The Limits of a Safe Withdrawal Rate

The question, and why it has an appealing answer

Somebody who has built a corpus wants to know one thing: how much can be taken out each year without running out.

A single percentage would answer it, and one is widely quoted. Its origin is a body of research on United States market history: test every historical starting year, withdraw a fixed amount rising with inflation, and find the largest rate that survived every thirty-year period on record. That is a genuine piece of work and the method is sound.

The trouble is what happens when the number travels. It gets repeated without the market it came from, without the period, and without the several conditions attached to it. This page is about whether that number — or any number of its kind — can be justified from Indian data.

We cannot produce one, and the reason is worth showing rather than asserting.

What the method needs

The method's strength is entirely in its sample. To find a rate that survived every period, you need many periods, each as long as a retirement, and ideally not overlapping each other so heavily that they are really one observation wearing several hats.

Three requirements, then. Complete runs of data at least as long as the retirement being planned. Enough distinct starting points that the answer is not an accident of one. And returns measured in what the money actually buys, because a retiree's spending rises with prices and a rate computed before inflation answers a different question.

Set against those requirements, here is what the Indian record we hold contains.

What the record actually contains

A retirement lasting	Start dates available	After inflation
10 years	17.2 years’ worth	3.5 years’ worth
15 years	12.2 years’ worth	none
20 years	7.2 years’ worth	none
25 years	2.2 years’ worth	none
30 years	none	none

The bottom row is the answer to this article's question. For a thirty-year retirement — which is what a person retiring in their early sixties should be planning for — the number of complete periods in the Indian data is none. Not too few to be reliable. None. The record is shorter than the thing being tested.

Move up the table and it does not improve much. A twenty-five-year horizon leaves 2.2 years’ worth of start dates, meaning every period tested overlaps almost entirely with every other — it is effectively one observation, not a distribution. And the last column, which is the one that matters because it is measured in purchasing power, says none for every horizon beyond ten years. Indian consumer price data on the basis we hold begins in 2013.

So the honest position is not that an Indian safe withdrawal rate is uncertain. It is that the calculation cannot be performed. Any figure presented as one has either used a different market, extended a short record by assumption, or worked before inflation — and each of those is a different number from the one the reader thinks they are getting.

What the data does support

Something can still be shown, and it is more useful than a rate.

The sequence-risk experiment established that the order of returns decides a retiree's outcome, using a rearrangement of the same 26 years rather than an inference about the future. Running that same rearrangement across a range of withdrawal rates shows how quickly the risk grows.

Taking out each year	Orders in which the money ran out	Typical amount left
3%	0.0%	₹1879.31 lakh
4%	0.2%	₹1666.85 lakh
5%	1.7%	₹1454.40 lakh
6%	5.0%	₹1241.94 lakh
7%	9.8%	₹1029.49 lakh
8%	16.4%	₹817.03 lakh

Read this as a demonstration and not as a table of safe rates — the caption says so and it is worth repeating, because a table of percentages next to a column headed "ran out" is exactly the shape of thing that gets screenshotted without its caveats.

Three reasons it is not a safe-rate table. The orderings are shuffles, which destroys any tendency of markets to rebound after falls, so it includes sequences markets would rarely produce. The run is 26 years rather than thirty, so it is a shorter and easier test than a real retirement. And the withdrawal is held flat instead of rising with prices, which flatters every row.

All three biases point the same way. The real risk at any given rate is higher than this table shows, and the table already shows the risk rising steeply.

What the quoted number leaves out even where it applies

Even in the market it was derived from, the familiar figure carries conditions that rarely travel with it, and each one moves the answer.

It assumes a particular mix of shares and bonds, held through the whole retirement, and rebalanced. A different mix gives a different rate. It assumes a fixed real withdrawal — the retiree takes the same purchasing power every year regardless of what the portfolio has done, which is precisely what a sensible person would not do. It assumes no fees, and fees come straight off the sustainable rate. It assumes a fixed thirty-year horizon, so it says nothing about someone who lives longer. And it reports the rate that survived the worst historical period, meaning that in most periods the retiree died with a very large unspent balance — a success by the test's definition and a failure by most people's.

The tax treatment differs by country too, and it matters here: money withdrawn from an Indian equity portfolio is taxed on realised gains, so a given spending requirement needs a larger withdrawal than the headline rate implies.

What to do instead

Lakshmi is living off her corpus now, and she cannot wait for a better dataset. The absence of a reliable rate does not leave her without a method — it leaves her with a different one, and arguably a better one.

Start from spending, not from a percentage. What does her year actually cost, split into what must be paid and what she would like to pay? A rate applied to a corpus produces a number with no relationship to her life. Working from the essential total tells her what the portfolio genuinely has to deliver, and that is the figure worth protecting.

Cover the essential part with income that does not depend on markets, as far as it can be covered, which is the argument in retirement income flooring. What that floor does not reach, the portfolio funds — and the portfolio's job is then the discretionary part, where a bad year can be absorbed by spending less rather than by selling more.

Hold several years of spending outside equity, so that an early fall is met from the stable part. That is the bucket structure, and the sequence-risk finding is its justification.

Make the withdrawal respond to the portfolio. A rule that takes less after a bad year — even a crude one, like skipping the annual increase — breaks the mechanism that does the damage, because the damage comes from selling into a decline. A retiree with any flexibility at all is in a fundamentally safer position than the fixed-withdrawal assumption behind every published rate.

And review it annually rather than setting it once. A withdrawal rate chosen at retirement and never revisited is a thirty-year forecast. Checked each year against the actual balance and the actual spending, it becomes a series of one-year decisions, each made with information the original choice did not have.

What to take away

The safe-withdrawal-rate method needs many long, distinct, inflation-adjusted periods. The Indian record we hold contains no complete thirty-year period at all, only about 2.2 years’ worth of start dates at twenty-five years, and nothing in real terms beyond ten. The calculation cannot be done here, and a figure offered as an Indian safe rate has come from somewhere other than Indian data.

What can be shown is that the risk rises steeply with the rate and that the order of returns decides the outcome — and that every simplification in the demonstration makes it kinder than reality.

So do not plan around a rate. Work from what the year costs, floor the essential part with income that does not depend on markets, keep several years of spending out of equities, keep the ability to take less after a bad year, and check it every year. That is a method that works without the number nobody can supply.