A Monte Carlo retirement plan with a 90% success rate sounds robust. Most retirees who see that number conclude the plan is safe. The actual interpretation is more complicated, and the difference matters when the plan is the one you have to live on.
A 90% success rate is not a 90% chance the plan works. It is conditional on a long list of assumptions that the model treats as fixed and that real life does not. It says nothing about when the 10% of failures happen, how badly they fail, or how robust the success in the other 90% is. Two plans with identical 90% success rates can have very different real-world risk profiles.
This is not a critique of the metric. It is a critique of how it gets read. A Monte Carlo success rate is a useful summary of what a model says under a specific set of assumptions. Treating it as a probability of plan success in real life confuses the model output with the underlying truth it is approximating.
What "Success" Actually Means
In a Monte Carlo retirement simulation, the model runs N paths (commonly 10,000-50,000) where each path is a different sequence of random returns drawn from an assumed distribution. For each path, the simulator tracks the portfolio balance from retirement to end of horizon, applying the withdrawal rule each year.
A path "succeeds" if the portfolio balance is above zero at the end of the horizon. A path "fails" if the balance hit zero before the horizon ends. The success rate is the fraction of paths that succeeded.
That definition has three implicit choices that almost no one inspects:
The horizon is fixed. Most simulators use 30 years from retirement age. A 65-year-old retiree is implicitly modeling to age 95, not their actual life expectancy distribution. If you live to 97, the simulator never tested that scenario; if you die at 80, the simulator counted the post-80 portfolio balance even though it was not relevant to your actual life.
"Success" is binary. A path that ends with 5,000 left is just as successful as one that ends with 3,000,000. A path that runs out at year 29 is just as failed as one that runs out at year 5. The metric collapses an enormous amount of variation into one bit per path.
The failure mode is binary too. "Above zero" or "below zero." There is no concept of "partial success" or "income shortfall." A path where the retiree had to cut spending 50% in years 20-30 to stay above zero is a "success." So is a path where the retiree was rolling in money the entire time.
The success rate captures one specific question: did the portfolio survive the horizon? It does not capture how it survived, how much was left, or how close some near-failures came.
The Probability Is Conditional
Every Monte Carlo success rate carries an implicit footnote: given these assumptions. Change the assumptions and the number changes, often dramatically.
The assumptions that matter most:
| Assumption | Typical effect of small change on success rate |
|---|---|
| Expected return: -1 percentage point | -15 to -20 points |
| Distribution: normal vs fat-tailed | -5 to -10 points (fat-tailed is worse) |
| Inflation: +1 percentage point | -5 to -10 points |
| Asset correlation: independent vs correlated | +/- 3 to 5 points |
| Withdrawal rule: static vs dynamic | +5 to +12 points (dynamic is better) |
| Volatility: +5 percentage points | -3 to -6 points |
A plan that is 90% success at "historical US returns, normal distribution, independent assets, static 4%" might be 70% success at "forward-looking returns, fat-tailed distribution, correlated assets, static 4%." Same retiree, same portfolio, same withdrawal strategy. Different assumption set.
The 90% number is not wrong; it is precise. It says exactly what the model produces under those assumptions. The mistake is treating it as the probability of the plan working in real life, which is a probability over the assumption uncertainty, not just within it. The honest version of any Monte Carlo result is "90% success at these assumptions, 70% at those, 95% at the optimistic case." The single headline number compresses out the assumption uncertainty.
Failure Distribution Matters
The 10% of "failures" in a 90% success plan are not equivalent. A plan that runs out of money in year 29 of a 30-year horizon is a success in everything but the technicality. A plan that runs out in year 12 is a catastrophe.
Two plans with identical 90% success rates:
| Plan | Median failure age (in failed paths) | Worst 10% income, year 15 |
|---|---|---|
| A | year 28 | 36,000 |
| B | year 14 | 18,000 |
Plan A is more robust in any meaningful sense. The failures happen late, the worst-case income is livable, and the retiree has time to adjust if conditions deteriorate. Plan B fails early and brutally - a retiree on the wrong path is broke at age 79 with no time to recover.
The success rate metric is identical for both. Reading only the success rate, the two plans look the same. Looking at the failure distribution, they are completely different.
This matters because Monte Carlo simulations are usually used to compare alternatives. If you are choosing between two 90% plans, the success rate cannot distinguish them. Failure age distribution and worst-case income paths do.
Median vs Average vs Worst Case
A complete reading of Monte Carlo output looks at four numbers, not one:
Success rate. The summary metric. Useful as a first filter but never sufficient.
Median ending portfolio. The 50th percentile final balance. Tells you the typical leftover capital. A high success rate with a low median ending portfolio means most paths succeed but barely - the plan is borderline.
10th percentile income at year 15. The worst-decile spending level halfway through retirement. Tells you what the bad paths actually feel like. A 10th percentile income of 18,000 in real terms is a very different reality from one of 35,000.
Failure age distribution. Among the failure paths, when do they fail? Late failures are recoverable; early failures are not. A 90% success rate where the median failure age is year 28 is dramatically more robust than one where the median is year 12.
A retirement plan that performs well on all four is genuinely robust. A plan that excels on success rate but is mediocre on the others is fragile in ways the headline number hides.
How to Read the Output Honestly
Three habits separate honest interpretation from naive reading.
Read the assumption set first, the success rate second. A 95% success rate at "7% real return, normal distribution, no fat tails, no black swans" is not as informative as 88% success at "5% real return, fat-tailed distribution, correlated assets." The second number is more conservative because the assumptions are. Compare numbers under matched assumption sets, not in isolation.
Run the same plan under multiple assumption sets. Optimistic, central, and bear cases. The spread between them tells you how sensitive the plan is to your beliefs about the future. A plan that is 95% success at the bull case and 70% at the bear case is more fragile than a plan that is 90% across all three.
Look at the failure distribution, not just the rate. Most simulators show a histogram of ending portfolio values or path-by-path balances. The shape of the failure distribution tells you how the plan dies when it dies. Late failures with gentle declines are different from early failures with cliff drops.
Stress-test specific scenarios on top of the random distribution. A black swan crash at age 67 is not a random event the simulator might or might not generate. Force it explicitly and see how the plan handles the worst plausible timing. Random Monte Carlo tells you the average case; deterministic shocks tell you the worst-case behavior.
What This Means for Your Plan
Stop reading the success rate as a probability of plan success. Read it as "what the model says under these assumptions." The two are related but not identical. The gap is the assumption uncertainty.
Aim for 85-95% under conservative assumptions, not 99% under optimistic ones. A high success rate from optimistic inputs is not robustness; it is a model that is hiding the risk. The honest plan has a lower headline success rate at conservative inputs, but the number reflects something closer to reality.
Look at the 10th percentile income, not just the median. The plan you can live on is the plan that delivers acceptable income in the worst 10% of paths, not the one that has a great median outcome. If the 10th percentile income is below your survival floor, the success rate does not save you - the plan fails functionally even if it succeeds on paper.
Test the assumptions, not just the parameters. The right expected return and distribution shape matter more than the withdrawal rate. A plan calibrated to optimistic inputs is fragile by construction.
Use the success rate to compare plans, not to certify a single plan. If Plan A has 90% success and Plan B has 92%, the difference is meaningful only if the assumption sets are identical and the failure distributions are similar. Otherwise, the comparison is between different things.
The success rate is a useful summary. It is not the answer. The complete picture comes from reading the assumption set, the failure distribution, and the worst-case income paths together. A plan that looks good across all of those is a plan that has a real chance of working. A plan that looks good only on the headline number is a plan that may not survive contact with the actual market.
Frequently Asked Questions
- What does a 90% Monte Carlo success rate actually mean?
- It means that across the simulated paths the model ran, the portfolio remained above zero at the end of the horizon in 90% of them. It does not mean a 90% chance the plan works for your real life. The 90% is conditional on every assumption in the model (returns, volatility, inflation, withdrawal rule, correlations, distribution shape). Change any of those and the number changes.
- Why are some 90% success plans riskier than others?
- Because the 10% of failures are not equal. Some failure paths deplete the portfolio in year 28 (mostly fine - you lived a full retirement). Others deplete in year 12 (catastrophic). The standard success rate metric does not distinguish them. Two plans with identical 90% success rates can have completely different median failure ages and very different practical risk profiles.
- What success rate target should I aim for?
- Most planners suggest 85-95% as the practical range. Below 85% is fragile; above 95% means you are likely under-spending and over-saving. The right target depends on your tolerance for income reduction (a plan that fails should fail with warning, allowing spending cuts) and on whether you are confident in the underlying assumptions. A 90% success rate at conservative assumptions is more robust than a 95% rate at optimistic ones.
- Should I look at success rate alone or other metrics too?
- Always look at multiple metrics. Success rate tells you whether the portfolio survives. Median ending portfolio tells you the typical leftover capital. The 10th percentile path tells you the worst-case income you might need to live on. Failure age distribution tells you when failures happen. A complete reading uses all four. Looking at success rate alone is the most common Monte Carlo modeling error.
- Why does my success rate change so much with different assumptions?
- Because Monte Carlo success rates are extremely sensitive to inputs. A 1 percentage point reduction in expected return drops success by 15-20 points. Switching from a normal distribution to a fat-tailed distribution drops it by another 5-10. Adding correlated asset returns shifts it by 3-5 points. Each adjustment compounds. The headline number is honest about what the model says; it is not the same as the probability of your plan working.