What Is a Monte Carlo Simulation in Retirement?

By the Retirement Odds Lab editorial team · Updated November 4, 2026

Quick answer: A Monte Carlo simulation in retirement runs your plan through thousands of invented market histories, each built by random draws from a set of assumed returns, and counts how often the money lasts to the age you chose. The count is a description of the model, not a forecast. In our example plan, 4,367 of 5,000 paths lasted, a success rate of 87% under the default assumptions.

Education, not financial advice. This page explains a method and shows what a model produced from stated assumptions. It does not tell you what to buy, sell, save or spend. For your own situation, ask a qualified professional.

The idea in plain terms

The name comes from a casino district, and the connection is chance. A Monte Carlo method answers a question by playing it out many times with random inputs and looking at the pile of results. You do not solve for one answer. You roll the dice ten thousand times, record where each roll landed, and read the shape of the pile.

Retirement suits this approach because the thing you most want to know, whether the money lasts, depends on something nobody can know in advance: the order and size of market returns over the next thirty or forty years. A single guess at that future is a story. Five thousand different futures, each equally plausible given past data, is a map of how wide the uncertainty is.

Think of it as a stress rehearsal rather than a crystal ball. The simulation does not say what will happen. It says what happened to your plan in a large collection of made-up scenarios that behave statistically like history, and it lets you count the ones that ended badly.

How one simulated path is built

Every path follows the same recipe, repeated year by year from today until the plan-to age.

First, the model needs inputs about you: current age, the age you stop working, the age you plan to last until, what you have saved, what you add each year, what you expect to spend after retiring, and any steady income such as Social Security with the age it starts. Second, it needs inputs about markets: for each asset class, an average return, a measure of how much the return bounces around from year to year (the standard deviation, or volatility), and how the assets tend to move in relation to each other (correlation).

Then, for each year of a path, the model draws one random return for stocks and one for bonds. The draws are not independent coin flips; they respect the averages, the volatility and the correlations you supplied. The balance grows by that year's return. During working years, savings go in. After retirement, spending comes out and income comes in. If the balance reaches zero before the last year, that path is a failure. If money remains at the plan-to age, it is a success.

Do that 5,000 times with fresh draws each time and you have 5,000 paths. The success rate is the number that lasted divided by the number you ran. That is the entire trick, and everything else is bookkeeping. How the calculator orders the cash flows within a year, which random distribution it uses and which checks we ran are written up on how this simulation works.

A worked example with the default plan

Here is the example plan that loads on the Monte Carlo simulation retirement calculator. It is age 45, retiring at 65, planning to 95; $300,000 saved, $10,000 saved a year, $65,000 a year of spending in today's dollars, $24,000 a year of Social Security from 67, 60% stocks and 40% bonds. All figures here are in today's dollars, which is the same as saying the returns have inflation already removed. The assumptions come from annual data for 1971 to 2025, and the stock, bond and gold figures in the table of sources on the methodology page are computed from that data.

For stocks, those assumptions are an average real return of 8.4% with a volatility of 17%. For bonds, they are 2.5% with a volatility of 10%. A 60/40 mix blends the two. Because these averages are arithmetic means of yearly returns, the compound rate an investor would actually have seen is lower: 7.0% for stocks and 2.0% for bonds. Volatility drags the compounded figure below the simple one, which is another reason a single average line can mislead.

Run the example 5,000 times and 4,367 of 5,000 paths still had money at 95. That gives 87%. The other 633 paths ran out of money somewhere along the way. Among those failures, the median age at which the balance hit zero was 86, and 2% of the failures started in the first ten years of retirement.

The remaining detail is the spread of balances. At retirement, the paths were spread from $730,000 at the 10th percentile to $2,132,000 at the 90th, with $1,235,000 in the middle. At the plan-to age, the 10th percentile was $0, the median was $2,507,000 and the 90th percentile was $10,354,000. A percentile is a rank: the 10th percentile means 10 of every 100 paths ended at or below that figure. These numbers are what the model produced from the inputs shown, no more.

Notice what the example is not. The success rate is not a prediction for a real 45-year-old with these figures. It is how the model counted the paths it generated. A different data window moves it: with the longer 1928 to 2025 assumptions the same plan scored 83%.

What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.

How this differs from a single average-return line

The older way to project a retirement is a spreadsheet with one growth rate. You assume a steady number each year, subtract spending, and watch the line fall to zero or not. It is clear, quick and easy to check by hand. It is also the same plan run exactly once, in a world where nothing ever goes wrong in a particular order.

A simulation keeps the average but adds the spread. That matters mainly because of timing. Early in retirement the account is at its largest, and withdrawals come out of whatever balance it has just become; a poor first few years hurts more than the same poor years late on. The order of returns, in other words, is part of the answer, and an average-return line has no order. If you would like to try changing the inputs yourself, the free Monte Carlo retirement calculator on the home page does exactly this. Our page on sequence of returns risk demonstrates this with the same ten returns played in opposite directions.

A smooth line also hides a second thing: some paths do far better than the average, and those paths pull the average up. In the example plan the median balance at the plan-to age was $2,507,000 while the 90th percentile was $10,354,000. Looking only at the average of results, as opposed to the median, can make a plan look sturdier than most of its paths were.

Changing one input at a time

One of the most useful things a simulation does is let you move a single input and watch the count respond. Using the same example plan and the 1971 to 2025 assumptions, a few changes show how unevenly the inputs matter. Retiring five years earlier took the success rate to 57%, and retiring five years later took it to 98%. Taking out the Social Security income altogether gave 64%. Raising yearly spending by a tenth gave 81%, and by a fifth, 74%.

Market assumptions moved the result too, though by less than some of those life choices. Lowering the assumed average stock return by one percentage point gave 80%, by two points 69%, and by three points 57%. Raising it by one point gave 93%. Those are all outputs from one example plan, so read them as an illustration of which levers are long and which are short, not as a ranking that holds for every household.

Three common misreadings

The first is treating the percentage as a weather forecast. A score of the kind this site prints counts simulated paths. It does not say that a given share of retirees will run short, and it says nothing about the chance that you personally will. The second is reading precision into a single run. With 5,000 trials, a rerun can move the figure by a point or two without anything about the plan changing, so differences smaller than that are noise.

The third is assuming the model has seen everything. It has seen the 55 yearly observations that fed its averages. A decade unlike any in the sample can't appear in the draws unless the fitted distribution happens to produce it. When a simulation reports that nearly every path survived, it is reporting about its own world, which is a simplified copy of ours.

Why the answer is a rate and not a yes or no

A Monte Carlo result is a proportion because the question has no deterministic answer. Under identical inputs, some random paths survive and some do not. Reporting the share is the honest summary of that.

Two features of that share are worth holding onto. It depends on how many paths you run, and a different random seed will shift it slightly. In eight runs of the example plan with different seeds, 5,000 trials gave 87% to 89%. The share also depends heavily on the assumptions, which are a human choice. People who want to know how to read a particular number, and which numbers people tend to aim for, will find that discussed on the good Monte Carlo success rate page, and the 4% rule page shows how a well-known rule of thumb looks when run through the same machinery.

Random draws versus replaying history

There are two common ways to fill a simulation with returns. One draws random numbers from a distribution fitted to history, which is what this calculator does. The other replays actual historical sequences: start in 1966, run the plan through the real returns of each following year, then start in 1967, and so on. The two approaches answer slightly different questions, and each has strengths. A replay uses real events but only a few dozen starting years, and neighboring windows overlap heavily. Random draws can generate many more paths but depend on the fitted distribution being a fair summary. For a side-by-side test on the same dataset, see Monte Carlo vs historical simulation.

What a simulation leaves out

The model is only as good as what you give it. Every simulation carries assumptions that are easy to forget. This one uses annual steps, so it does not see what happens inside a year. It uses real returns, so it assumes inflation is already removed and does not let inflation spike in the future, which is a gap we cover in what a Monte Carlo simulation cannot see. It does not model taxes, health costs, or the way a person might cut or raise spending as markets move.

It also treats the past as a guide to the shape of the future. The returns come from a particular 55-year window in one country's markets. Whether future decades resemble that window is unknowable, and a model that draws from it can understate extremes that happened rarely or not at all.

Where do these gaps leave you? Mostly with a cautious reading rather than a dismissal. Knowing what a model omits is part of reading its output. When you compare tools, the questions to ask are things like whether the assumptions are visible, whether returns are real or nominal and how many trials were run; the page on how to pick a Monte Carlo retirement calculator turns them into a checklist.

What this can't tell you

This page can't tell you whether your own plan will work. It cannot tell you what markets will do, which year you will retire into, or what you will actually spend. It can't tell you whether 87% is a number you should be comfortable with, because that depends on your other income, your flexibility, your health and your tolerance for risk, none of which a page knows. And the example plan is an example; its inputs are not a recommendation for anyone. What it can do is show how a simulation works so that the next score you see makes more sense.

What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.

Frequently asked questions

Keep reading

• the methodology page: the inputs, the data and the checks behind this site's calculator
• what a success rate means: how to read a score without treating it as a verdict
• the order of returns: the same average return, in a different order, with a different ending