Sequence of Returns Risk: Why the Order of Returns Matters More Than the Average

During accumulation with no cash flows, order is irrelevant: multiplication commutes, so a −30% year followed by +40% ends in the same place as +40% followed by −30%. Withdrawals break that symmetry. Every withdrawal taken during a drawdown sells a larger fraction of the portfolio, and those shares are gone — they don't participate in the recovery.

Consider spending 40,000 from a 1,000,000 portfolio. In a flat market that is 4% of the portfolio. After a 40% crash it is 6.7% — the same lifestyle now consumes the portfolio two-thirds faster. A few years of this early in retirement can dig a hole that even a strong subsequent bull market cannot refill, because the money that would have compounded is no longer there.

Sequence risk is heavily front-loaded. Early in retirement the portfolio is at its largest and faces the most remaining years of withdrawals, so damage done then compounds through the whole plan. In historical simulations, plans that survive their first 10-15 years without major erosion almost always survive to the end; the failures are nearly all seeded early.

This is why the worst US retirement cohorts were not those who merely experienced crashes, but those who hit sustained bad real returns immediately: the 1929 retiree who walked into a deflationary collapse, and the mid-1960s retiree who faced fifteen years of stagflation grinding down real portfolio values while withdrawals kept rising with inflation. Retirees who saw the same events mid-retirement — with a decade of growth already banked — sailed through.

A projection built on average returns will never show sequence risk — it is by construction a single smooth path. Historical backtesting shows it, but only in the handful of orderings history actually produced. Monte Carlo simulation with block bootstrap sampling generates thousands of orderings built from real multi-year market dynamics, which is what makes the resulting failure probabilities meaningful.

The sampling method matters here. Simulations that draw each year independently shuffle history too finely and dilute the very clusters of bad years that create sequence risk. Sampling history in multi-year blocks preserves those clusters — the 1970s stay the 1970s — so the simulated bad tails look like the kind of bad decades that actually occur.

Three defenses do most of the work. First, a lower initial withdrawal rate: the smaller the withdrawals, the less a bad early sequence can amplify them; this is the blunt but reliable tool. Second, spending flexibility: dynamic strategies like guardrails cut spending precisely in the states where withdrawals do the most damage, which is why they raise sustainable starting rates so effectively — they neutralize bad sequences at their source.

Third, allocation. Bonds and other diversifiers cushion the early-retirement drawdowns that seed failures, which is why moderate bond allocations often improve worst-case outcomes even when they lower average ones. Some retirees formalize this as a 'bond tent' — holding more bonds around the retirement date, then re-equitizing as the danger zone passes. Cash buffers held to 'avoid selling low', by contrast, mostly reshuffle the same trade-off: the drag of holding cash costs about as much as the selling it prevents.

Accumulators face the mirror image. Early in your saving career, crashes are nearly irrelevant — there is little capital exposed and decades of contributions ahead buying cheap shares. In the final years before retirement, a crash hits maximum capital with no time to recover before withdrawals begin. The riskiest window is roughly the five years either side of the retirement date.

The practical implication: your true risk capacity falls as the retirement date approaches, regardless of your risk tolerance. Planning tools should treat the retirement date as the center of the danger zone, not as a finish line after which risk starts.