Retirement Income Sequencing

Two retirees with identical $2M portfolios and identical average returns can have radically different outcomes. The difference is sequence: when the bad years hit, where the cash comes from, and which tax bucket gets drained first. A retirement-income engine that ignores ordering produces plans that look right on paper and fail in production.

The classic illustration is two return paths with the same arithmetic mean. Path A starts with three down years; path B ends with them. Same average, same securities, same withdrawals — wildly different terminal balances. The math is unforgiving: a 30% drawdown while you are also withdrawing 4% of the portfolio compounds against you, and there is no later sequence of good years that fully repairs it.

This is the problem your engine has to model. Aggregate return assumptions are not enough.

“

The retiree who runs out of money usually didn't run out of returns — they ran out of returns at the wrong time.

What a working sequencer must model

A correct retirement-income engine tracks three intertwined ledgers per household: an account ledger (taxable, traditional, Roth, HSA, annuity), a tax ledger (ordinary income, LTCG, RMD-driven distributions, IRMAA bracket position), and a liability ledger (Medicare premiums, state taxes, healthcare share). Withdrawal decisions depend on the joint state of all three.

The IRMAA cliff

Medicare Part B and D premiums step up at five Modified Adjusted Gross Income brackets. Cross a bracket by a single dollar and you owe the full step-up — for two years, because IRMAA uses the MAGI from two tax years prior. This is the most overlooked variable in retirement-income engines.

The planning consequence: a Roth conversion sized to the 22% federal bracket can quietly trip an IRMAA tier and add $1,300+/year of Medicare premium for two years. Your engine's optimal_conversion_amount cannot just look at marginal tax rate — it has to look at the joint marginal cost across federal, state, IRMAA, and (for some clients) ACA premium tax credits.

A worked timeline

Here is what a realistic 10-year drawdown plan looks like for a couple retiring at 65 with $2.0M across a 40/40/20 split (taxable / pre-tax / Roth):

1. Age 65–66
   Bridge to Social Security on the taxable account
   Spend down the most tax-inefficient assets first; harvest gains in the 0% LTCG bracket while ordinary income is still low. Begin small Roth conversions sized to the top of the 12% bracket.
2. Age 67–69
   Roth conversion window
   Pre-RMD, pre–full Social Security. The cleanest 3-year window to fill the 22% bracket and still avoid the first IRMAA tier. Convert ~$80k/yr of pre-tax to Roth.
3. Age 70
   Claim Social Security at the 32% delayed credit
   Switch the income mix toward SS and taxable yield. Continue smaller conversions but model IRMAA carefully — 2-year lookback now binds.
4. Age 73
   RMDs begin
   Pre-tax balance has been pruned by 6 years of conversions; RMDs are smaller and the household stays a bracket below where it would have without the conversion ladder.
5. Age 75–80
   Roth withdrawals + qualified charitable distributions
   QCDs from the IRA satisfy RMDs without raising MAGI — the only legal way to keep IRMAA tiers low while still meeting the RMD requirement.

What goes wrong

This is the honest production failure mode. Engines built on iid return draws underestimate sequence risk by 30–50%. Engines built on historical bootstrapping underestimate it less but still miss the long correlated drawdowns (2000–2002, 2008, the 1970s for bonds). The fix is not a fancier returns model — it is stress sequences as a first-class input, not a what-if button.