Monte Carlo simulation for decision making

Learn how Monte Carlo simulation turns uncertain inputs into a range of outcomes, so you can compare options and choose using clear decision criteria.

9 min read8 viewsMonte Carlo simulation for decision making
— · 0 votes
Contents
  1. What a Monte Carlo simulation does for a decision
  2. When a simulation adds value to a choice
  3. Turning estimates into probabilistic inputs
  4. Running the simulation step by step
  5. Reading simulated outcomes and expected value
  6. Setting decision criteria for risk under uncertainty
  7. Using simulation to rank a shortlist
  8. Limits and mistakes to watch for
  9. Monte Carlo decision analysis questions
  10. How does Monte Carlo simulation help with decision making?
  11. How many iterations does a Monte Carlo simulation need?
  12. Can I run a Monte Carlo simulation in a spreadsheet?
  13. What is the difference between expected value and the most likely outcome?
  14. Is Monte Carlo simulation worth it for small decisions?

Monte Carlo simulation for decision making replaces single guesses with ranges, then runs the decision many times. Each run draws random values for the uncertain inputs and records what happens to every option. The spread of simulated outcomes shows which choice holds up across likely futures, not only on average.

  • Describe each uncertain input as a range or distribution instead of one fixed number.
  • Run enough random samples that the results stop shifting when you add more runs.
  • Compare options on expected value and on the chance of a bad outcome.
  • Set your decision criteria before you look at the simulated results.
  • Treat the output as a guide whose quality depends on your input assumptions.

What a Monte Carlo simulation does for a decision

A Monte Carlo simulation tests a decision by replaying it many times with randomly drawn inputs. Each replay, called an iteration, uses one possible set of values for the things you cannot know. The model then calculates the result for that set, such as cost, profit or time. After all iterations, you have a full picture of outcomes rather than a single estimate.

A single-point estimate hides risk because it shows only one version of the future. Two options can share the same average result while one of them carries far more downside. Simulation exposes that difference by showing how widely the results spread around the middle. That spread is often what separates a comfortable choice from a risky one.

Monte Carlo simulation takes its name from the Monte Carlo casino, a nod to games of chance. The core idea is plain: let random sampling stand in for real-world uncertainty. You need no advanced math to read the output, only a clear question to answer. Spreadsheets with random functions and dedicated simulation tools can both run the calculations.

When a simulation adds value to a choice

Monte Carlo simulation adds value when a decision depends on several uncertain numbers at once. Project timelines, budgets, demand forecasts and investment returns all fit that description. Uncertainties combine in ways that are hard to judge by intuition alone. A simulation does the combining for you and shows the joint effect.

A simulation is less useful when the inputs are known or the stakes are small. Choosing between two listings with fixed, published prices rarely needs a probability model. In that case, a side-by-side comparison of the facts answers the question directly. Effort spent on modeling should match the size and reversibility of the decision.

Monte Carlo simulation also helps when people disagree about a forecast. Each person's view can become a range, and the simulation shows whether the disagreement matters. Sometimes the same option wins across all plausible ranges, which settles the debate quickly. Other times the ranking flips, which tells you where better information is worth gathering.

Turning estimates into probabilistic inputs

Probabilistic inputs are uncertain values described as ranges with likelihoods instead of single numbers. A probability distribution is the shape that says which values are likely and which are rare. For a renovation cost, you might give a low, a most likely and a high figure. The simulation then draws values from that shape during each iteration.

The choice of probability distribution should reflect what you know about each input. A uniform distribution treats every value in the range as equally likely. A triangular distribution adds a peak at the value you consider most likely. A normal distribution suits values that cluster around an average with symmetric spread.

Inputs that move together need to be linked in a simulation model, a property called correlation. Higher fuel prices, for example, can raise both shipping costs and travel costs at once. Treating linked inputs as independent can make the results look safer than reality. Where you lack data, a simple range with honest bounds is better than false precision.

DistributionWhat you supplyFits inputs that
UniformMinimum and maximumCould fall anywhere in a range
TriangularMinimum, most likely, maximumHave a clear likely value
NormalAverage and standard deviation (spread)Cluster symmetrically around an average
LognormalAverage and spreadCannot go below zero and have a long upper tail

Running the simulation step by step

Running a Monte Carlo simulation starts with a model that links inputs to the outcome you care about. The model can be a simple formula, such as total cost equals price plus repairs plus fees. Every uncertain input in that formula gets its own distribution. Inputs you know for certain stay as plain numbers in the formula.

Random sampling is the engine that drives each iteration of the simulation. In each iteration, the software picks one value from every distribution and computes the outcome. The results are stored, and the process repeats until the output distribution stabilizes. The number of iterations needed depends on the model's complexity and how precise you need the tails.

Checking convergence tells you whether the simulation has run long enough. Convergence means the summary figures barely change when you add more iterations. If the average or the low-end result still drifts, run more iterations. Rerunning with a different random seed, the starting point for the random numbers, is a quick test.

  • Define the decision and the outcome to measure
  • Build a formula linking inputs to that outcome
  • Assign a distribution to each uncertain input
  • Link inputs that move together
  • Run iterations and store each result
  • Check that summary figures have stabilized
  • Compare options on the same simulated scenarios

Reading simulated outcomes and expected value

Simulated outcomes can be shown as a histogram, a bar chart of how often each result appeared. Tall bars mark the results that came up often across the iterations. A wide histogram signals high uncertainty, while a narrow one signals a more predictable option. The shape also reveals skew, meaning a long tail of rare but extreme results.

Expected value is the average of all simulated outcomes, weighted by how often each occurred. Expected value is useful for comparing options that you face repeatedly or across many small bets. For a one-time decision, the average alone can mislead you about what might actually happen. An option with a higher expected value may still carry an unacceptable chance of loss.

Percentiles turn the simulation into plain statements about risk. A percentile is the value below which a given share of simulated outcomes falls. The tenth percentile, for example, is the result that only one in ten outcomes falls below. A cumulative chart, often called an S-curve, shows the chance of staying under any target.

Setting decision criteria for risk under uncertainty

Decision criteria are the rules you use to turn simulated results into a choice. Writing them down before running the model keeps the results from bending your judgment. A criterion might be the highest expected value, or the lowest chance of exceeding a budget. Another might be the option with the least bad worst-case outcome.

Risk under uncertainty means you know the possible outcomes only as probabilities, not certainties. Your tolerance for that risk shapes which criterion fits your situation. Someone who cannot afford a large loss may favor a narrow range over a higher average. Someone who can absorb setbacks may accept a wide range for a stronger expected result.

Combining decision criteria often gives a more balanced choice than relying on one figure. One approach is to set a minimum acceptable outcome at a chosen percentile. Options that fail that threshold drop out before averages are compared. The remaining options can then be ranked on expected value or another stated goal.

Using simulation to rank a shortlist

Monte Carlo simulation can rank a shortlist by testing every option against the same uncertain conditions. Each listing gets its own model, but shared inputs like interest rates draw identical random values. Using the same draws for shared inputs makes the comparison fair and reduces random noise. Differences in the results then reflect the options themselves, not luck in the sampling.

Keeping each option's inputs side by side makes a simulation model easier to check and explain. A table listing each option's ranges, sources and assumptions shows where every number came from. You can then see which assumptions drive the ranking and which barely matter. Bringing your own information, such as quotes or past bills, narrows ranges that would otherwise be guesses.

Win counts offer a simple way to read a shortlist simulation. For each iteration, note which option produced the preferred result under your criteria. The share of iterations each option wins shows how robust its lead is. An option that wins narrowly on average but rarely wins outright deserves a closer look.

Limits and mistakes to watch for

A Monte Carlo simulation is only as reliable as the inputs and the model behind it. Ranges that are too narrow produce results that look far more certain than they are. A missing input, such as a fee you forgot, shifts every outcome without warning. Reviewing assumptions with someone who knows the subject helps catch these gaps.

Sensitivity analysis shows which inputs move the outcome the most. One method is a tornado chart, which ranks inputs by how much they swing the result. Inputs near the top deserve better data or a second opinion. Inputs near the bottom can stay rough without changing the decision.

Simulation output describes possibilities, not predictions of what will happen. Rare events outside your ranges will not appear in the results at all. For high-stakes financial, medical or engineering decisions, a qualified specialist should review the model. Treat the results as one input to judgment rather than a replacement for it.

Monte Carlo decision analysis questions

How does Monte Carlo simulation help with decision making?

Monte Carlo simulation helps by showing the full range of outcomes each option could produce. You see the average result alongside the chance of good and bad results. That view lets you choose based on risk as well as expected value.

How many iterations does a Monte Carlo simulation need?

The number of iterations depends on the model and the precision you need. A Monte Carlo simulation has run enough when its summary figures stop changing as you add runs. Tail results, such as rare losses, need more iterations to settle than the average.

Can I run a Monte Carlo simulation in a spreadsheet?

Yes, a spreadsheet can run a simple Monte Carlo simulation using its random number functions. You build the outcome formula, fill it with random draws, and copy it down many rows. Larger models with linked inputs may be easier in dedicated simulation software.

What is the difference between expected value and the most likely outcome?

Expected value is the probability-weighted average of all simulated outcomes. The most likely outcome, or mode, is the single result that appears most often. The two figures differ when a long tail skews the results. For a one-time decision, compare both alongside the full range.

Is Monte Carlo simulation worth it for small decisions?

No, a small or easily reversed decision rarely justifies building a Monte Carlo simulation. A clear side-by-side comparison of known facts is faster and just as useful there. Simulation pays off when several uncertain inputs combine and a wrong choice is costly.

Read next

Latest articles