Time Machine runs your plan through the year-by-year returns that actually followed a real starting year. Not an average of history, and not a simulation drawn from it: one specific stretch, in the order it happened. Retiring in 1966 meant a decade of inflation before any recovery; retiring in 1982 meant the opposite. Same plan, opposite lifetimes.

Inflation is the window's AVERAGE, not a year-by-year path. Returns are replayed one year at a time; inflation is not. That is a measured decision rather than a shortcut - decomposing the two found the inflation LEVEL worth 15% to 19% of the terminal balance while the year-to-year variation netted to roughly nothing, because Social Security's COLA is indexed and tracks it, and nominal returns correlate with inflation over long windows. So 1966's punishing inflation is present in full as a level; what is not modeled is the exact year each print landed.

Everything else about your plan is modeled exactly as it is elsewhere - taxes, Social Security, RMDs, withdrawal order. Only the market returns and the inflation rate are replaced.

Two ways to read a stretch

The toggle under the scrubber decides whether history supplies the returns or only their order.

• As recorded replays the stretch untouched, at the inflation it actually ran. Your expected-return assumptions do not apply, because history supplies them. This answers "what happened".
• At my assumptions keeps the same sequence of good and bad years but shifts the whole record so its long-run return after inflation matches what your plan assumes, then reads it at your own inflation rate. This answers "what if that order of years happened at the returns I am planning on".

Expect the second to be harsher, and read the section below for why: as recorded is systematically kinder to a plan than the plan's own projection, which makes it a weak stress test however alarming the year sounds. Starting in 1966 with a 60/40 mix, the record delivered about 4.1% a year above inflation; shifted onto the default assumptions the same sequence delivers about 1.8%.

The shift is one constant applied across the whole 1872 to 2025 record, not a per-window adjustment. Each window keeps its own character - the point of the mode is that 1966 still looks like 1966, with its crash in the same place - while the general level of returns becomes yours rather than history's. The worst year in a 1966 start moves from about -26% to about -34%; the best moves down too.

The choice is remembered in your browser rather than stored in the plan, so it does not travel with a shared plan and does not mark your results stale. It also lets this page and the simulator beside it disagree, which is deliberate.

Picking a year

Drag the scrubber, or use the shortcut chips above the chart for the years people have heard of. Amber chips mark stretches history made punishing, green ones the kind stretches. That colour describes what markets did next, not how your plan handles it - a well funded plan can come through an amber year comfortably. The line under the chips names what the selected year is known for.

Which years you can pick depends on your plan's length, because a window has to fit entirely inside the record. A 30-year plan can start no later than 1996, so the 2000s chips do not appear for it at all.

Reading the result

• Outcome - whether the money lasted, and if not, the age it ran out and how many years short of your plan's end that was.
• Funded, in the chart's corner - the same funded ratio used elsewhere, measured on this replay rather than on the simulation.
• Ending balance, average annual return, average inflation - each shown against your plan's own figure, because the comparison is the answer. The return is the compound rate the portfolio actually earned over the window, not a start-to-end balance ratio.

The chart is in nominal dollars, and 1966 shows why that matters. A plan replaying 1966 to 1990 as recorded looks like it treads water for years, but inflation over that window averaged close to 6% against the 2 to 3% most plans assume. The balance was not holding steady, it was losing purchasing power while the withdrawals climbed with prices. That is exactly why the sweep measures outcomes in years of spending rather than in dollars.

Why most stretches beat your plan's own projection

This surprises people, and it is worth understanding rather than treating as a contradiction. Two separate things are going on.

The assumptions are forward-looking, not historical. The default 8% for stocks and 4.5% for bonds sit below what markets actually delivered - US stocks returned about 10.9% a year on average since 1872, and 11.9% since 1928. That is a deliberate stance, in line with the forward-looking capital market assumptions most large firms publish, because starting valuations today are high by historical standards. It is not an attempt to describe the past. If you have raised these on the Assumptions page, this half applies to you less, or not at all - but the volatility effect below still does.

A simulated median compounds below the average you enter. Returns vary year to year, and variation costs compound growth: an 8% average with the default volatility produces a median path that grows at about 6.5%, not 8%. Nothing is being taken away from you, this is arithmetic - a portfolio that gains 30% then loses 20% averaged 5% a year but grew 4%.

Put together, the default assumptions give a median path compounding at roughly 2.7% above inflation for a 50/50 retirement allocation. For comparison, the worst 25-year stretch in the record - starting in 1966 - delivered about 2.6% above inflation. Starting in 1929 gave 3.0%, in 2000 about 3.7%, and in 1982 about 8.1%.

So your plan's middle case is calibrated near history's worst case. That is why a plan showing a 95% success rate can come through every stretch on record: the simulation's unlucky tail is worse than anything that has actually happened. Neither view is wrong. The simulation answers "what range of futures should I prepare for", and it prepares you for a future somewhat worse than the past. Time Machine answers "what would the past have done to me", and the past was, on the whole, kinder than the assumptions.

The quickest way to close this gap is the toggle, which is what "At my assumptions" is for: it removes the first half entirely by putting history's returns on your plan's footing, leaving only the sequence. If instead you would rather plan against historical norms than against forward-looking estimates, raise the expected returns - either on the Assumptions page, or in the strip at the top of this page, where the plan line redraws against the same replayed history while you nudge. Raising the inflation rate does the opposite of what you might expect here: returns are entered in nominal terms, so a higher inflation assumption lowers your real growth and widens the gap rather than closing it.

Testing every starting year

One year on its own proves very little. The sweep runs your plan through every complete stretch in the record and paints the result as a band per year. The colours are continuous, not pass/fail: deep red for running out a decade early, through amber for finishing on fumes, to green for finishing with years of spending to spare. The scale is fixed rather than fitted to your plan, so you can compare two plans, or the same plan before and after an edit, and trust that a greener track means a better plan.

Failing years arrive in contiguous bands rather than scattered ticks, because neighbouring windows share the same downturn. That is also the honest caveat: these windows overlap heavily, so 131 of them are nothing like 131 independent tests. Treat the count as a stress test against what has happened, not as odds - and remember that history contains nothing worse than its own worst.

Historical Paths

The Historical Paths tab draws the same sweep as a grid of small charts, one per starting year, each showing your portfolio from that year to the end of the plan. The tiles are laid out a decade to a row, with each year in the column of its last digit, so 1966 sits directly under 1956: a run of bad starting years reads down a column as well as along a row. Each tile is coloured exactly like its band on the track, and a tile that ran out shows the age it did. Click any tile to open that year in the single-year view.

Unlike the single-year chart, the grid is in today's dollars, so the same year can look different in the two views. That is deliberate: every stretch ran at its own inflation, so in nominal dollars a 1970s start and a 1950s start are in different currencies and could not share an axis. Deflating each one by the inflation it ran at (or by your own rate, with At my assumptions) puts them all in today's money, and the tiles share a single scale so their heights can be compared directly. The scale tops out at the 90th percentile of the tiles' peaks, so one runaway stretch cannot flatten the rest; the few taller tiles are cut off at the top, marked with a dashed edge, and the legend says how many.

Historical Outcomes

The Historical Outcomes tab summarizes the same results on two charts that share one starting-year axis. The top one, portfolio longevity, shows the age the money ran out for each starting year that ran out. A year that stayed funded is not given a made-up depletion age: its bar reaches the "funded through end of plan" line, in green. The bottom one shows what was left at the end, in today's dollars like Historical Paths. It uses the same capped scale, so the few largest endings are cut off at the top; each of those bars carries an up arrow, and its tooltip always gives the real figure. Above the charts, a short summary gives how many starting years stayed funded, the historical success rate, and the earliest age any of them ran out. Hover a year on either chart to see it outlined on both, and click it to open that year in the single-year view. The year already open there stays outlined in blue, with a dashed line through both charts, so you keep your place when you switch tabs; the legend names it.

What it would take

After a sweep, Time Machine can search for the smallest uniform change in your recurring spending that moves the plan across the line, and name the year that sets that number. It applies to recurring expenses only: taxes follow from the plan, debt payments are contractual, and one-time expenses are events rather than a spending level.

If your plan already comes through every stretch, the question inverts and it searches for how much more you could spend. That answer is checked against the tightest stretches rather than all of them, which is why the wording says so.

This is the historical cousin of the Spending Optimizer, which asks the same question of 2,000 simulated futures. The two will disagree, sometimes by a lot. That gap is information rather than a bug: one tells you what a distribution of possible futures supports, the other what the recorded past would have allowed.

Where the data comes from

US large-cap stocks, 10-year Treasury bonds, and CPI inflation, 1872 through 2025. 1928 onward comes from Damodaran's published series at NYU Stern. The years before that are derived from Robert Shiller's dataset, with bond returns computed by Damodaran's own method and checked against his published figures across the 98 years where both exist.

Two honest caveats about the older era. Shiller's prices are monthly averages rather than year-end closes, so those years are slightly smoothed compared with the modern ones, and inflation before 1913 comes from a spliced historical index rather than from CPI-U. The 1872 to 1927 window is included because it carries conditions the modern record barely does - deep deflation, and equity losses alongside double-digit inflation - not because it is measured as precisely.

Time Machine vs the other simulation modes

The Historical Market Simulation mode on the main chart also uses this data, but it resamples it: it stitches together contiguous blocks from different eras to build thousands of plausible-but-invented sequences. Time Machine does the opposite. It runs one sequence exactly as it occurred, start to finish. Use the simulation for the odds, and Time Machine for the specific question "what would have happened to me?"

Time Machine is a Plus feature.