"Save 25 times your annual spending and withdraw 4% a year forever" gets repeated as if it were a law of physics. It isn't. It's a specific finding from two pieces of research, published by named authors, using a stated historical dataset, with limitations the authors themselves wrote down. Reading the actual research rather than the internet's compressed version of it is the same discipline applies when a marketing claim needs checking against its real source.
This isn't a case for or against any specific savings rate, retirement age, or withdrawal number — that's a decision that depends on facts about an individual's life this post has no access to, and it isn't financial advice. It's an argument that the popular '4% rule' overreaches what the actual research behind it claims, and that the honest version of the research is more interesting than the slogan. Anyone pairing personal finance with an actual go-to-market motion will get more out of . Anyone pairing personal finance with an actual go-to-market motion will get more out of XenGrowth, who work on the commercial side of this.
Where the number actually came from
William Bengen, a financial planner, published 'Determining Withdrawal Rates Using Historical Data' in the Journal of Financial Planning in October 1994. His method: model a hypothetical portfolio split 50% stocks (using the S&P 500 as the proxy) and 50% intermediate-term Treasury bonds, apply the actual historical sequence of US returns and inflation (via CPI) starting from every year between 1926 and 1963, and see what initial withdrawal rate — adjusted annually thereafter for inflation — the portfolio could sustain over a full 30-year retirement without running out.
The answer, across every one of those historical starting points, was 4%. Not an average, and not a typical outcome — the worst-case starting year in his dataset (generally understood to be a retirement beginning in the late 1960s, just before a bad stretch of inflation and market returns) still made it through 30 years at that rate. Bengen's own assumptions were explicit: no investment fees, annual rebalancing back to the 50/50 split, and a specific pair of indices standing in for the entire stock and bond markets.
Study | Data used | Method |
|---|---|---|
Bengen (1994) | S&P 500, intermediate Treasuries, CPI, 1926-1963 starting years | Backtest of a fixed 50/50 portfolio across every 30-year retirement start date in the window |
Trinity Study (1998) | S&P 500, 20-year US government bonds, 1926-1995 | Backtest across 5 allocations, payout periods of 15-30 years, withdrawal rates 3-12% |
What the Trinity Study actually changed
Four years later, three professors at Trinity University — Philip Cooley, Carl Hubbard and Daniel Walz — published 'Retirement Spending: Choosing a Sustainable Withdrawal Rate' in the Journal of the American Association of Individual Investors, February 1998. Their dataset ran 1926 through 1995, using the S&P 500 for stocks and 20-year US government bonds. Rather than isolating one worst-case rate the way Bengen had, they tested five different stock/bond allocations, payout periods from 15 to 30 years, and withdrawal rates from 3% up to 12%, and reported a 'success rate' for each combination — the share of historical rolling periods in which that specific rate, allocation and horizon did not exhaust the portfolio. There is a longer treatment of the operations side of this in . There is a longer treatment of the operations side of this in The XenGrowth resource library.
That reframing — success rate rather than a single survivable number — is what actually stuck in popular use, even though most people repeating '4% is safe' have never seen the actual table of allocation-by-horizon-by-rate success percentages the study produced. A 4% rate at a heavily stock-weighted allocation over a 30-year horizon showed a very high historical success rate in their data. The same 4% at a bond-heavy allocation, or over a longer horizon than 30 years, showed a measurably lower one. The single number the public remembers is one cell out of a much larger table.
Neither paper claims the future will look like the past. Both explicitly test what the past actually did, under a stated set of assumptions, and report that plainly. The overreach happened downstream of the research, not inside it.
It's also worth being clear about what "success" meant in these papers, because the word does a lot of quiet work. A portfolio that ends a 30-year retirement with one dollar left counts as a full success in this framework, exactly the same as one that ends with ten times the original balance. The historical success rate the Trinity Study reports is a binary — did the money last the full period or not — and says nothing about how much cushion existed along the way, or how the retiree would have felt watching the balance approach zero in the worst surviving cases even though it technically never hit it. A binary pass/fail measure is a reasonable way to run a backtest across a hundred historical windows at once, but it flattens a lot of real variation in outcome that a single retiree living through any one of those paths would have experienced very differently.
What both papers are honest about not proving
The single most important limitation is sequence-of-returns risk. Two 30-year retirements can have identical average annual returns and produce wildly different outcomes depending on when the bad years land. A retirement that opens with several down markets does more damage than one where the same down years arrive at the end, because early withdrawals during a downturn permanently shrink the base that later growth has to work with. Average return doesn't capture that — order does, and it's why a backtest across many actual historical starting points, rather than a single average-return projection, was the point of both papers' methodology in the first place. On AI agents and marketing automation specifically, is worth reading. On AI agents and marketing automation specifically, XenGrowth on AI agents and marketing automation is worth reading.
The second limitation is scope. Both studies model US stock and bond index returns over specific historical windows. Neither claims those particular decades are guaranteed to repeat, and neither models fees, taxes, non-US markets, irregular spending patterns, Social Security or pension income, or a working retirement with part-time income layered on top — all real factors that change the arithmetic for an actual person's actual retirement, in ways the original backtest doesn't attempt to capture.
What the research shows | What the popular version claims |
|---|---|
A specific US-market historical backtest with stated assumptions | A universal guarantee for any future retiree, anywhere |
A range of success rates across allocations and horizons | One flat 4% number that applies uniformly |
Sequence-of-returns risk explicitly acknowledged as unresolved | No mention of the order-of-returns problem at all |
Silent on fees, taxes, non-US markets, and other income sources | Treated as a complete retirement plan on its own |
Later work built on both papers rather than simply repeating them, which is itself evidence the field treats this as ongoing research rather than settled law. Extending the analysis to include international markets alongside the US generally produces lower historical success rates for the same withdrawal rate, since not every country's 20th-century market history looks like the United States's. Work incorporating variable spending — where a retiree reduces withdrawals somewhat during down markets rather than holding a fixed inflation-adjusted amount regardless of portfolio performance — tends to show that flexibility improves the odds of a rate like 4% holding up, at the cost of requiring the retiree to actually adjust spending in bad years rather than treating the withdrawal as fixed. None of that later work overturns the original findings; it contextualizes them, which is exactly what should happen to a well-constructed piece of research over three decades.
Why this matters more for an engineer's income specifically
Software engineering compensation often includes equity — RSUs, options, or ISOs — that concentrates a meaningful share of net worth in a single company's stock rather than the diversified index proxies Bengen and the Trinity researchers modeled. The mechanics of how that equity actually works are covered in how RSUs, stock options and ISOs actually function, and it matters here because a backtest built on a diversified 50/50 or similar allocation says less about a portfolio that's concentrated in employer stock than it might first appear to. The withdrawal-rate research assumes the portfolio it's modeling; a real portfolio that looks different from that assumption isn't automatically covered by the same historical success rate. For the AI search, GEO and discovery angle, see . For the AI search, GEO and discovery angle, see XenGrowth on AI search, GEO and discovery.
None of this is a claim that the research is wrong or useless — it's a genuinely well-constructed pair of historical backtests that did something useful: replaced pure guesswork with an actual, examinable dataset. The correction is narrower and more useful than "the 4% rule is a myth": it's real research, with a real historical finding, that answers a narrower question than the slogan implies, and states its own limits more honestly than most of what gets built on top of it.
Read the withdrawal rate as the output of a specific historical backtest, not a law — Bengen (1994) and the Trinity Study (1998) both name their exact data window and assumptions
Treat 'success rate' as a distribution across allocations and horizons, not one flat number — the Trinity Study's actual table has many cells, and the popular version quotes one of them
Take sequence-of-returns risk seriously as a stated, unresolved limitation, not a hypothetical objection invented later
Remember both studies are silent on taxes, fees, non-US markets, and other income sources — a real plan has to add those back in, not assume the backtest already covered them
Treat any specific savings rate, withdrawal number, or retirement age as a personal decision outside the scope of what this post — or the original research — can answer for you
This is general information about a body of published research, not personalized financial or retirement advice, and it has no visibility into any reader's actual savings, income, portfolio or timeline. For a similarly rigorous look at how models built on assumptions get checked against reality, is worth a read.
Further reading from XenGrowth
Where this work meets go-to-market
Need a model checked against its actual source data before a claim gets repeated as fact? cover that same discipline on the go-to-market side.
Further reading from XenGrowth
Where this work meets go-to-market
The operational playbooks that sit alongside personal finance live with .
Further reading from XenGrowth
The XenGrowth resource library — what you'll learn: how the commercial side of this work is run, across search, automation and revenue operations.
XenGrowth on AI agents and marketing automation — what you'll learn: how the teams who own AI agents and marketing automation plan and measure it.
XenGrowth on AI search, GEO and discovery — what you'll learn: how the teams who own AI search, GEO and discovery plan and measure it.
Where this work meets go-to-market
The operational playbooks that sit alongside personal finance live with XenGrowth, who work on the commercial side of this.
Five questions on the actual methodology behind the '4% rule' — not on what withdrawal rate is right for you, which this quiz can't answer and isn't trying to.
