What If You'd Retired in 2000? Or 2008?

A different question than "what's the probability"

Most stress-testing tools — including the Monte Carlo simulation behind this app's own Stress Test — answer a probability question: generate many randomized possible sequences of market returns, run your plan through each one, and report the percentage that survive. That's a genuinely useful lens, and it's broad by design — it explores territory that never actually occurred, to give you a sense of the range of things that plausibly could.

A historical back-test asks something narrower and, in a different way, more useful: forget hypothetical markets — what if you'd retired in a specific real year, and lived through the actual sequence of returns that actually happened to real investors? Not a simulated 1966. The real one — the one where the market went nowhere for over a decade while inflation ran hot. Not a modeled 2008. The real crash, the real recovery, the real numbers.

The difference matters because a real historical year carries something a randomized scenario can't: it happened. It has a name, a story, a reason. That makes it concrete and visceral in a way an abstract probability number isn't — "your plan would have survived 1929, 1973, and 2008" lands differently than "your plan has an 87% success rate," even when both statements are pointing at related information.

How it actually works: replaying real sequences, one starting year at a time

The mechanics are straightforward, even though the data behind them runs deep. The engine uses the Shiller dataset — historical U.S. market and inflation data reaching back to 1928 — and treats each year in that range as a possible retirement start date. For each eligible starting year, it takes your plan's actual inputs (savings, spending, contributions, retirement age) and runs them forward using the real returns and real inflation that occurred starting in that specific year, for as long as your plan needs to run.

Someone who "retired" in the back-test's 1966 scenario experiences the real 1966 return, then the real 1967 return, then the real 1968 return, and so on — the actual historical sequence, in the actual order it happened, applied to your numbers. Do that for every eligible starting year back to 1928, and you get a set of outcomes: some starting years your plan comfortably survives, some it depletes early, some it survives but only barely.

Cohort survival: the honest way to report it

It would be tempting — and misleading — to report a back-test as a single dramatic headline: "your plan would have survived the 2008 crash." That's one data point dressed up as an answer. A single famous bad year makes a good headline but a bad measure of resilience, because it only tells you about one path through history, and it's usually not even the hardest one.

The more honest framing is cohort survival: out of every eligible historical starting year in the dataset, what fraction actually got your plan to the end without running out? That number treats 1929 the same as 1955 the same as 1974 the same as 2008 — every real starting year counts once, and the result is a genuine track record across the full sweep of market history the data covers, not a cherry-picked scare story or a cherry-picked reassurance.

A single bad year makes a memorable headline. A cohort survival rate is the actual answer.

This is also why an era can matter more than a single crash. 1966 through the early 1980s is a genuinely brutal stretch by this measure — not because of one dramatic crash, but because weak market returns and persistently high inflation combined for well over a decade, quietly eroding a retiree's purchasing power year after year with no single headline moment to point to. A back-test that only tested famous crash years would miss it entirely. A cohort approach that walks through every starting year in the stretch catches it, because each of those starting years gets its own honest test.

The honest limitation: history only had one path

Here's the tension worth sitting with. A historical back-test is more concrete than Monte Carlo — real years, real numbers, real events — but it's also more limited, for exactly the same reason. History gave us one path through the 20th and early 21st centuries. It didn't give us the thousand alternate paths that could have happened instead. A Monte Carlo simulation explores that wider space of hypothetical futures; a back-test can only walk through the specific years that are actually in the record.

That has a real consequence: a back-test can't price in a market condition that has never occurred before in the dataset — an unprecedented inflation regime, a genuinely new asset class reshaping how portfolios behave, a structurally different economy shaped by demographic shifts the historical record doesn't contain. Past returns are not a guarantee of future returns, and a back-test, however grounded in real events, is still fundamentally a reality check against the past rather than a forecast of what's ahead.

None of that makes the exercise less valuable. It makes it a different kind of evidence than a probability score — evidence about how a plan holds up against the specific hard stretches that have actually occurred, which is worth knowing, alongside the honest acknowledgment that the future doesn't owe us a repeat of any particular decade.

A worked example: 1966 versus 1982

To make "sequence matters, not just the era's average return" concrete, consider two hypothetical retirees with identical plans — same savings, same withdrawal rate, same retirement age — who happen to start in different real historical years.

Same plan. Same withdrawal amount. Same starting balance. The only variable that differs is which real 15-to-20-year stretch of market history the retiree's clock happened to start in — and it's enough to separate a plan that finishes comfortably from one that finishes depleted. That's the whole case for testing against many real starting years instead of trusting a single average or a single remembered "bad year."

Historical back-test versus Monte Carlo: two lenses, not competitors

It's worth being precise about how these two tools relate, because they're easy to conflate. Monte Carlo simulation — covered in more depth in what happens if the market crashes the year you retire — generates many randomized possible return sequences and reports the percentage that survive. It's broader than a back-test: it can explore combinations of good and bad years that have never actually occurred together in the historical record, which is genuinely useful for understanding the full range of plausible risk.

A historical back-test is narrower but grounded in fact: it only tests sequences that actually happened, which means every result it reports corresponds to a real, nameable stretch of history a real retiree lived through. Neither approach is strictly better — they answer different questions. Monte Carlo answers "across a wide range of plausible futures, how often does this plan hold up?" A back-test answers "across the real history we actually have, how often would this plan have held up?" A plan worth trusting should hold up reasonably well under both lenses, not just the more forgiving one.

Try it in the app

Common questions