Most calculators project a single average return. This one runs 3,000 complete market histories and shows where your plan lands across all of them — including the bad ones.
The reason one number isn't enough comes down to timing. Picture two people with the same savings, the same plan, and the same amount drawn each year. One retires in 2000, just before two long downturns; the other in 2009, into the recovery. Over a full career their markets average out to about the same return — but the first spends the early years selling into falling prices and never quite recovers, while the second barely has to worry. Same plan, different decade.
A calculator that assumes one fixed return — say 7% a year — quietly treats you as the average of those two people. But no one retires into an average. You retire into one particular sequence of years, and the order they arrive in matters as much as the numbers. Running 3,000 histories is how the model accounts for that: it tests your plan against good sequences and bad ones, early crashes and late ones, rather than betting on a single tidy path.
An 85% result means your plan lasted to the end of your planning horizon in 85 of every 100 simulated histories, with each year's spending covered along the way. It isn't a promise about your particular future — it's a count of how many possible futures, out of thousands, your plan held up against.
In October 1987, stock markets fell 22% in a single day. Under a standard bell curve, that's an event so unlikely it shouldn't happen once in the lifetime of the universe. It happened on a Tuesday.[1]
The bell curve treats years like that as nearly impossible. But −30% years and +40% years occur more often than a normal distribution allows, and they tend to cluster. Two bad years in a row don't feel twice as hard as one — they feel far harder, because you're drawing down assets at a discount while waiting for a recovery that hasn't arrived. A model that underweights those years will tell you a plan is sturdier than it is.
So the model draws its returns from a heavier-tailed shape — a Student's t-distribution with 4 degrees of freedom — that treats bad years as less of a surprise.[2] The practical effect is that the 3,000 histories include the kind of sequences that actually happen: a −31% shock early in retirement, a five-year grind that starts the week you stop working, a single bad year that several decent ones still don't fully undo.
At 4 degrees of freedom the distribution still has finite variance (which requires more than 2), so the math stays well-behaved — but the tails are heavy enough to match the empirical frequency of extreme annual returns in long-run equity data. A plan that survives 85% of these fat-tailed sequences has been tested against conditions a normal-distribution model barely considers. Two tools can show the same 85% and be testing against very different histories.
Because all three run on the same engine, they check each other. The required nest egg from the second question, used as starting capital in the first, should reproduce the target confidence. When the three don't line up, the cause is almost always a mismatch in the inputs, not the model.
The analysis also computes 75% and 90% so you can see how far the answer moves between positions. Planning at 90% usually takes more capital, a lower spend, or a later start than the same plan at 75%. The model prices that difference; it doesn't decide for you how cautious to be.
What the threshold doesn't mean. A 15% non-survival rate is not a 15% probability that your portfolio is depleted. It's the share of the model's simulated sequences in which the plan didn't last. Real outcomes depend on things the model can't see — spending adjustments, work flexibility, inheritance. Most people don't hold spending perfectly fixed through a long downturn; they adapt, and the model can't capture that.
The threshold also says nothing about which sequences are likely. The model treats every simulated path as equally plausible. Some of the failing 15% reflect conditions that may be extremely rare; others reflect conditions that have already happened in living memory. The model doesn't tell them apart.
One thing worth knowing before you compare this number to another tool's: because the model uses fat-tailed returns, its probabilities run more conservative than most consumer calculators. Two tools can both show 80% and be testing against very different conditions. If a number here looks lower than you expected — lower than another tool, or lower than your advisor's — that gap is worth examining rather than assuming one side is simply wrong.
Starting balance and contributions. The model grows your current savings at the pre-retirement return, adds contributions each year until retirement, then stops. Contributions are a fixed annual dollar amount — not inflating. For couples where both partners work, the working spouse's half continues as a portfolio inflow after the primary partner retires, reducing what the portfolio has to fund on its own.
One thing worth being explicit about: the confidence number the model shows is a conditional result. It answers the question "given that your savings reach the projected balance at retirement, how does the plan fare from there?" The 3,000 simulated sequences cover the drawdown phase — retirement through the planning horizon — not the accumulation phase that gets you there. Pre-retirement markets can run ahead of or behind the modeled pace. If you arrive at retirement with a larger or smaller nest egg than projected, the plan's odds shift accordingly. This is the model's second-largest simplification, after taxes. It's one reason the output is most useful as a range of scenarios to stress-test, not a single forecast to optimize toward.
Expected return and volatility. Return runs at three levels: Conservative (5%), Default (7%), Aggressive (9%) — all nominal, pre-retirement. Volatility is held fixed at 12% and isn't user-adjustable. At retirement, the model lowers the expected return by 1.2 points and volatility by 2.0 points, reflecting the shift toward a less aggressive allocation most investors make in drawdown. A bit over a percentage point of return, compounded across thirty years, isn't a rounding error. Run all three return settings before anchoring on any single output.
Inflation. The rate at which spending grows: Conservative 4%, Default 3%, Aggressive 2%. Applied uniformly to gross spending, Social Security, and retirement income each year. The spread between 2% and 4% over thirty years of retirement isn't a different variant of the same plan. It's a different plan.
Target spend. Monthly spending in today's dollars, after tax — the take-home lifestyle cost you actually live on, not the pre-tax portfolio draw. Each year the model estimates the federal income tax on your income stack — the portfolio withdrawal and other income as ordinary income, plus the taxable portion of Social Security under the provisional-income rule, less the standard deduction, through the marginal brackets — and grosses the withdrawal up to net your target. Social Security and retirement income are modeled as inflows that reduce what the portfolio funds. The model doesn't have a view on whether the number is achievable. It reports what the simulation found.
Planning horizon. The age the portfolio is modeled through. Paths that exhaust before this age count as failures; paths that survive count as successes. The model has no view on longevity — it runs to the boundary either way. What changes with a longer horizon is the fraction of paths that make it.
Social Security. Monthly benefit in today's dollars at your elected claiming age. The model inflates to nominal at the claim date and applies SSA delay factors: 70% of the full retirement age benefit at 62, 100% at 67, 124% at 70.[3] Waiting until 70 is one of the largest single-decision effects available in the model — and one of the few that doesn't require saving more.
Retirement income. Other contractual monthly income beyond Social Security — pension, annuity income, rental income. Modeled as an inflation-adjusted inflow each year. Even a modest fixed income stream changes what the portfolio has to do.
One more thing worth knowing: the same inputs always produce the same outputs. The simulation uses a deterministic random number generator seeded from your inputs. There's no randomness between sessions — only between simulated paths.
Taxes. The model estimates federal income tax each year — marginal brackets, the standard deduction, and the provisional-income taxation of Social Security — using federal tax figures dated in the report. It does not model state or local income tax, Medicare (IRMAA) surcharges, the difference between ordinary income and capital gains, account types (taxable, traditional, Roth), required minimum distributions, or the tax change when one spouse outlives the other. Every portfolio withdrawal is treated as ordinary income — so for someone drawing largely from Roth or low-basis taxable accounts the estimate runs high, and for high earners in a high-tax state it runs low. Treat it as a sound federal estimate, not a tax plan.
Investment fees. Expense ratios and advisor fees reduce effective returns without appearing in the simulation. A 1% annual fee compounded over 30 years can reduce ending portfolio value by 25% or more. If your expected return is 7% gross and your actual costs are 0.75%, the effective input the model would use is closer to 6.25%.
Behavioral risk. The model assumes you hold through every simulated sequence, including the ones that lose 35% in year two of retirement. In practice, most investors don't hold perfectly through extended drawdowns. The gap between modeled returns and actual investor returns — caused by timing decisions, panic selling, and drift — has historically been material.[4] The model cannot quantify your own behavioral risk.
Healthcare and long-term care. Medical cost inflation has consistently exceeded general CPI. Long-term care costs — home care, assisted living, memory care — are among the largest unmodeled risks in a retirement plan and can restructure a portfolio in ways the simulation doesn't see. The model applies one uniform inflation rate to all spending.
Variable spending. The headline assumes real spending is flat across the planning horizon. Research suggests spending isn't flat: it tends to ease through mid-retirement as physical activity slows, then rise again in the latest years as health and care costs grow. This U-shaped path is sometimes called the retirement spending "smile" (the go-go, slow-go, no-go phases).[5] Flat spending is a conservative baseline — it assumes no mid-retirement dip, and it does not assume late-life care costs are any lower than mid-life spending. The report models the smile explicitly as a what-if scenario.
Longevity and survivorship. Every path runs to the same fixed planning horizon, with both people in a couple assumed alive the whole way through. The model doesn't draw on mortality tables, doesn't model the chance one spouse outlives the other, and doesn't step Social Security down to a survivor benefit. A real plan has to weigh how long the money needs to last against how long it's likely to be needed.
One asset, not a portfolio. The simulation uses a single aggregate return and volatility — not a mix of asset classes with their own correlations, drift, and rebalancing. Allocation enters only as the small return-and-volatility reduction the model applies at retirement. Two portfolios with the same expected return but different compositions behave identically here; in practice they don't.