Judgment in Probabilistic Analysis
- D.C. Purvis (The Strickland Group, Inc.)
- Document ID
- Society of Petroleum Engineers
- SPE Hydrocarbon Economics and Evaluation Symposium, 5-8 April, Dallas, Texas
- Publication Date
- Document Type
- Conference Paper
- 2003. Society of Petroleum Engineers
- 4.1.2 Separation and Treating, 4.2 Pipelines, Flowlines and Risers, 5.7.4 Probabilistic Methods, 4.1.5 Processing Equipment, 5.6.9 Production Forecasting, 4.3.4 Scale, 5.7 Reserves Evaluation, 1.10.1 Drill string components and drilling tools (tubulars, jars, subs, stabilisers, reamers, etc), 5.1.7 Seismic Processing and Interpretation, 5.6.4 Drillstem/Well Testing
- 6 in the last 30 days
- 524 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 9.50|
|SPE Non-Member Price:||USD 28.00|
The chief use of probabilistic methods is to assess risk and opportunity, making them most applicable to situations of significant uncertainty. Hence, the cardinal sin unique to probabilistic analysis is to underestimate the range of outcomes. Unfortunately, the situations of the greatest uncertainty are also the ones where poor judgment is most likely to create unreliable results and dangerous decisions. The best judgment in probabilistic analysis is that which recognizes the full range of uncertainty by carefully framing the problem and by avoiding pitfalls which artificially reduce the range of results.
Probabilistic analysis has become an essential tool1 of the practicing reservoir engineer and reserve evaluator because its benefits are undisputed and well-documented.2,3,4,5 They include the following:
Probabilistic analysis forces the practitioner to think more completely, thoroughly, and thus clearly about the issues at hand.
Probabilistic analysis can reveal the drivers of value or risk more clearly, making it possible to focus on risk mitigation efforts, data acquisition, further analysis and upside potentials.
The results of a probabilistic study give a decision-maker more information about upside and downside uncertainty to inform his business decisions, future plans and portfolio analysis.
Probabilistic analysis communicates the uncertainty unambiguously (to those conversant in the terms of statistics).
The first two uses add the greatest value. It is only in the framing, interrogation and audit of the model that the user obtains these advantages. Moreover, those benefits must be actively pursued in the process in order to obtain a meaningful quantitative result. Without judicious implementation of the model, the quantified results may mislead and endanger the decision-maker with unmerited confidence.
Much effort has been spent on discussion of input distributions to the probabilistic analysis, i.e., the form and range of the uncertain variables. Unfortunately, these considerations are dwarfed in importance by the architecture of the model. Model architecture represents the way the model is set up, e.g., the type of calculation, the number of parts, the correlation of parts, and the rules in the model. The discussion below deals first with issues related to model architecture and second with issues related to the input distributions.
In designing the architecture of a probabilistic model, it is essential to identify those drivers with the greatest impact on uncertainty, to consider all possible sources of uncertainty, to select an appropriate calculation methodology and level of detail.
Input distributions are defined by range and form. Definition of these two parameters, however, is predicated upon proper understanding of biases and types of uncertainty. All of these are discussed below.
Poor model-building causes an excessively narrow distribution of results and higher estimates of "reasonably certain" values. For example, the three most commonly cited pitfalls of implementation (aggregation, correlation, and range of input variables) all tend to reduce the range of outcomes. Ironically, though the high confidence end of the distribution is used to define Proved reserves, the extremes of a resultant distribution of outcomes are more poorly defined and subject to change than central estimates. Identifying the judgment calls which impact the range of results makes it possible to appreciate the limitations and subjectivity of probabilistic analysis.
The single most important factor in the construction of a probabilistic model is the conceptual framework. This consideration more than any other affects the predicted range of outcomes.
|File Size||213 KB||Number of Pages||8|