A withdrawal rate is a test result, not a rule

What the number actually measures

A withdrawal rate is the amount you take out in the first year of retirement, expressed as a share of the corpus you started with. It is not a forecast of returns, and it is not what you take out in later years.

That last clause is where most of the confusion lives, and it is worth being exact. Under the rule the famous studies actually tested, you fix a rupee amount in year one, raise it each year by inflation, and never look at the portfolio again. The percentage is a label for the starting amount. After that it plays no part.

So the question the rate answers is narrow and testable: given a starting corpus, a set of assets, a length of retirement and a rule for raising withdrawals, how large could the first withdrawal have been without the money running out before the end? The answer is a number produced by a test, on one body of history. It is not a property of money.

Read that way, the widely quoted figure stops being a rule you obey and becomes a result you can question — which conditions produced it, and whether yours resemble them. Its provenance is covered in the FIRE article; this one is about what kind of number it is and how you would derive your own.

Two different rules wear the same number

Ask ten people what a 5% withdrawal rate means — the figure is picked at random here, to keep the argument about the rule rather than the number — and you will get two incompatible answers. One says ₹2.5 lakh out of a ₹50 lakh corpus, rising with inflation every year regardless of what the portfolio does. The other says 5% of whatever the portfolio is worth this year.

These are not two variants of one strategy. They are opposite strategies, and only one of them can run out of money.

The tested rate belongs to row one and to no other row. Quoting it while running row two, or the reverse, is the commonest error on this topic — and it is not a small one, because the two rules trade the same risk in opposite directions. Row one protects your income and risks your capital; row two protects your capital and risks your income. No rule protects both. That trade is the whole subject.

The floor-and-upside row is a structural answer rather than a withdrawal rule, and it is the subject of layering retirement income.

Why the order of returns decides it

Because a withdrawal taken during a fall sells more units to raise the same rupees, and those units are not there for the recovery. Two portfolios can earn the identical set of returns and end at different balances purely on the order they arrived in. Here it is in three years of arithmetic.

Take a corpus of ₹50 lakh, a withdrawal of ₹2.5 lakh at the start of each year, and three annual returns of +20%, +20% and −40%. The figures are arbitrary — chosen because they are easy to redo on paper — and the only thing that changes between the two cases below is the order they arrive in.

• Good years first. ₹50 lakh less ₹2.5 lakh is ₹47.5 lakh; up 20% is ₹57 lakh. Less ₹2.5 lakh is ₹54.5 lakh; up 20% is ₹65.4 lakh. Less ₹2.5 lakh is ₹62.9 lakh; down 40% leaves ₹37.74 lakh.
• Bad year first. ₹50 lakh less ₹2.5 lakh is ₹47.5 lakh; down 40% is ₹28.5 lakh. Less ₹2.5 lakh is ₹26 lakh; up 20% is ₹31.2 lakh. Less ₹2.5 lakh is ₹28.7 lakh; up 20% leaves ₹34.44 lakh.

Same three returns, same three withdrawals, ₹3.3 lakh apart after three years. Now remove the withdrawals and run both again: 1.2 × 1.2 × 0.6 is 0.864 either way, so both end at ₹43.2 lakh exactly. Order only matters when money is moving.

Which way it matters depends on the direction of the flow, and this is the part that gets stated backwards. Money going in makes an early fall useful — the contributions that follow buy at lower prices, so a bad start helps the accumulator. Money coming out reverses the sign: a withdrawal in a fallen market sells more units to raise the same rupees, and those units are not there for the recovery. The same three returns help the accumulator and damage the retiree.

Three years shows a modest gap. A retirement compounds it, because every subsequent year's growth applies to the smaller base — the early loss is not repaid by a later good year, it is merely grown from a lower start. That asymmetry is sequence-of-returns risk, and it is the mechanism a withdrawal-rate study exists to measure.

It follows that a plan tested against an average return has not been tested. The average is the one statistic that discards the ordering, which is the only thing that mattered here. The unit-level view of the same effect is set out in how a systematic withdrawal plan behaves, and the standard structural response to it is the bucket strategy.

What kind of number a tested rate is

It is a minimum, not an average. The method that produces the famous figure is roughly: take a long historical series, start a retirement in every year of it, apply the fixed-real rule, and find the largest starting withdrawal that survived every start — including the worst one. The answer is set by the single worst start date in the record.

Minima behave differently from averages in two ways worth knowing before you lean on one.

A minimum moves when the sample grows. Add one historical episode worse than anything already in the series and the answer falls; add a hundred ordinary ones and it does not move at all. The figure is therefore hostage to whether the worst thing that can happen has already happened inside the window studied.

The windows overlap, so there are fewer of them than there appear to be. A series of N years yields N − H + 1 start dates for a horizon of H years, which sounds like plenty. But those windows share most of their data with each other: a 30-year window starting in one year and the one starting the next year differ by two years out of thirty. The count of genuinely independent windows is N divided by H. For a horizon of thirty years, a market with a few decades of usable history offers one independent window, or one and a bit — not hundreds.

Historical outcomes describe what happened, not what will happen; past performance does not indicate future results. That caution is usually printed as a formality. Here it is the substance of the argument.

The four conditions that change when the number moves

A tested rate carries its conditions with it. Four of them differ enough here that transplanting the answer is not a small approximation. Each is worth naming with its mechanism attached, because "conditions in India are different" on its own tells a reader nothing they can act on.

• The inflation path. The fixed-real rule indexes withdrawals to inflation, so a higher or more variable inflation record means withdrawals grow faster than the tested rule assumed and the corpus is drawn down harder. The gap widens further if your own basket — healthcare heavy in later life — rises faster than the published index.
• The asset mix and its record. The tested rate belongs to a specific equity-and-bond split in a specific market. Both the return series and the volatility series differ here, and a rate derived from one and applied to the other is an assumption rather than a finding.
• The length of the retirement. Thirty years was a modelling choice, not a law. A rate that survives thirty years does not automatically survive forty-five, because failure probability rises with the number of years the corpus must span — which is the arithmetic behind retiring early being a different problem rather than the same one started sooner.
• Tax. A rate tested with no tax drag is a rate for a tax-sheltered account, whatever the study meant it to be — because a rupee withdrawn from a taxable portfolio is not a rupee received. This one is mechanical and gets its own section below.

None of that produces a replacement figure, and this article does not offer one. Offering an Indian number derived the same careless way would be the identical error with a local accent.

Withdrawals are gross; spending is net

To spend ₹5 lakh you must redeem more than ₹5 lakh, because part of what you redeem is taxable gain. The withdrawal rate that matters to your life is the net one; the rate that depletes the corpus is the gross one.

Only the gain embedded in the units sold is taxed, not the whole redemption — and that proportion is not constant across a retirement. Take a unit bought at a per-unit price of ₹100 — the NAV. Redeemed at ₹120, it carries a gain of ₹20, one-sixth of what you received. The same unit redeemed at ₹400 carries a gain of ₹300, three-quarters of what you received. The gain share of a redemption rises as retirement goes on. A constant nominal withdrawal therefore carries a rising tax bill with no rate having changed, and a plan built on the first year's tax drag understates every later year's.

The rates that apply to the two structures a retiree is most likely to be drawing from, under the Income-tax Act 2025:

• Equity-oriented funds — a fund holding more than 65% of proceeds in domestic equity. Gains are long-term after 12 months and taxed at 12.5% above ₹1.25 lakh of such gains in a year, under s.198 (the section formerly numbered 112A). Below that holding period the rate is 20% under s.196.
• Specified mutual funds — those holding more than 65% in debt and money-market instruments, for units acquired on or after 1 April 2023. Section 76 taxes these at slab rates, with the gain always treated as short-term, so waiting longer before redeeming buys nothing.

Two consequences that change the arithmetic and are routinely missed. First, the annual equity threshold is an allowance to be used, not a floor: a redemption schedule that realises gains across years rather than in one large event uses it more than once, and the difference is entirely mechanical. Second, and against the intuition of a retiree with little other income: the s.156(2) rebate — up to ₹60,000 where total income does not exceed ₹12 lakh — cannot shelter special-rate income, which s.156(3) states explicitly, and long-term equity gains are special-rate income. "My total income is small, so there is no tax" is therefore wrong in exactly the case where a retiree is most likely to believe it.

The new regime is the default under s.202. Which regime applies changes the slab arithmetic on any non-capital-gains income sitting alongside these withdrawals, so the comparison has to name a regime to mean anything. Income for the financial year to March 2026 — FY 2025-26 — is still assessed under the repealed 1961 Act, so both numberings are live during the transition.

Deriving your own, without importing anyone's answer

The useful output of all this is not a number to adopt. It is a test to run, with your own inputs. Seven of them, and every one is a decision rather than a lookup.

1. Annual expenses in retirement, not today's — the total is usually not much lower, because what falls in commuting and dependants tends to be met by what rises in health and help. This is the input the corpus calculation turns on.
2. A horizon, chosen from a longevity assumption you are willing to be wrong about in only one direction. An average life expectancy sits in the middle of a distribution, not at its far end, so a plan built to it is a plan a large share of that cohort outlives.
3. The withdrawal rule — pick a row from the table above and commit to it. This choice does more to the outcome than the percentage does.
4. An asset mix, and the return and inflation series you will test it against. Say where the series came from and how long it is; that determines how much independent evidence you actually have.
5. Costs and tax, subtracted before anything else. The expense ratio and the gross-up described above — redeeming more than you spend, because part of the redemption is taxable gain — both come off the top, every year.
6. Every start date, not the average. Run the plan beginning in each year the series allows, and record the outcome of each.
7. Read the worst window. The median start date tells you what a comfortable retirement looked like. The worst tells you whether your plan is a plan.

Two honest limits on that exercise, since it would be inconsistent to argue the above and then oversell the method. It inherits the overlapping-windows problem — running every start date in your series still tests against only a handful of genuinely distinct histories. And a binary survived-or-not result treats ending with a rupee and ending with several times the starting corpus as the same success, which hides everything a person would want to know about the range in between.

The variable that does the most work

Of everything a retiree can change, spending flexibility is the one that acts on the mechanism directly — and it is the input most discussions leave out, because it is not a number in a table.

Sort the available responses by what each one actually changes and the difference is plain. A larger corpus or a different asset mix changes how much damage a given withdrawal does. A cash buffer changes which asset you sell in a fall, which is worth having, but the same rupees still leave the portfolio and the buffer itself has to be refilled. Reducing the withdrawal is the only response that changes the size of the sale that does the damage.

The trade-off is real and should be stated before the technique is admired. Flexibility means accepting a variable income, and the years the rule asks for a cut are the years cutting feels worst. A retiree whose essential spending already consumes most of the withdrawal has no flexibility to offer, whatever the model says — which is why the size of the non-negotiable layer, not the withdrawal rate, is often the binding constraint. That layer is the subject of matching certainty to necessity, and the cash-buffer response is the bucket strategy.

Whichever rule is chosen, the argument for writing it down before retiring is the same as for the bucket refill rule: its entire purpose is to be followed in the year when following it feels wrong, so it has to exist before that year arrives. A rule invented during a drawdown is a decision made during a drawdown.

Running the test rather than arguing about the rate

The mechanism above is arithmetic, but the size of the effect is not — how deep the falls were, and how much the answer moved with the start date, is something you ask of a price history rather than argue about.

FNOTrader's Mutual Funds app runs on the full AMFI NAV history — around 34 million NAV rows — and reports rolling-return distributions across every start date available, alongside maximum drawdown. That gives two of the things a withdrawal test turns on: how much the answer moves with the start date, and how deep the fall was that a withdrawal would have been made into. Reading the worst window rather than the median is the whole discipline.

FNOTrader is not a SEBI-registered investment adviser and this is not retirement advice. What withdrawal rate suits a particular person depends on facts an article does not have.

Common questions

Continue reading

More in Retirement Planning · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article