By Jim Gouveia, Marc Bond, Jeff Brown, Mark Golborne, Bob Otis, Henry S. Pettingill, and Doug Weaver
Statement of the problem
Few industries are fraught with more uncertainty than prospect exploration in E&P. Our formal education guided us with the notion that unless we provide a precise answer, we have ‘failed’ to meet expectations. This is exasperated by investor and leadership’s need for certainty in their investment decisions. When we face an uncertain prediction, we need methods that decouple our minds from trying to jump to ‘the answer’ and instead capture a pragmatic range of possible outcomes.
The present value of our drilling prospects is primarily driven by their probability of realizing a commercial success, our corporate discount rate, commodity prices, capital expenses, operating expenses, and our share of the commodity’s cash flow after taxes and royalties. Each of these key parameters are riddled with uncertainty through a project’s lifetime.
Modelled ranges better inform decision makers by fairly representing the spectrum of possible outcomes. Many experts argue that decision makers simply require confidence in the mean outcome. Portfolio theory advises that given a great number of repeated trials and an unbiased estimation, our firms will deliver the aggregated mean outcome. Whilst portfolio theory is sound, it presupposes two realities which do not exist in the world of exploration. First, that our predictions are free of bias. Without a probabilistic basis, grounded by ‘reality checks,’ our forecasts have repeatably proven to be optimistically biased. Second, that there are enough repeated trials to make the aggregate prediction valid over time. No one (especially currently) is drilling enough exploration wells to support high statistical confidence in a program’s ability to deliver a mean outcome.
As decision makers we need standardized evaluation techniques upon which we can confidently make our best business investment decisions. That requires that subjective words and phrases such as ‘good chance,’ ‘most likely,’ ‘excellent,’ ‘low risk’ and ‘high confidence’ be eradicated from our presentation of E&P opportunities and replaced with probabilities which have a common definition across all disciplines and projects. The traditional industry consensus is the use of P10 and P90, which in the predominantly used ‘greater than’ convention, represents our optimistic but reasonable high side and our pessimistic but reasonable low side values respectively.
In a prior blog, we introduced our industry standard method of providing ’P90’ and ‘P10’ values to bracket the ranges of all possible prediction outcomes. Studies have consistently shown that we are not particularly good at making such predictions and tend to underestimate the uncertainty in what we are assessing. For most E&P parameters, this presents itself as having our P10 to P90 ranges too narrow and optimistically high. Until the Petroleum Resources Management System (PRMS) update in 2010, industry guidance for validation of a probabilistic distribution was for the user to compare their probabilistically derived P50 to their deterministic (based upon their best guess) P50. It should not come as a surprise to learn that early probabilistic methods were flawed, as they were based on the belief that in the face of all the inherent subsurface uncertainty, we as subsurface professionals (even those of us who were newly graduated) had an innate ability to directly estimate a P50. Unfortunately, this antiquated belief persists to this day.
So how do we better derive our probabilistic ranges? Let us first bear in mind that we are trying to pragmatically capture the full range of possible outcomes in our predictions. We know that many of our subsurface distributions are often best represented by normal or lognormal distributions. We also know that both normal and lognormal distributions go to a positive infinity, and that an infinite reserve or rate is neither possible nor pragmatic. On the low end, lognormal distributions approach zero and normal distributions go to negative infinity. At the high end, both lognormal and normal distribution go to infinity. As we are building a distribution to characterize a geological success, we can eliminate the known low and high ends of both distributions either by truncation of outcomes above and below certain thresholds, or our preference at R&A, spike bounding. In spike bounded distributions, randomly sampled values in excess of the selected high-end limit or low-end limit are respectively set as equal to the selected high- and low-end boundary values. As such the values are ‘bounded’ at the low and high ends of the distribution.
In the exploration realm, we are dealing with a relative dearth of data. Industry experience shows that in the “Exploration space” we can use the P1 and P99 values of our distributions as the pragmatic end members for our distribution. In practice, we estimate our P90 and P10 values. We then extrapolate these values to a P99 and P1 value. We try to assure that the P99 value represents the smallest meaningful result – the minimum geological success value – and the P1 value represents the largest geologically defensible result. Our P1 and P99 values are intended to represent a blending of geologic pragmatism and conceptualization. It is a trivial academic debate as to whether these high -end members are or should be our P2, P3 or P0.05 outcomes, the same logic applies to the low end P99. The use of P1 and P99 should be thought of as a pragmatic spike bounding of the end members (‘reality checks’) of our input distributions.
In summary, our input ranges must capture the entire range of possible outcomes, and industry experience has taught us that to effectivity capture that range, we should consistently employ specifically defined low-side and high-side inputs (e.g., P99, P90, P1 and P10)
In our next article in this series, we will work through an example which addresses Average Net Pay.