwide projects, one major oil and gascompany reported the following actual versuspredicted measures: capital expenditure over-runs averaged 95%, with a maximum of 974%;operating expenditure overruns averaged 140%;first-oil production starting between one andthree years later than predicted; and averageproduction rates 65% of those predicted.21

Correcting predictive bias is an organizationalproblem, related to people, incentive systems,consistent procedures, corporate culture and lead-ership. It is not primarily a technology problem,although new technologies can help reduce bias.Another aspect of the bias problem deals with pre-ferred styles of decision-making. Executives, whomay have succeeded through their intuitive, sub-jective decision-making style may be understand-ably reluctant to begin relying on a probabilistic,systematic portfolio-management style.

Several recent independent studies tell thesame tale: integrated, probabilistic, systematic,centrally coordinated management of explorationand production portfolios produces better perfor-mance than traditional methods.22 Most of thesystematic, probabilistic companies work hard tominimize predictive bias, and they track past pre-dictive performance in order to improve it.Unbiased forecasting is a recognized objective.

The risk-analysis tools discussed have realand enormous potential for improving explorationand production performance. But the human partof the equation must be improved if this potentialis to be fully realized. —MB, RH

11 10 100 1000 10,000

10

100

10,000

1000

Size

of d

isco

very

, mill

ion

barre

ls o

f oil

Expected size before licensing, million barrels of oil

Predictive accuracy = 38%

> Recalibrating prospect forecasts. Operators in the Norwegian North Seaconsistently have been overly optimistic when forecasting the size of prospectsfor inclusion in applications for a concession. The X-axis shows the size of expected prospects operators reported to the Norwegian PetroleumDirectorate during the last ten years. The Y-axis shows the actual discoveriesrecorded. The central diagonal line (blue) represents a perfect calibration. Theupper diagonal line (yellow) represents forecasts underestimated by a factorof 10. The lower diagonal line (red) represents forecasts overestimated by a factor of 10. The vast majority of all the forecasts fall below the centraldiagonal line, which by definition represents bias in the estimations of theoperators. The sum of all the actual discovery sizes equals about 38% of the sum of the forecast discovery sizes. Only discovery data are included.(Adapted from Discoveries on the Norwegian Continental Shelf. Produced in collaboration with the Norwegian Petroleum Directorate: Stavanger,Norway, 1997.)