Monte Carlo Simulation | Oil Authority

Monte Carlo simulation is a computational technique that models the uncertainty inherent in reservoir and economic parameters by running thousands of iterative calculations, each time drawing input values at random from defined probability distributions. Rather than producing a single deterministic answer, it generates a full probability distribution of outcomes, allowing engineers and decision-makers to quantify risk and characterize the range of possible results from P10 (optimistic) through P50 (median) to P90 (conservative). In oil and gas, it is the industry standard for volumetric resource estimation, EUR forecasting, and economic project evaluation.

Why Deterministic Estimates Fall Short

A conventional single-point volumetric calculation multiplies best-guess values for area, net pay thickness, porosity, water saturation, and formation volume factor to produce one number for STOIIP or GIIP. The problem is that each input carries its own uncertainty range, and multiplying best-guess values together produces a result that is almost always optimistic. When four or five uncertain parameters are compounded in a product, the most likely outcome of the combined expression is generally lower than the product of the individual most-likely values, a phenomenon known as the flaw of averages. Monte Carlo simulation resolves this by treating each input as a distribution rather than a fixed number and propagating uncertainty through the calculation mathematically.

Inputs and Probability Distributions

For a STOIIP calculation, the key inputs and their typical distribution shapes include: gross rock volume (often log-normal, reflecting asymmetric uncertainty about the structural closure), net-to-gross ratio (uniform or triangular between analogue well data bounds), porosity (log-normal or normal from core and log statistics), water saturation (triangular from Sw versus depth trends), and oil formation volume factor Bo (narrow normal distribution from PVT data). Geologic risk factors, the probability that the trap is filled, the seal is intact, and the reservoir is connected, are often applied as discrete multipliers outside the volumetric product but can be included as Bernoulli random variables within the simulation. Correlation between inputs must be specified carefully: thick reservoirs tend to have higher porosity in many formations, so treating those as independent variables introduces error. Modern tools such as Petrel, Crystal Ball, @RISK, and Python-based frameworks (scipy.stats, NumPy) support explicit correlation matrices between inputs.

Running the Simulation and Reading Results

A typical Monte Carlo run requires 5,000 to 50,000 iterations to produce stable percentile estimates. At each iteration the model draws one value from each input distribution, computes STOIIP or NPV, and stores the result. After all iterations the stored results are ranked to produce the cumulative probability distribution. P90 is the value exceeded 90 percent of the time, representing the low or proved case. P50 is the median, representing the most likely single estimate. P10 is the value exceeded 10 percent of the time, representing the high or optimistic case. The ratio P10 to P90 is a standard measure of spread and uncertainty; a ratio above 10 signals that the estimate is highly uncertain and additional data acquisition should be considered. Tornado charts and sensitivity plots identify which input variables drive the most variance in the output, directing further data collection effort to the highest-value parameters.

Reserve Classification and Regulatory Use

Both the SEC and SPE-PRMS reserve frameworks incorporate probabilistic language that maps directly to Monte Carlo outputs. Under SPE-PRMS, Proved reserves correspond to the P90 estimate of recoverable volumes, Probable reserves span P90 to P50, and Possible reserves span P50 to P10. The SEC 2009 rules allow probabilistic methods for reserve estimation provided the inputs and methodology are documented and auditable. Independent reserve evaluators routinely require the Monte Carlo model file, input distribution justifications sourced from analogue fields or local well data, and evidence that correlation structures were considered. For unconventional plays, Monte Carlo is applied to EUR distributions derived from type-curve fitting across hundreds of wells, with log-normal EUR distributions being the most commonly observed shape in tight oil and shale gas datasets.

Economic Modeling Applications

Beyond volumetrics, Monte Carlo is applied to full-cycle economic models to estimate NPV, IRR, and payout distributions. Uncertain inputs include well costs (subject to rig rate volatility and AFE overruns), commodity price paths (often modeled using price decks from strip pricing or stochastic mean-reversion models), royalty and working interest percentages, operating cost escalation rates, and facility capital timing. The output is a probability distribution of project NPV that allows management to evaluate risk-adjusted returns and compare projects on a consistent basis. Portfolio-level Monte Carlo, which aggregates multiple projects accounting for correlation between shared cost drivers such as steel prices or regional labor markets, is used to optimize capital allocation across a drilling program. Value-at-risk metrics derived from Monte Carlo are increasingly required by equity investors and lenders for project finance.

Key Takeaways

Monte Carlo simulation replaces single-point deterministic estimates with full probability distributions of STOIIP, EUR, NPV, and IRR by running thousands of iterations with randomly sampled inputs.
P90, P50, and P10 outputs map directly to SPE-PRMS Proved, Probable, and Possible reserve categories, making the technique central to reserve reporting and SEC filings.
Input distribution shape, correlation between parameters, and iteration count all significantly affect result quality; log-normal distributions are most common for volumetric and economic inputs.
Tornado charts derived from Monte Carlo runs identify which uncertain parameters drive the most variance, focusing data acquisition budgets on the highest-value measurements.
A P10/P90 ratio greater than 10 signals high uncertainty and typically justifies additional appraisal drilling or seismic before sanction.