Every module before this one, implicitly, has been about accumulation: building, valuing, and holding assets that grow. Decumulation is a different problem with a genuinely different arithmetic, because a retiree earns a return on a portfolio while also withdrawing from it while it is trying to compound, and the order in which good and bad years arrive, not just their average, determines whether the money lasts.
What is the 4% rule, and why can it fail?
The best-known heuristic, popularized from research associated with financial planner William Bengen and later broadened in the "Trinity study" tradition, is some version of a "safe withdrawal rate", commonly cited around 4% of the initial portfolio, adjusted for inflation each year thereafter, as a rate historically shown not to exhaust a portfolio over a multi-decade US retirement. Treat the number as a heuristic anchored to a specific historical sample and starting-valuation regime, not a law of markets: it is sensitive to the valuation level at retirement (a retirement that begins at rich equity valuations has, historically, supported a lower safe rate than one beginning at cheap valuations), to the inflation and interest-rate regime of the retirement period, and to the country and asset mix studied, a rate calibrated to one market's long-run history does not automatically transplant to another market's return distribution. The heuristic's real value is as a starting conversation, not a promise, and the reason it can fail even when its long-run average return assumption is correct is exactly sequence-of-returns risk: withdrawals taken during the early bad years of a retirement permanently shrink the base that later good years compound from, in a way that identical returns arriving in the opposite order do not. Worked Example 4 computes this precisely: two return sequences with an identical arithmetic mean, run in opposite order against the same withdrawal schedule, produce meaningfully different terminal wealth purely from the order the returns arrived in, the retiree who happens to face losses in the first year or two of retirement is structurally worse off than one who faces the same losses near the end, even though "the market averaged the same return either way" sounds, on the surface, like it should not matter.
Advertisement
What is the money-weighted return?
Because a decumulation portfolio has cash flowing both directions (an initial balance, then a stream of withdrawals, then a terminal value) its true realized return is not the simple average of the yearly returns; it is the money-weighted return, the same concept an XIRR measures for any irregular cash-flow stream, and R3's law for this branch is explicit that an IRR must tie to a hand-computed check, not just a spreadsheet function. Worked Example 4 sets up exactly this: a portfolio's initial value as an outflow, each year's withdrawal as an inflow to the retiree, and the portfolio's terminal value added to the final year's cash flow as a liquidation, then solves for the discount rate that sets the stream's net present value to zero, by hand, using Newton-Raphson iteration, and cross-checks the converged answer against an independent bisection search on the same cash flows.
How do US and Indian withdrawal rules differ?
The two markets' decumulation rules, as opposed to the arithmetic above (which is market-agnostic), diverge sharply and are each perishable enough to need a primary-source check before use. In the United States, tax-deferred retirement accounts carry a Required Minimum Distribution (RMD), currently beginning at age 73 under the SECURE 2.0 Act (confirmed directly from the IRS's own retirement-plan FAQ page), forcing a minimum annual withdrawal whether or not the retiree needs the cash that year, while a Roth IRA carries no RMD at all during the original owner's lifetime (also stated directly by the IRS), a genuinely useful sequencing lever: draw down the RMD-bound account first, or manage withdrawal order across account types, to control both the tax bill and the timing. In India, the National Pension System (NPS) has historically required at least 40% of the accumulated corpus be used to purchase an annuity at exit, with up to 60% withdrawable as a lump sum, and, under Section 10(12A) of the Income Tax Act, that 60% lump sum is fully tax-exempt. As of recent PFRDA regulatory changes, subscribers with a corpus above a stated threshold may now elect to take up to 80% as a lump sum (only 20% mandatorily annuitized), but the Income Tax Act's Section 10(12A) exemption has not yet been amended to match, so the additional 20% now permitted as lump sum (the slice between 60% and 80%) is, as things currently stand, taxable at the subscriber's slab rate even though PFRDA now permits withdrawing it. This is a live regulation-versus-tax-law lag, not a settled rule, and it is worth holding up as a concrete, current illustration of exactly why this module insists on primary-source verification rather than "the rule I remember": a regulator changing a withdrawal ceiling does not automatically change what the tax authority taxes, and an adviser working from last year's summary would get a client's tax bill wrong today. (For corpora at or below a small threshold, PFRDA permits full withdrawal without any annuitization requirement at all, check the current threshold, which has itself moved, against the live PFRDA circular before advising a specific client.)
Modern Data Analysis: scaling the single-path lesson into a real simulation
Worked Example 4's sequence-of-returns demonstration uses exactly two hand-picked return paths, on purpose, the point of a by-hand illustration is to make the mechanism undeniable, not to survey every path retirement markets could take. A real retirement plan needs the mechanism run across thousands of plausible paths, not two, to say something like "there is an 85% chance this withdrawal rate lasts 30 years given this asset mix", and that is a Monte Carlo simulation, built either from an assumed return distribution (drawing random annual returns from a fitted normal or bootstrapped historical distribution) or from a historical block-bootstrap (resampling real historical multi-year blocks of returns to preserve their actual sequencing and correlation structure, which the by-hand exercise above should have convinced you matters enormously). In pandas/numpy terms this is a loop (or a vectorized array operation) over, say, 10,000 simulated 30-year paths, each applying the same withdrawal logic as the hand calculation, then reporting the fraction of paths that never hit zero, the exact simulation-construction skill QM1.03 hands off to the quant-code region rather than building here. The discipline this module insists on before you trust that dashboard: reproduce the by-hand, single-path sequence-of-returns result inside the simulation code as a unit test first (feed it exactly the two return sequences from Worked Example 4 and confirm the code reproduces the same two terminal-wealth figures), a simulation that cannot replicate a result you already verified by hand is not ready to be trusted on the thousands of paths you cannot check by hand.