Living off a portfolio without a crash cutting your living expenses

The idea: cover the need instead of setting a share

The problem first. A portfolio from which a fixed amount is withdrawn every year depends not only on the average return but on its order: two bad years at the start force sales at the lows, and the capital missing afterwards is not recovered by any later rally. The common remedies are withdrawal rules with built-in cuts, or a cash bucket for two or three years that runs dry in a long drought. Both accept that living expenses fluctuate. The construct described here does not, for the part of spending that must not fluctuate: it is taken out of equity risk for twenty years. Only the remainder may fluctuate afterwards, and that remainder is honestly labelled as fluctuating.

The usual question is "what percentage in equities". The bridge model does not ask it. Instead it splits annual spending into two parts: a floor that must not be cut, and a variable remainder that may fluctuate. The floor holds everything non-negotiable such as rent, insurance, basic living. The variable part is travel, hobbies, comfort.

The floor gets its own pot, the buffer. It holds as many annual floor amounts as the window is long, twenty years in the example, in inflation-linked government bonds so that the purchasing power of the pot matches the purchasing power of the need. Everything else sits in a broad equity ETF and carries the variable part. The equity share is thus no longer a dial but a result: wealth minus buffer.

The bridge does not end at retirement. From then on the buffer only pays the difference between floor and pension; the window keeps rolling. Only towards the end of the plan does it shrink. That is the difference from a "bridge to the pension" that hands everything over to the equity rule afterwards.

The floor is a trade, not a gain

The larger the share of spending that is immunised, the more capital sits at roughly one percent real instead of in equities. That costs late income. In the measurement over 20,000 random paths, median income after retirement with full immunisation was about a third below that with a 75 percent floor. You buy early smoothness with later prosperity. Which share is right depends on how much of your spending you truly cannot cut, not on a return expectation.

Why the buffer rolls and is refilled only at all-time highs

A buffer that expires at retirement abandons the late years. A buffer that is rebalanced to target every year sells equities after the crash as well. The measurement over 20,000 paths with the same starting capital and the same need, tenth-percentile income in year 40:

The all-time-high condition is the active ingredient, not the reallocation itself. Both active rules move similar amounts over a lifetime; only one of them never sells at lows.

The trigger hangs on the price, not on portfolio value

Anyone who has sold equities to fill the buffer sees their portfolio value fall although the market stands still. A trigger on portfolio value reads your own reallocation as a crash and never fires again. The trigger therefore has to sit on a price index of the securities held, weighted with today's quantities. A tolerance of one to two percent below the high is enough so that not every tick counts.

A band instead of constant activity

Topping up to the full window at every new high is too busy: eleven or twelve reallocations per lifetime. Better a band: the buffer may run down. Only when the remaining coverage falls below a trigger, 15 years in the example, is the next high used. If it falls below a floor, say 8 years, you buy even without a high. That halves forced sales and cuts reallocations to a third. The floor is a risk dial, not a seat belt: the lower it sits, the more is in equities on average, with more terminal wealth and more uncovered years. Drawdown stops, tolerance bands and partial refills all land on the same curve in the measurement. They just turn the same dial differently.

Saving in the drought beats every rule variant

The only tested variant that improves both axes is not a rule on the portfolio but on consumption: if the floor is cut by 15 percent as soon as the buffer falls below the trigger, the share of uncovered years drops by more than half and terminal wealth rises, because fewer equities are sold at bad prices. Timing is decisive: early and mild delivers almost four times as much as late and drastic. At the floor the buffer is almost gone and hardly any saving years remain.

Why inflation-linked bonds, and why as an ETF

The reflex for the safe pot is a savings account or a money market fund. That is the wrong kind of safety. The liability, a real amount over twenty years, itself has a duration of around nine years. Cash has a duration of zero. If the real rate falls by one percentage point, the cash buffer is short by a multiple of what the linker is short by. The unhedged instrument is the cash, not the bond.

The second channel is inflation itself. Cash pays nominal, the need is real. In the model test with an inflation shock like 2022 without full pass-through by the central bank, the money market buffer loses about two years of coverage, the linker none. In isolation, cash wins the interest-rate risk and loses the inflation risk; against a real liability both act at the same time, and then there is no constellation in which cash comes out ahead without having bet on the direction. The linker does not win the bet, it ends it.

ETF instead of a ladder

A ladder of individual bonds matches duration more exactly and defers tax on the inflation adjustment until maturity. Both are small: at just over one percent real return, deferred money contributes almost nothing, and the ETF recovers a rate shock over roughly its duration while the buffer only sells about five percent per year. Against that stand two big arguments: Germany has not issued linkers since the end of 2023, a Bund ladder cannot be built. And a ladder would have to be managed by hand for forty years, at 85 as well. The construct is deliberately built to carry without ongoing maintenance: one security, one order in December.

What the linker does not capture

• Tax on phantom gains. The inflation adjustment is taxable, and the deduction grows with inflation, precisely in the scenario the linker is for. At high inflation the after-tax real return turns negative. No instrument in the euro area is free of this.
• Redenomination. If a country leaves the euro, its bonds drop out of the index at the then-current price, after the spread shock and devaluation. The exclusion does not protect, it locks in the loss. For a large member state that would be a one-off ten to fifteen percent of the buffer. No euro instrument helps against it; anyone who wants to hedge it needs something outside the system and must book it as insurance, not as an investment.
• The maturity band. As the remaining window shortens, the duration of the liability drops below that of the ETF. Then the buffer belongs in a shorter maturity band. That is an appointment in twenty years, not today.

The variable part swings more than the market

The variable part is withdrawn as a fixed percentage of the free equity portion, i.e. equity wealth minus what is bound as the refill reserve for the buffer. That reserve is a fixed block; it does not shrink in a crash. The entire decline lands on the free part: ten percent less price is about 18 percent less variable part, a third less price halves it.

The median is flat over ten years, not rising. That is by construction: the withdrawal rate is exactly the one at which capital is preserved on the median path. Anyone expecting a rising trend is confusing the arithmetic with the geometric mean. In return the band is wide: after ten years the worst tenth stands at about 40 percent of the starting value, and in almost nine out of ten paths there is at least one year with a drop of over 20 percent. That belongs on display, so that normal movement is not read as an incident.

Why no smoothing pot

A third pot that holds surpluses from good years for bad ones works arithmetically but fails three times over: it collides with the core rule that everything that is not equities counts as buffer. Folded into the existing buffer, the variable part eats the floor. And it is one more decision, every December. The gain would be about one percent of total income. The cheaper answer: in good years, do not spend the full variable part; the money stays in equities and keeps working.

Where the construct breaks

• A single return assumption carries everything. Reserve, variable part and coverage hang on the assumed real equity return. One percentage point less can tip the floor coverage from "covered" to "not covered". The budget figure is as accurate as this one number, and that belongs as a sensitivity table next to every result.
• Everything is a present value on the median path. Half of all outcomes lie below it. The dispersion has to be measured separately; reading the covered need alone is the most common mistake.
• History is no more dangerous than dice, but it is the history of a winning market. A block bootstrap over 150 years of US data gives the same results as independent random returns, because droughts are followed by recoveries. A sequence like Japan from 1990 does not occur in it.
• Silent errors. One implementation of this model contained four calculation errors that no formula would have reported: distributions counted twice, a trigger on the wrong basis, a cap that created capital need out of nothing, two opposing orders on the same security. They were only found by cross-checks that recompute year by year. Anyone building something like this needs self-tests, not a spreadsheet.
• It is not a model for the accumulation phase. It assumes the capital is there and the need is known. For the question "is it enough" there are better tools; this one answers "how do I split it so that the floor stands".

Questions that came up