Forty years of advice about putting money in, and almost none about taking it out. This is the other half: how much you can actually withdraw, which account to pull from first, what the IRS forces you to take at 73 or 75, and the one risk that ends more retirements than bad investing ever has.
While you are saving, a market crash is a sale. You are buying every month, so cheap prices are good news you will not feel for twenty years. This is why "stay the course" is genuinely good advice during accumulation, and why it is repeated so often it stops sounding like advice at all.
The day you start withdrawing, that reverses completely. You are now a seller every month. A crash means selling more shares to raise the same rent money, and those extra shares are gone — they are not there to recover when the market does. The recovery happens to a smaller pile than the one that fell.
This is sequence-of-returns risk, and it is the single most important idea on this page. It says that the order of your returns, not their average, decides whether the money lasts.
The clean illustration comes from the fact that a portfolio that falls 50% and then doubles ends exactly where it started — as long as nothing is being taken out. Add withdrawals and the symmetry breaks. Take money out during the 50% year and there is less capital left to participate in the doubling, so the two orderings no longer land in the same place. Run it far enough and one of them reaches zero while the other is still comfortable, on identical average returns.
You can watch this happen with your own numbers. The retirement withdrawal calculator runs your plan twice — once smoothly, once with a single bad year bolted onto the front and nothing else changed. On a $500,000 balance at 65 taking $20,000 a year, moving one 25% loss to the front of the sequence moves the finish line by decades. Same withdrawals. Same expected return from year four onwards. Different retirement.
The cash-reserve or "bucket" approach was built by planner Harold Evensky, used at his firm from 1985 and written up formally in 1997. The point is not the return on the cash — the cash loses to inflation and that is accepted. The point is that a bad first year becomes something you wait out instead of something you sell into. Evensky's own study put the improvement in 30-year plan survival at up to about six percentage points.
A plan you can trim by 10% for two years is structurally safer than a lower fixed number you treat as untouchable. This is the whole idea behind the guardrail systems in the next section. Flexibility is worth more than precision, because precision was never available.
Social Security, a pension, and in some cases an annuity are not investments — they are income that keeps arriving no matter what the market does or how long you live. The more of your essential spending they cover, the less the market gets to vote on whether you keep the lights on.
Almost everyone has heard of the 4% rule. Almost nobody has heard what it actually is, which matters, because the misunderstanding runs in both directions — some people treat it as a promise and others dismiss it as a myth, and it is neither.
William Bengen published Determining Withdrawal Rates Using Historical Data in the Journal of Financial Planning in October 1994. He took US market history from 1926 to 1992, ran rolling 30-year retirements through it, and asked a specific question: what is the highest first-year withdrawal rate, raised for inflation each year afterwards, that would have survived every single starting year in the sample — including the worst one there is?
The answer was about 4%. Bengen called it the SAFEMAX.
Read that definition again, because it is the part that gets lost. 4% was never the typical result. It was the number that survived the worst 30 years American markets have produced. In the large majority of historical starting years, a retiree taking 4% died with more money than they retired with — often far more. The Trinity Study (Cooley, Hubbard and Walz, AAII Journal, February 1998) made this visible by publishing a whole matrix instead of a single figure: withdrawal rates from 3% to 12%, five stock and bond mixes, and horizons of 15, 20, 25 and 30 years, with the historical success rate for each combination.
Bengen has revised his own number upward twice since. He raised it to roughly 4.5% in the mid-2000s, and in his 2025 book he puts it at 4.7% using a more diversified portfolio than the 50/50 stock-and-bond split he originally tested.
And the twentieth-century US was one of the best-performing equity markets on earth, which is not a coincidence you can invest in. Javier Estrada tested the same 4% withdrawal against 1900–2019 data across 22 countries: it failed 4.4% of the time in the United States, up to 67% of the time in Italy, and about 22% of the time for a world-market portfolio. The rule is not wrong so much as it is a statement about one country's past.
Retire at 55 and you may need 40. Retire at 75 and 20 may be plenty. The horizon is an input, not a constant, and it moves the answer a lot.
Bengen tested raw index returns. Michael Kitces' work on the question found that a full percentage point of annual investment and advisory cost lowers the sustainable withdrawal rate by roughly 0.4 percentage points — less than the full point people assume, but not nothing, and it comes straight off the top of a number that is already a worst case.
David Blanchett's 2014 research in the Journal of Financial Planning found real household spending falls through retirement at roughly 1% a year, in a shape he called the retirement spending smile: a faster decline early, a trough in the middle, then a modest late-life rise driven by healthcare that still sits below where it started. Government survey data points the same way — Bureau of Labor Statistics figures for 2024 show households headed by someone aged 65–74 spending an average of $65,354 a year against $55,834 for households 75 and older.
Guardrails. Jonathan Guyton and William Klinger published a set of decision rules in the Journal of Financial Planning in March 2006: instead of one fixed withdrawal, you set boundaries and cut spending when the portfolio drops through the lower one, raise it when it climbs through the upper one. Their own results claim initial withdrawal rates in the region of 5.2% to 6.2% for portfolios holding at least 65% equities — bought entirely with the willingness to take a real spending cut in a bad year.
The RMD method. Take the balance and divide it by the IRS life-expectancy divisor for your age. It is crude, it produces income that swings with the market, and it has one unarguable property: because you are always taking a fraction of what is left, it can never fully empty the account. Some people use the required minimum distribution as the whole withdrawal policy rather than as a floor.
Annuitising part of it. A single-premium immediate annuity converts a lump sum into income for life. What you are buying is not a return — it is the pooling of longevity risk, so that living to 100 becomes the insurer's problem rather than yours. Economists have known since Yaari's 1965 paper that for someone with no wish to leave money behind, this is theoretically the right thing to do, and the fact that so few people do it is known in the literature as the annuity puzzle. Payout rates differ meaningfully between insurers for identical age and amount, so this is a quote-comparison purchase, not a relationship purchase.
Most people finish their working life with money in three different tax buckets, and which one a withdrawal comes out of changes what it costs.
You already paid tax on the money going in. Selling triggers capital gains tax on the growth only, and long-term gains have their own lower rate schedule with a 0% band at the bottom: for 2026 that band runs to $49,450 of taxable income for a single filer and $98,900 for married filing jointly. Anything still in this account when you die gets a step-up in basis, meaning the built-in gain disappears for your heirs.
Nothing has been taxed yet. Every dollar out is ordinary income at your marginal rate, and this is the bucket the IRS eventually forces open with required minimum distributions.
Already taxed, and qualified withdrawals are tax-free. Roth IRAs have no required distributions while the original owner is alive, and since 2024 designated Roth accounts inside a 401(k) or 403(b) do not either.
The textbook sequence is taxable first, tax-deferred second, Roth last. The logic is sound: spend the account with the smallest tax drag first and leave the one that will never be taxed again to compound longest.
The problem is that following it mechanically wastes the most valuable years you will ever have.
Look at what happens to a typical person's income between retiring and turning 73 or 75. The salary has stopped. Social Security may not have started. Required minimum distributions have not begun. For many people these are the lowest-income years of their entire adult life — and if you spend them living off a taxable account, you report almost no income and leave your low brackets completely empty.
Those brackets do not roll over. The dollars you did not pull at a low rate are still sitting in the tax-deferred account, growing, until the IRS forces them out later at a higher rate — possibly large enough to make more of your Social Security taxable and to trigger a Medicare surcharge two years afterwards.
The alternative is to fill those brackets deliberately: take tax-deferred withdrawals, or convert part of the balance to Roth, up to the top of a low bracket each year. For 2026 the standard deduction alone is $16,100 for a single filer and $32,200 for married filing jointly, with an extra $2,050 for a single filer aged 65 or older and $1,650 per qualifying spouse otherwise, and the 12% bracket runs to $50,400 of taxable income for a single filer. That is a meaningful amount of income a year that can be realized at a very low rate, every year, if you are paying attention.
Retirement law is a series of thresholds attached to birthdays. Miss one and you either pay a penalty you did not need to pay or give up an option you did not know you had. Here is the whole sequence.
Qualified public safety employees — including firefighters — can take distributions from a governmental plan without the 10% early-distribution tax at age 50 or after 25 years of service under the plan, whichever comes first. Employer plans only; this exception has never applied to IRAs.
If you separate from service in or after the calendar year you turn 55, distributions from that employer's plan escape the 10% additional tax. Two traps: it does not apply to an IRA, and rolling the 401(k) into an IRA hands the exception back. If you are retiring in your fifties and might need this money, leave it in the plan first and ask questions second.
A schedule of equal payments, calculated by IRS-approved methods, avoids the 10% at any age. The catch is severe: modify or stop the series before the later of five years or age 59½ and the penalty is applied retroactively to every payment in the series, with interest for the deferral. One switch to the RMD method is permitted. This is a commitment, not a tap.
From here, withdrawals from any retirement account are free of the extra 10%. Income tax is still owed on anything pre-tax — the penalty going away is not the same as the tax going away, and this is the most common misreading of this birthday.
And the most expensive. For anyone with a full retirement age of 67, claiming at 62 cuts the benefit permanently by 30% — a $1,000 benefit becomes $700, for life. If you are still working, the earnings test also withholds $1 of benefit for every $2 earned above $24,480 in 2026. Those withheld benefits are not lost: at full retirement age the benefit is recalculated to credit the months that were withheld.
Your initial enrollment period runs seven months: the three months before your birthday month, the month itself, and the three after. The 2026 standard Part B premium is $202.90 a month with a $283 annual deductible. Higher earners pay an income-related surcharge on top — see the next section, because it is calculated on a tax return from two years earlier.
67 for everyone born in 1960 or later; earlier birth years have their own slightly lower figure. This is the age at which you get 100% of your calculated benefit and the earnings test stops applying entirely. In the year you reach it, the test is looser: $1 withheld for every $3 above $65,160 in 2026, counting only earnings before the month you reach full retirement age.
Every year you delay past full retirement age adds 8% to the benefit, and that stops accruing at 70. There is no reason at all to delay past 70 — a claim filed at 71 is simply a year of payments you did not collect. 70 is also, not coincidentally, the strongest longevity insurance most people can buy, and it costs nothing but patience.
From 70½ you can send money directly from an IRA to a charity — up to $111,000 for 2026, a figure that is now indexed to inflation. It never appears in your income at all, which is better than a deduction, and once RMDs begin a QCD counts towards satisfying them.
The next section is entirely about this. The short version: the IRS stops letting you decide.
SECURE 2.0 moved the age again for younger cohorts. Two extra years of untouched compounding sounds like a gift; it also means two more years of balance to be forced out later, which is exactly why the gap years in Section 03 matter.
Every dollar in a traditional 401(k) or IRA is money the government let you keep untaxed on the understanding that it would eventually be taxed. Required minimum distributions are the government collecting on that understanding, on its schedule rather than yours.
Age 73 if you were born between 1951 and 1959. Age 75 if you were born in 1960 or later. That is the SECURE 2.0 schedule, and there is nothing to elect — it follows from your birth year.
Your first distribution can be delayed to April 1 of the year after you reach that age. This is the single most common expensive mistake in this whole area, because delaying does not skip a year — it stacks two distributions into one tax year, which can push you into a higher bracket, make more of your Social Security taxable, and trigger a Medicare surcharge on top. Every distribution after the first is due by 31 December.
Take the account balance on 31 December of the previous year and divide it by the distribution period for your age from the IRS Uniform Lifetime Table. Those divisors are set in the Treasury regulations and reproduced in IRS Publication 590-B:
About 3.8% of the balance. A $500,000 IRA has to distribute roughly $18,900.
About 5.0%. The required percentage climbs every year for the rest of your life.
About 8.2%, which by then is likely well above what you actually want to spend.
A different table — the Joint Life and Last Survivor Table — applies instead if your sole beneficiary is a spouse more than ten years younger than you. It produces a larger divisor and therefore a smaller required distribution.
A missed or short distribution carries a 25% excise tax on the amount you failed to take, cut to 10% if you correct it inside the two-year correction window. SECURE 2.0 reduced this from the old 50%. The distribution itself is still owed on top.
A Roth IRA has never had required distributions during the original owner's lifetime, and since 2024 designated Roth accounts inside a 401(k) or 403(b) do not either. Every dollar you move to the Roth side before your RMD age is a dollar that never joins this schedule.
If you are still employed past your RMD age, are not a 5%-or-greater owner of the business, and the plan document allows it, you can defer distributions from that employer's plan until you actually retire. There is no equivalent for an IRA.
From 70½, sending money directly from an IRA to a qualifying charity keeps it out of your income entirely and counts towards the required distribution — up to $111,000 in 2026. For anyone who already donates, this is strictly better than writing a check and claiming a deduction.
Most non-spouse beneficiaries now have to empty an inherited account within ten years. Under the final regulations, if the original owner had already reached their required beginning date, the beneficiary also has to take an annual distribution in years one through nine — first applicable for the 2025 distribution year. Surviving spouses, minor children of the owner, disabled and chronically ill beneficiaries, and anyone not more than ten years younger than the deceased are treated differently.
Covered in Section 01, and it belongs first here too. The defense is not forecasting. It is holding a year or two of spending somewhere that cannot fall, and being willing to trim in a bad year.
People plan to life expectancy, which is wrong roughly half the time by construction — half of any group outlives its own average. CDC data for 2023 puts life expectancy at age 65 at a further 18.2 years for men and 20.7 for women. But the tail is what matters for planning: on the Social Security Administration's 2023 period life table, about 24% of 65-year-old men and 35% of 65-year-old women are still alive at 90. For a couple, the relevant question is not whether either of you reaches your own life expectancy — it is how long the second of you lives, which is materially longer than either figure alone.
Longevity is the risk that makes the others worse, because it stretches every other exposure. It is also the only one you can transfer: delaying Social Security to 70 and, for some people, buying a plain immediate annuity are both purchases of income that cannot run out.
Two separate problems here, and the second one surprises people.
The first is coverage before 65. If you retire at 60, you have five years to fund before Medicare, and marketplace coverage is the usual route. The subsidy rules changed for 2026 — the enhanced premium tax credits that ran from 2021 expired at the end of 2025 and the older income cap on eligibility is back in play. This is exactly the sort of rule that moves, so check healthcare.gov against your own numbers rather than planning from anything you read a year ago, including this.
The second is IRMAA. Above certain income levels, Medicare charges an income-related monthly adjustment on both Part B and Part D. For 2026 the first threshold sits at $109,000 for a single filer and $218,000 for married filing jointly, and it is a cliff rather than a slope — one dollar over moves you to the next tier for the whole year. And it is calculated on the tax return from two years earlier, so 2026's surcharge comes from your 2024 return. This is why a large Roth conversion or a stacked double RMD has a delayed cost most people never see coming. If your income dropped because of a life-changing event such as retiring itself, Form SSA-44 exists to have the surcharge reconsidered.
Ordinary inflation is the visible half: at 2.5% a year, something costing $1,000 today costs about $2,100 in thirty years, which is why a withdrawal plan has to be defined in real terms.
The invisible half is that some tax thresholds do not move at all. The income levels at which Social Security benefits become taxable — $25,000 and $34,000 for single filers, $32,000 and $44,000 for married filing jointly — were set in 1984 and 1994 and have never been indexed. Every year of inflation pulls more people over them. A separate temporary deduction for people 65 and over, worth $6,000 per qualifying individual and running for tax years 2025 through 2028, was added by the 2025 federal tax law and phases out above $75,000 of modified adjusted gross income for a single filer and $150,000 for a joint return. Note what it is: a deduction stacked on top of the standard deduction, not an exemption of Social Security from tax. The taxation formula and those unindexed thresholds are unchanged.