How accurate is a Monte Carlo probability of success?

It is precise about its own assumptions and uncertain about the future. In this guide's example, the same plan read anywhere from 56% to 83% depending on the stretch of history and the method behind it. Treat a single figure as one estimate among several.

Do more simulations make the result more reliable?

Only up to a point. More paths shrink the sampling noise (under one percentage point at 10,000 paths), but they cannot fix the assumptions. If the model learned from an unusually good or bad stretch of history, more paths just repeat that view more precisely.

Which history should a Monte Carlo use?

There is no single right answer. Longer histories include more kinds of markets, crashes among them, while recent years may reflect today's conditions better but can be dominated by one long rally or one bad year. Looking at several windows side by side is more honest than picking one.

What is a good probability of success?

Many planners aim for something like 80% to 90%, but the number matters less than how sturdy it is. A plan that stays above your bar under less favourable assumptions is in better shape than one that shows 95% under a hopeful reading and 60% under a sober one.

Why is the success rate different every time I run it?

Each run draws a new random set of futures, so the figure moves by a fraction of a point between runs. Our tools store the random seed, so the same inputs and seed reproduce the same result exactly. Bigger differences come from changing the inputs, the method or the history window.

How Sure Is Your Success Rate? Why One Monte Carlo Number Can Mislead You

7 min read · Updated 2026-10-28

Add us as a preferred source on Google

Retirement and goal calculators love a single number: "you have an 87% chance of success." It looks precise, and it is easy to read as a fact about your future. It is not. It is the answer to a narrower question: if markets behave the way this particular model, fitted to this particular stretch of history, says they do, how often does the plan work?

Change the model or the history and the number changes, sometimes a lot. This guide shows how much with a worked example you can reproduce, separates what matters from what does not, and explains how to use a success rate without being fooled by it. Educational, not advice.

Check how sure your plan is →

What a success rate really measures

A Monte Carlo simulation runs your plan through thousands of possible futures and counts how many reach the goal. Each future is built from assumptions: how much the portfolio earns on average, how much it swings, and how good and bad months follow one another. Those assumptions come from history, and there is more than one reasonable way to read history.

So the success rate inherits two choices you rarely see: which stretch of the past the model learned from, and what kind of model turned that past into futures. Neither choice is wrong. They are just choices, and a single percentage hides them.

One plan, nine answers

Take a saver with $100,000 today who adds $1,000 a month and wants $1,000,000 in 20 years, invested in a classic 60/40 mix of stocks and bonds. We ran the plan 10,000 times under each of nine sets of assumptions: three stretches of the mix's real monthly history (the full history from Oct 2003, the last 10 years and the last 5 years), and the three ways our simulators can turn history into futures.

The answers run from 56% to 83%. Read one way, the plan is a coin flip that needs a rethink; read another, it is comfortably on track. Same saver, same money, same portfolio.

Chance of reaching $1,000,000 in 20 years, starting from $100,000 plus $1,000 a month
History the model learned from	Bell curve (normal)	Reshuffled months	12-month blocks
Full history (from Oct 2003)	56%	56%	60%
Last 10 years (from Oct 2016)	79%	79%	83%
Last 5 years (from Oct 2021)	56%	57%	58%

Goal: $1,000,000 in 20 years. 10,000 simulated paths per cell, calibrated on monthly returns of a 60/40 (SPY 60%, AGG 40%), rebalanced yearly, dividends reinvested, Oct 2003 to Sep 2026. Contributions flat, figures in future (not inflation-adjusted) dollars.

What moved the number, and what did not

The model barely mattered. Within any one stretch of history, the three methods landed within 4 points of each other. Whether the simulator draws from a bell curve, reshuffles real months, or replays whole years at a time changed the answer only a little here.

The history mattered enormously. With any one method, switching the stretch of history moved the answer by 23 points or more. The last 10 years, a strong decade for stocks despite the 2020 and 2022 drops, averaged about 10.3% a year and gave the most hopeful answers. The full history since 2003, which includes the 2008 crash, averaged about 8.5%, and the last 5 years, which include the 2022 drop in both stocks and bonds, averaged about 8.8% with bigger swings. Less than two points of average return, compounded for 20 years, is the difference between a coin flip and a comfortable plan.

Running more simulations would not help. At 10,000 paths, the sampling noise in each figure is under one percentage point. The spread in the table is not noise. It is uncertainty about the assumptions themselves, and no amount of computing removes it.

Chance of reaching $1,000,000 by stretch of history (bell-curve method, percent)

Full history (from Oct 2003)

56%

Last 10 years (from Oct 2016)

79%

Last 5 years (from Oct 2021)

56%

When the models agree

Disagreement is not inevitable. Give the same saver more room, $200,000 today and $1,500 a month toward the same $1,000,000, and all nine sets of assumptions land between 94% and 99.5%. When a plan has a real margin, it stays on track however you read history.

That is the practical lesson. A plan whose success rate depends on which years you believe is a fragile plan, even if the number you happened to see was high. A plan that clears the bar under the less favourable readings is a sturdy one.

Chance of reaching $1,000,000 in 20 years, starting from $200,000 plus $1,500 a month
History the model learned from	Bell curve (normal)	Reshuffled months	12-month blocks
Full history (from Oct 2003)	97%	96%	95%
Last 10 years (from Oct 2016)	99%	99%	99.5%
Last 5 years (from Oct 2021)	94%	94%	95%

How bad is a miss?

A success rate only counts how often a plan falls short, not by how much. For the first saver, the simulated paths that missed fell short by $155,000 to $233,000 on average, depending on the assumptions, and the worst 1 in 20 outcomes ended below $562,000 to $796,000.

Missing a million-dollar goal by ten thousand dollars and by a quarter of a million are very different futures, yet both count as one failure. Look at the size of the shortfall as well as its odds.

How to use a success rate without being fooled

A success rate is still useful. It just needs company.

•Ask for the range, not the point. Re-run your plan with a different stretch of history and a different method, or use a tool that does it for you, and read the lowest answer as seriously as the highest.
•Know which history you are trusting. A recent bull market makes almost every plan look safe; a window with a crash in it is a harder test, and arguably a fairer one.
•Build a margin. If your plan only works under the hopeful readings, save more, aim a little lower, give it more time, or plan to adjust spending along the way.
•Check the size of a miss. A plan that sometimes falls a little short is a different thing from one that sometimes falls far short.
•Re-run it every year. Your balance, the market and your goal all change, and a success rate is a snapshot, not a promise.

Check your own plan

The Goals planner, the Monte Carlo tool and the Retirement lab now show this for you. Under the headline success rate, a "How sure is this number?" card re-runs your plan under several models of how markets move, each calibrated on different stretches of your portfolio's own history, and shows the spread, whether the models agree, and how severe a miss could be.

To reproduce the table above yourself, open the Goals planner, enter the plan with SPY 60% and AGG 40%, untick growing contributions with inflation, and switch the method and the lookback start date. Your figures will differ by a point or so, because every run draws its own set of futures. Educational, not advice: what a success rate means for you depends on your whole financial picture.

Try it yourself

Financial Goals PlannerSee your odds of reaching a target, and how sure that number is.Open tool →Monte Carlo SimulationRun thousands of futures for any contribution or withdrawal plan.Open tool →Retirement Withdrawal LabCheck whether your savings last, and the income left in weak markets.Open tool →

FAQ

How accurate is a Monte Carlo probability of success?: It is precise about its own assumptions and uncertain about the future. In this guide's example, the same plan read anywhere from 56% to 83% depending on the stretch of history and the method behind it. Treat a single figure as one estimate among several.
Do more simulations make the result more reliable?: Only up to a point. More paths shrink the sampling noise (under one percentage point at 10,000 paths), but they cannot fix the assumptions. If the model learned from an unusually good or bad stretch of history, more paths just repeat that view more precisely.
Which history should a Monte Carlo use?: There is no single right answer. Longer histories include more kinds of markets, crashes among them, while recent years may reflect today's conditions better but can be dominated by one long rally or one bad year. Looking at several windows side by side is more honest than picking one.
What is a good probability of success?: Many planners aim for something like 80% to 90%, but the number matters less than how sturdy it is. A plan that stays above your bar under less favourable assumptions is in better shape than one that shows 95% under a hopeful reading and 60% under a sober one.
Why is the success rate different every time I run it?: Each run draws a new random set of futures, so the figure moves by a fraction of a point between runs. Our tools store the random seed, so the same inputs and seed reproduce the same result exactly. Bigger differences come from changing the inputs, the method or the history window.

Stop guessing. Run the numbers on your own portfolio, free.