What a Monte Carlo Simulation Cannot See
Quick answer: A simulation can only vary what it was built to vary. Ours varies yearly market returns and nothing else. Your spending, your tax bill, your health, your lifespan, your reaction to a bad year and the correctness of the model itself are all held still or left out. Those gaps matter because a score can look precise while resting on a fixed spending line and a short window of history. The table later on this page shows how far the default plan's score moved, under the default assumptions (1971 to 2025 real returns, 60/40), when we shifted a few of those held-still inputs: from 87% to as low as 57% and as high as 98%.
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.
What the score actually covers
Before listing the gaps, it helps to recall the design. What a Monte Carlo simulation is comes down to drawing many random sequences of yearly returns and counting how many paths still hold money at the end. The randomness is confined to returns. Everything else you type into the Monte Carlo simulation retirement calculator, such as your age, savings, spending and fixed income, stays exactly where you put it for the whole run.
That is a sensible choice for a first model, and the output is useful for what it is. But the reader who sees a single percentage and treats it as a measurement of their retirement is reading more into it than the method supports. Every item below is a place where reality has a say and the model does not.
Inflation that does not behave like the model
The calculator works in today's dollars with real returns, meaning returns already adjusted for inflation. In the dataset, each year's nominal return is divided by one plus that year's change in the consumer price index, and the index is the FRED series CPIAUCNS. The Bureau of Labor Statistics page we read on 2026-11-04 presents the CPI as an average of price changes for urban consumers across a basket of goods and services.
Two consequences follow. First, inflation is already inside the return figures, as it happened in the years the dataset covers. There is no separate inflation path that can run hot for a decade and cool for the next. Second, your own costs may not track a broad basket. If the things you buy rise faster than the average, the spending line you entered understates your real need. We make no claim about current inflation, and this page prints no inflation figure.
The closest stand-in the calculator offers is the spending input. In the table below, raising the default plan's yearly spending by one tenth moves the score from 87% to 81%, and raising it by one fifth moves it to 74%. Those are not inflation forecasts; they are a way of asking how sensitive the answer is to being wrong about the cost of living.
Spending that is not a straight line
The model withdraws the same real amount every year, from the first year of retirement to the last. Real spending is lumpier. There is a year with a large repair, a year with help for a family member, a year when travel is the priority and a later year when it is not. Averages smooth these out; sequences do not, and a large cost that lands right after a market fall is more damaging than the same cost five years on, for the reason set out under sequence of returns risk.
The retirement date is another held-still input. Moving it is not a small tweak, because it changes both how long savings are added and how long they are drawn on. For the default plan, retiring five years earlier gives 57% and five years later gives 98%. The swing is large because a single choice acts on both sides of the plan at once.
Here is the full set of sensitivity runs for the default plan, each changing exactly one input.
| Change | Success rate |
|---|---|
| Default example plan | 87% |
| Retire 5 years earlier | 57% |
| Retire 5 years later | 98% |
| Spend 10% more in retirement | 81% |
| Spend 20% more in retirement | 74% |
| No Social Security or pension income | 64% |
| Plan to age 100 | 84% |
| Plan to age 105 | 83% |
| Stock return 1 point lower | 80% |
| Stock return 2 points lower | 69% |
| Stock return 3 points lower | 57% |
| Stock return 1 point higher | 93% |
| Stock volatility 3 points higher | 82% |
| Stock volatility 3 points lower | 93% |
| Bond return 1 point lower | 82% |
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 long you will live
A simulation takes the planning age as a given. It has no notion of the chance that you live past it. Moving the default plan's end age from 95 to 100 gives 84% and moving it to 105 gives 83%, against 87% at 95. The fall is small. It is consistent with the default plan's middle path still holding $2,507,000 at 95, so most paths have plenty left for a few more years, while the paths that fail were already failing.
Do not read that as comfort, though. The default plan has fixed income that starts at 67, which carries much of the late-life load. Remove it and the same plan scores 64%, a drop that would make the later years matter more. The page on how long your money will last in retirement runs the same balance to several end ages so you can see the shape. It prints no life-expectancy figure, and neither does this page: the Social Security Administration pages we tried to read would not load for us, so any official period life table is something to look up yourself.
Taxes and required withdrawals
The calculator treats a dollar withdrawn as a dollar spent. In real life, money taken from a traditional retirement account may be taxed on the way out, so a bigger gross withdrawal may be needed to cover the same spending. We do not model that and we make no estimate of it.
A related rule can force withdrawals whether or not you need the cash. The IRS page we read on 2026-11-04 defines required minimum distributions as the minimum amounts an owner must withdraw from retirement accounts each year. The same page says the first one is due for the year in which the owner reaches age 73; check the current rule before relying on that age. The model has no such trigger. If your required amount would exceed the spending line you entered, the model neither knows nor adjusts.
Health and long-term care
Healthcare costs usually rise with age, and the spending line in the model does not. We have no cost figures to offer and will not make any up. What we can report is narrow. Medicare.gov states that Medicare does not pay for long-term care, including nursing-home or community care, according to the page we read on 2026-11-04. A reader who is counting on public coverage for that kind of care should check what applies to them.
In simulation terms, a large care cost late in life behaves like the spending shock described above, arriving at a point when the portfolio has little time to recover. The calculator holds spending constant, so it cannot add a late-life cost in one step. A rough way to explore the idea is to raise the spending input and compare the runs, though that raises every year, not only the last ones.
What you actually do when markets fall
The model rebalances by rule once a year, keeps the same mix and spends the same amount in every path, whether the market is up or down. People are not like that. Some spend less after a loss, which helps the portfolio; some change their mix or stop following a plan, which can help or hurt. The simulation includes none of it.
This is the largest blind spot and also the hardest to quantify. We can't tell you how you would act, and neither can a calculator. What you can do is read your score as describing a version of you that never changes course, then consider whether that version resembles you. A plan that depends on never flinching is a different object from one that bends.
The model can be wrong in its own terms
Every result rests on estimates. The means, volatilities and correlations of stocks, bonds and gold come from 55 yearly observations in the default set and 98 in the alternative, a small sample for a question about decades. The sensitivity of the output to those estimates is not small. Lowering the stock return assumption by one point took the default plan from 87% to 80%, by two points to 69%, and by three points to 57%. Raising stock volatility by three points gave 82%, and lowering it gave 93%.
The shape of the model matters as well. Each year's return is drawn independently of the year before, so there are no multi-year trends or slumps beyond what chance produces. Returns follow a lognormal pattern, a smooth bell shape in logarithms, which real markets only approximate. The steps are annual, so anything that happens inside a year is invisible. And random draws bring noise: for the default plan, eight different seeds at 1,000 trials ranged 87% to 89%, at 5,000 trials 87% to 89%, and at 10,000 trials 87% to 88%.
Full details of what the engine does are on the page about how this simulation works, and the checklist at how to pick a Monte Carlo retirement calculator turns these points into questions you can ask of any tool.
What history did that the average hides
Compare the simulation with a plain replay of the past. Take the example of $1,000,000 and a first-year withdrawal of $40,000, repeated each year in real terms, with a 60/40 mix, 30 years, and real returns from the same dataset. Of the 69 start years from 1928 onward that allow a full 30 years, 65 lasted, and the ones that did not began in 1965, 1966, 1968, 1969. The start year that ran dry soonest was 1966. Restricting the starts to 1971 or later, 26 of 26 windows lasted.
The simulated scores for that same example, 30 years, were 95% on the 1971 to 2025 assumptions and 92% on the 1928 to 2025 assumptions. Three numbers, three slightly different answers, all of them built from one dataset. The point is not that one is right. It is that the answer depends on which slice of history sets the assumptions, a choice the score itself does not display. The windows also overlap, which means they are not independent trials, and a handful of years do most of the work.
Where diversification comes in
A model that knows about only a few asset types can only discuss a few. People who worry about the gaps above sometimes ask whether holding a different kind of asset would matter. We can describe what the dataset shows and leave the decision to you.
In the dataset's real (after-inflation) series, the ten years from 1999 to 2008 were the worst ten-year stretch for stocks, with a cumulative return of -32%. Over the same ten years the gold column gained +136% and the bond column +48%. That is one decade, and one decade is a thin basis for a claim.
The same series has a less comfortable stretch. Gold's real value peaked in 1980, fell by 77% to a low in 2001, and did not regain that peak until 2011, which is 31 years under water. Gold had a negative real year in 44% of the years from 1971 on, against 25% for stocks. The gold column is built from yearly prices, an average for the year before 1970, which smooths real swings, and before 1971 gold was fixed under the gold standard and was mostly stable.
So the data shows gold rising sharply in some windows and sitting below an earlier level for decades in another. Whether it raises or lowers a score depends on the return you assume for it, which is why the page on gold in a retirement portfolio runs the calculator both ways. Readers who want to see how one provider describes its gold IRA, including its stated minimum, can open the Augusta Precious Metals gold IRA write-up.
What this can't tell you
Which of these gaps matters most for you is something we can't tell you, because that depends on facts about your life that no form on this site collects. It can't tell you what inflation, taxes or care will cost, how long you will live, or what you would do in a downturn. It can only point at the places where the score is silent and show how far it moved when we nudged a few inputs. Treat the number as a question to carry into a conversation, not an answer to bring away from it. If you would like to experiment, the retirement Monte Carlo calculator on the home page lets you change any input and watch the result move.
Frequently asked questions
Does a high Monte Carlo score account for inflation?
Partly. The calculator uses real returns, so historical inflation is already inside them, and spending stays constant in today's dollars. It does not let inflation run hot or cold for a long stretch, and it does not know whether your own costs rise faster than the average basket.
Are taxes included in the calculator?
No. A withdrawal is treated as spending, with no tax taken out and no required-withdrawal rule applied. If your accounts are taxed on the way out, the gross amount you need may be larger than the spending line, and the model will not show it.
Can a simulation include long-term care costs?
Not in our calculator, which holds spending steady. You can approximate a larger late-life cost by raising spending or shortening the plan and comparing the runs, but that is a rough stand-in, not a care forecast, and we offer no cost figures.
Which blind spot is the biggest?
We can't rank them for you. In our sensitivity runs, the retirement date and a three-point cut in the stock return assumption moved the default plan's score the most, while a longer end age moved it the least. Your own spending and behavior may matter more than either.