Building a corpus number honestly

Start from what you spend, not what you earn

Estimate your annual spending in retirement, inflate it to the year you retire, then find the amount that would fund an inflation-indexed income for as long as you need it — discounting at your return after inflation, never before.

Four steps, and the first one is where most calculations go wrong before any arithmetic happens. The common shortcut is to take a share of current income — 70% is the number people repeat — and call that retirement spending. Income is the wrong base, because income is not spending. It is spending plus tax plus debt repayment plus saving that stops the day you retire.

Work through a household earning ₹20 lakh a year, of which ₹5 lakh goes into investments and ₹3 lakh into a home loan that finishes before retirement. Spending is at most ₹12 lakh, and less once tax is taken out. A 70% rule asks for ₹14 lakh a year. That gap is not a rounding difference: at the annuity factor derived below, ₹2 lakh a year of phantom spending is roughly ₹40 lakh of extra corpus in today's money, bought to fund contributions you will no longer be making.

The error runs the other way too, which is why the fix is a line-item estimate rather than a different fraction. Commuting, work clothes, dependants and the loan stop. Healthcare, paid help and the premium on cover an employer used to provide start or rise. Netted, retirement spending for most households lands close to current spending rather than far below it — but "close to" is a finding the line items produce, not an assumption to start from.

One practical note on the estimate itself. Build it in today's prices, from what you actually spend now, and keep it there for the moment. Mixing today's prices with a future rupee is the error the next section is about, and the cleanest defence is to know at every step which of the two you are holding.

Two consistent recipes, and the mix that halves the answer

Everything from here rests on one rule: a rate and a price must belong to the same year. A nominal return is a rate quoted in future rupees. Today's spending is a price quoted in today's rupees. Combine them and the answer is not approximately right — it is the answer to a question nobody asked.

Take five illustrative inputs, and substitute your own for every one of them: annual spending today of ₹9 lakh, 20 years to retirement, a 30-year retirement, inflation of 6% and a nominal return of 9%. Those last two are assumptions, not forecasts; nothing here claims to know either.

From them, three intermediate figures. Inflation over 20 years multiplies prices by 1.06²⁰, or 3.21, so ₹9 lakh of spending becomes ₹28.9 lakh in the first year of retirement. The return after inflation — the real return — is 2.83%, derived in the next section. And the amount needed at retirement to fund one rupee a year, rising with inflation, for 30 years at a real 2.83% is 20.04 rupees. That multiplier is the annuity factor, and it is the whole calculation in one number.

The first two rows are the same plan wearing different labels, and you can prove it in one multiplication: ₹1.8 crore × 3.21 is ₹5.78 crore. If your two versions do not reconcile by exactly the inflation factor, one of them is broken.

Row three deserves its own name, because it is the version that appears in most spreadsheets and in a good deal of published material. Discounting a stream at the nominal rate is exactly correct — for a stream that is flat in rupees. So ₹2.97 crore is the honest answer to a different question: what funds ₹28.9 lakh a year, never once increased, for 30 years? Call it the frozen-rupee retirement. It is a plan in which your income buys less every year for three decades, and it is short of the consistent answer by ₹2.8 crore, or 49%.

That is worth sitting with, because the mistake feels conservative while it is being made. A 9% return sounds ambitious; using it as the discount rate feels like a safe, optimistic-return assumption. It is the opposite. The high rate is not buying you headroom — it is silently deleting inflation from every year after you stop working, which is precisely the stretch where you have no salary to absorb it.

Every rate you have ever been quoted is nominal

The real return is the one input nobody is ever handed. The Public Provident Fund — PPF — pays 7.1%. The Employees' Provident Fund, EPF, carries a rate of 8.25% for FY 2025-26: recommended by the EPFO Central Board of Trustees, ratified by the Central Government, and only then credited by EPFO, which are three stages that coverage routinely collapses into one. The Senior Citizens' Savings Scheme pays 8.2%. A bank quotes its deposit rate in the same units. Every one of those is a number in future rupees, and none of them tells you what the money will buy.

Converting is a division, not a subtraction. Earning 9% while prices rise 6% does not leave you 3% better off, because the 9% is earned on rupees that have themselves shrunk. The correct form is (1 + nominal) ÷ (1 + inflation) − 1, so 1.09 ÷ 1.06 gives a real 2.83%, not 3%. Run that same division on the administered rates just quoted, against 6% illustrative inflation, and PPF lands a little above a real 1% and EPF a little above 2% — which is the figure that decides anything, and the one no scheme document prints.

Now the proportion, because this is where attention usually goes to the wrong place. Taking the subtraction shortcut — 3% in place of the true 2.83% — changes a 30-year corpus by about 2%. Using the nominal rate instead of the real one changed it by 49%. The shortcut is a rounding error at Indian rates of inflation; the unit mismatch is a halving. Fix the second first, and the first only if you enjoy it.

The real rate also decides how expensive a retirement is, and the effect is larger than it looks. At a real 2.83% the annuity factor is 20.04. At a real 1% — roughly where an administered rate lands against 6% inflation — it is 25.81, so the same spending, for the same 30 years, needs 29% more capital. What that extra capital buys is the absence of market risk, which is a genuine purchase and not a penalty. The trade is real in both directions, and it is stated plainly here because neither side of it is free.

One caution on using any administered rate as a 30-year planning input. These rates are set by notification rather than contracted for the life of a plan, so a projection built on today's number is assuming a number, not locking one. That is the same class of assumption as an equity return — smaller in range, not different in kind.

How many years it has to last

The annuity factor is where the horizon enters, and it behaves less alarmingly than most people expect. Here it is at a real 2.83%, alongside what each implies for the ₹28.9 lakh of first-year retirement spending derived above.

Read down the middle column and the shape is the useful part. Five extra years from 30 adds 10%; the next five add 8%; the five after that, less again. Each additional year is discounted harder than the one before it, so the factor climbs towards a ceiling instead of running away. That ceiling is 1 ÷ 0.0283, or 35.33 — the cost of an income that never stops at all.

Which produces a genuinely useful result: planning to never run out costs about three-quarters more than planning for 30 years. Not ten times more, not an unreachable multiple. For a household that finds a 30-year horizon uncomfortably close to a bet on its own longevity, that is a priced option rather than an impossible one, and it is the arithmetic underneath both a lifetime annuity and the "live off the real return, never touch capital" instinct.

Note also how little the horizon moves the answer either way. Dropping from 30 years to 25 saves 11%; adding five costs 10%, and the next five cost 8% on a bigger base. Five years in either direction is worth about a tenth — against the 49% the unit error costs and the 37% a point of inflation costs below — which is worth knowing before spending an evening on a life-expectancy table. The asymmetry in the consequences — unspent money against no money at eighty-five — is argued in the article on which assumptions decide the answer, and the structural response of covering essentials with income that does not depend on how long you live is annuities.

Two conventions worth stating rather than hiding. These factors assume withdrawals at the end of each year; taking them at the start multiplies the corpus by about 3%. And they assume a constant real return, which no portfolio delivers — the order in which returns arrive matters enormously once you are withdrawing, which is the subject of withdrawal rates and sequence risk.

Why the answer is a band, and what sets its width

Change one input at a time from the illustrative base and the answer moves like this. Everything else is held where it started, and the nominal return is held fixed while inflation varies — a choice that matters, for reasons directly below.

A single percentage point of inflation moves the answer roughly two and a half times as much as a point of return. That inverts where effort normally goes: return assumptions get argued over for hours, and the inflation figure gets picked in about four seconds.

Now the part that is almost never stated, and it changes how you should read every table like the one above. The sensitivity depends on what you held fixed. Hold the real return at 2.83% and vary inflation instead, and the corpus at the retirement date still moves — ₹6.98 crore at 7%, ₹4.78 crore at 5% — but the corpus in today's money does not move at all. It is ₹1.8 crore under every one of the three. Nothing about the plan changed; only the size of the rupee you quoted it in.

So inflation earns its reputation for a more precise reason than "inflation is scary". Under a fixed real return it changes the label and nothing else. What makes it bite is that no real household holds a real return fixed — you hold a nominal expectation, because a nominal number is the only kind anyone quotes you. A point of inflation is then a point off the real return, and it compounds the base as well. The argument to have is about the real return; the inflation figure matters because of what it does to that.

Which is also the reason the pinned-real version is not a second vote for the same conclusion. Its +21% and −17% at the retirement date are the rupee changing size, not the plan changing shape, and the today's-money figure sits still through all three. The ordering is a finding about the case a household actually faces — a nominal expectation held fixed — and not a general law about tables like this one. Quote a sensitivity, and say which of the two you computed it under.

Which is why the output is a band. Run your own inputs across the ranges you can defend and you get something like "between ₹4.2 crore and ₹7.9 crore, and here is what would push me to the top of it". A single figure carried to five digits claims a precision the inputs cannot support, and the band is more useful anyway — it tells you which assumption to watch, and recomputing it every few years against observed inflation narrows it in a way that guessing better never will. The published index is a starting point and not your basket; a retiree's basket is weighted differently.

Three checks that catch a broken corpus number

Corpus arithmetic fails quietly. There is no error message, and every wrong version produces a plausible-looking crore figure. Three checks catch nearly all of it, and each takes under a minute.

1. The reconciliation check. Compute the answer twice — once entirely in today's money, once entirely at the retirement date. Divide the second by the first. It must equal your inflation factor, here 3.21, to the last digit. Any other ratio means a rate and a price came from different years.
2. The factor check. Divide the corpus by the first year's retirement spending. The result should land in the range of the annuity-factor table — roughly 15 to 25 for ordinary real rates and horizons. A ratio of 10 means a nominal rate is doing a real rate's job. A ratio above 35 means the horizon or the rate is wrong.
3. The label check. Write the unit next to the number, every time: "₹5.78 crore in the rupees of the retirement year", or "₹1.8 crore in today's money". Most mixing errors are impossible to make once the unit is written down, and impossible to catch once it is not.

One more thing the calculation does not include, and it belongs in your estimate rather than in the factor. Withdrawals are gross and spending is net — to spend a rupee you must redeem more than a rupee, because part of what you redeem is taxable gain, and that share grows as a holding ages. A corpus sized against net spending is undersized by whatever the gross-up turns out to be. The rates that apply are set out in the withdrawal-rate article, and the mechanics of drawing an income down are in how a systematic withdrawal plan behaves.

And when the number comes out larger than expected — it usually does — that is the calculation working rather than failing. It has converted a vague worry into a specific gap, and a gap has levers, which are worked through in the companion article and in how much to save each month. Raising the return assumption is the one response that changes the spreadsheet without changing the plan.

Testing the plan rather than the number

Everything above runs on a constant real return, which is a modelling convenience and not a property of any portfolio. The gap between the two is not a detail: a plan that reaches the corpus on average may not reach it from the start dates that actually occurred, and a fall arriving in the first years of drawdown does far more damage than the same fall arriving later.

FNOTrader's Mutual Funds app runs on the full history of published daily per-unit prices — net asset values, or NAVs — from the Association of Mutual Funds in India, around 34 million rows of them, and reports rolling-return distributions across every available start date, alongside maximum drawdown. Two tests follow directly from the arithmetic above: whether the accumulation plan reaches the corpus using the worst historical window rather than the mean, and what a drawdown of that depth would do arriving in the first three years of withdrawals.

Historical outcomes describe what happened, not what will happen; past performance does not indicate future results. FNOTrader is not a SEBI-registered investment adviser and none of this is retirement advice. What corpus suits a particular household 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