How Accurate Is Monte Carlo Simulation for Retirement Planning?

Say your plan shows an 85% chance of success. It sounds precise, but one real-world back-test found that traditional Monte Carlo models were markedly overconfident in exactly that zone, between roughly 70% and 97%. So how accurate is Monte Carlo simulation for retirement planning? It's useful, and it's imprecise. It's a good way to see the odds your plan faces and a poor way to predict your one actual future. How much you can trust the number depends on the model behind it and on the assumptions you feed it.

This article covers what the research shows, why the models miss, and how to read your own success rate.

How accurate is Monte Carlo simulation for retirement planning? The short answer

• As a forecast, it's rough. Back-tests against real market history show significant error, especially in traditional models.
• As a stress test, it's valuable. It shows how your plan holds up across many possible markets, including bad ones, instead of assuming one average return.
• The method matters. Some kinds of Monte Carlo model tracked real outcomes noticeably better than others.
• The success rate leaves things out. It tells you how often a plan fails. It doesn't tell you how badly.

Treat the number as a probability, not a promise.

What Monte Carlo actually does, in plain terms

A simple retirement calculator assumes something like a 6% return every year and draws one smooth line. Real markets don't work that way. Some years are up, some are down, and the order they come in matters a great deal.

Monte Carlo simulation runs your plan many times. Each run, or "future," uses a different random sequence of market returns. Then it counts how many of those futures still have money at the end. If 850 of 1,000 stay funded, you see 85%.

The order of returns is the key. A crash in your first few years of retirement, while you're already withdrawing, does far more damage than the same crash at 85. This is called sequence-of-returns risk. Because Monte Carlo shuffles the order of returns in every trial, it captures that risk in a way an average-return projection can't.

Researchers use the same approach to set spending rules. Morningstar's 2025 research puts the safe starting withdrawal rate at 3.9%. "Safe" there means money is still left in 90% of simulated outcomes over a 30-year retirement, not that an average return works out. (Source: Morningstar, The State of Retirement Income 2025, Dec 2025.)

The real-world test: does "80%" actually mean 80%?

The best way to judge a probability is calibration. Take all the plans a model rated at 80%. Did about 80% of them actually succeed? A weather forecaster who says "70% chance of rain" should see rain on about 70% of those days. Retirement models can be checked the same way.

Researchers Derek Tharp and Justin Fitzpatrick ran this check for Kitces.com and Income Lab, assessing how Monte Carlo models perform against real outcomes. They scored each model with a Brier score, a standard accuracy measure where 0 is perfect and 1 is the worst possible.

They compared four types of model:

What this means for you: if your plan shows somewhere in the 70s, 80s or low 90s, a traditional model may be flattering you. Every model in the study showed significant error against actual results. The better ones were simply less wrong.

Why the models miss: the assumptions underneath

A Monte Carlo result is only as good as what goes into it. These are the usual weak spots.

Returns that behave too neatly

The Wharton/Pension Research Council paper Stress Testing Monte Carlo Assumptions notes that most simulations assume returns are independent and identically normally distributed. In plain English, each year is a fresh draw from the same bell curve, with no memory of the year before. Real markets show "fat tails," meaning extreme years happen more often than a bell curve predicts. Expected returns also change over time.

Return assumptions that are too rosy

If the model assumes future returns will match the past century's average, every result inherits that optimism. That's why "reduced-CMA" models exist. CMA stands for capital market assumptions, the expected returns a model is fed. As the back-test showed, trimming those assumptions helped but didn't remove the overconfidence.

Spending that never changes

Most simulations assume you withdraw a fixed amount, raised for inflation every year, no matter what. Real people don't usually do that. After a bad year, most retirees cut back a little: skip a trip, delay a car. A model that assumes you keep spending the same through a crash describes a more rigid retiree than most people are. That tends to make the success rate look worse than real life would be, though it depends on how flexible you really are.

A horizon that's too short

A plan that ends at 85 can show a great success rate simply because it stops early. Social Security's planning guidance says about one in three 65-year-olds may live to at least 90. (Source: Social Security Administration, Retirement Information for Medicare Beneficiaries, 2026.) A short horizon quietly bets against a third of retirees.

Costs left out

Health care is the big one. Fidelity estimates a 65-year-old retiring in 2025 will need $172,500 for Medicare premiums, deductibles and prescriptions over retirement, and that figure doesn't include long-term care. (Source: Fidelity Investments, 2025 Retiree Health Care Cost Estimate, Jul 2025.) If your spending input doesn't account for it, no model can fix that.

Too optimistic or too pessimistic? Explaining the fat-tails puzzle

There's a seeming contradiction here. The back-test above says traditional Monte Carlo is too optimistic. A separate Kitces analysis, Fat Tails In Monte Carlo Analysis vs Safe Withdrawal Rates, says it can be too pessimistic. In that analysis, a plan that succeeded in 100% of historical scenarios showed only about 95% in Monte Carlo. It also found that at annual horizons the downside fat tails "largely disappear," so Monte Carlo may overstate long-run risk.

Both can be true, because they measure different things:

• The fat-tails analysis asks whether random draws exaggerate the chance of disaster over a long retirement, compared with the worst stretches in history. Sometimes they do. Random shuffling can string together more bad years in a row than markets have actually produced, because real markets have tended to recover after bad runs.
• The back-test asks whether the model's confidence matched what actually happened to plans in the success range most people see. There, traditional models were too confident.

The practical takeaway: no single model is exactly right. A traditional Monte Carlo score can overstate the risk of total disaster and still overstate your odds in the 70% to 97% band. That's a good reason to look at a range of outcomes and to test more than one assumption.

What the success rate doesn't tell you

Even a well-calibrated success rate is a blunt measure.

It ignores how badly a plan fails. The CFA Institute's Enterprising Investor blog put it plainly in Rethinking Retirement Planning Outcome Metrics (2023): probability of success "completely ignores the magnitude of failure." Running out of money at 94 with Social Security still coming in is very different from running dry at 72. Both count as one "failure."

Aiming for 100% has a cost. Income Lab argues in Why Probability of Success Is Wrong for Retirement Planning that chasing near-certainty leads people to underspend. You give up real years of travel, help for family and peace of mind to protect against a small number of extreme futures. A very high score can mean you're likely to die with far more than you needed.

So a lower score isn't automatically bad, and a higher one isn't automatically good. What matters is what the failures look like and whether you'd adjust in time.

How to read your own Monte Carlo result

Some steady ways to use the number well:

1. Read it as a range, not a point. An 82% and an 86% are, for practical purposes, the same answer.
2. Be extra skeptical in the 70% to 97% band. That's where traditional models showed the most overconfidence.
3. Change one input at a time. Retire a year later. Spend a little less. Assume lower returns. If a small change swings your score a lot, your plan is fragile. That matters more than the score itself.
4. Run it long enough. Plan past 90. One in three 65-year-olds may get there.
5. Put health care in. Treat it as a real line item, not a rounding error.
6. Decide your adjustment plan now. Know what you'd cut, and when, if markets go against you early.
7. Ask what model the tool uses. Traditional, historical or regime-based? It changes how much weight to give the result.

The plan you build at Monte Climb was designed around several of these points. It runs to age 95 by default, and you can extend it. It charges Medicare premiums and income-based surcharges year by year instead of using a flat guess. It reports results as a success rate across 1,000 simulated futures, not a single average-return line. You can also ask William, our AI planner, follow-up questions such as "Am I on track?", "Can I retire at 62?" or "What's a sustainable spend?" It's planning software, not advice.

The bottom line

Monte Carlo simulation won't tell you what will happen. It tells you how your plan holds up across many things that could happen, and that's still far more honest than one average-return projection. The accuracy varies. Traditional models tend to run overconfident in the range most people land in, and every model rests on assumptions you should question.

Use it like a stress test. Push on it, change one number at a time and watch what breaks.

Three numbers to start: your age, your savings, your monthly spending. Show my answer and see how your plan does across 1,000 futures. It runs in your browser, and nothing is saved until you say so.

Frequently asked questions

How accurate is Monte Carlo simulation for retirement planning?

It's useful but imprecise. A Kitces/Income Lab back-test found significant error against real outcomes. Traditional models were markedly overconfident when they showed success rates above 50%, especially between 70% and 97%. Historical and regime-based models scored about 25% better on the Brier accuracy measure.

Is an 80% Monte Carlo success rate good enough?

It depends on what the failures look like and how flexible your spending is. A success rate ignores how badly a plan fails, and chasing 100% can lead you to underspend. Read 80% as a range, test how sensitive it is to small changes, and know in advance what you'd adjust if markets go badly early on.

Why do Monte Carlo and historical back-tests give different answers?

They measure different things. Random Monte Carlo draws can string together more bad years than history has produced, which can overstate the risk of total disaster over a long retirement. Separately, back-tests found traditional models were overconfident in the 70% to 97% success range. Both effects can exist at the same time.

What assumptions make Monte Carlo retirement simulations less accurate?

The common ones are returns modeled as independent draws from a bell curve (real markets have fat tails), expected returns that are too optimistic, fixed inflation-adjusted withdrawals regardless of markets, planning horizons that end too early, and missing costs such as health care.

How many Monte Carlo simulations are enough for a retirement plan?

Around 1,000 trials is a common standard. In practice, running many more adds little, because the quality of the assumptions matters far more than the number of trials.