How to Stress-Test a Retirement Plan

Ramesh's plan passed a test that asked almost nothing

Ramesh has had his retirement plan checked. The adviser lowered the assumed return by a couple of points, the plan still worked, and everybody was satisfied.

That test was close to worthless, and not because lowering the return is wrong. It is because the return is one input among several, it is not the one most likely to break the plan, and varying it alone tests the plan against a single narrow failure while leaving the others unexamined.

A stress test is not a way of confirming that a plan survives. It is a way of finding out how it dies — which failure arrives first, how much warning there is, and what would have to be done. A test that produces a reassuring verdict and no list of responses has not been run properly.

Test the failures separately

The first discipline is to vary one thing at a time, because a plan that fails needs a diagnosis rather than a verdict.

The failures worth testing are distinct from each other, and each has its own signature. General inflation running higher than assumed, which erodes everything slowly. Medical costs running higher than general inflation, which is concentrated late and arrives suddenly. Living longer than planned — which for a couple means the second death, not the first. A lower long-term return. A bad first few years, which is a different failure from a lower average and a far more dangerous one, for reasons demonstrated here. Expected income that is delayed, reduced or does not arrive. A single large expense — a medical event, a family obligation, a roof. And one spouse dying much earlier than the other, which changes both the spending and the income, usually not proportionally.

Running these separately tells Ramesh which of them his plan is actually sensitive to, and that is the output worth having. A plan barely moved by a lower return but broken by a five-year gap in one income source has told him precisely where to concentrate.

Record consequences, not verdicts

The second discipline is about what gets written down, and it is where most stress testing goes wrong.

A probability of success, or a pass, is very difficult to act on. What Ramesh needs from each scenario is four things: when the shortfall first appears, how far the portfolio falls at its worst, whether the essential spending is still covered, and how large a correction would be needed to repair it.

The fourth is the most useful and the least often recorded. "This plan fails at eighty-two" is alarming and unhelpful. "This plan needs about a tenth less discretionary spending from year eight, and we would see it coming by year six" is a plan. The same failure, described in a way that can be responded to.

The timing element matters more than it appears to. A failure that becomes visible fifteen years in advance is a manageable problem, because there is time for small adjustments. One that becomes visible two years out is an emergency, because only large and painful actions are still available. Some of the most valuable output of a stress test is knowing which of his failures announce themselves early.

Do not build one impossible disaster

A caution, because the instinct to be thorough produces a specific error.

Combining every adverse assumption at once — highest inflation, longest life, lowest return, worst sequence, no pension, a large medical event — produces a scenario that fails and tells Ramesh nothing. Almost any plan fails that test, including sound ones, and the natural conclusion from it is either despair or dismissal. Both are wrong.

The useful combinations are the plausible ones, and particularly the ones where the components are actually related. High inflation with lower real returns is a coherent pairing. A poor first decade combined with living a long time is coherent and is probably the single most dangerous realistic scenario a retiree faces. A medical event combined with the loss of one income is coherent, because the two often arrive together.

Test the coherent pairs. Leave the everything-at-once scenario out, and be clear that a plan is not required to survive it.

Decide the responses in advance

A stress test is finished when each failure has an action attached to it, and the actions should be ranked before any of them is needed.

The ordering is the same for most households. Essential spending is protected first, by income that does not depend on markets and by the near-term reserve. Discretionary withdrawals are what flexes, and a plan with a meaningful discretionary portion is far more robust than one where everything is essential — the flexibility itself is the safety margin.

Before retirement there are two more levers that disappear afterwards: the retirement date, and part-time work. Both are far more powerful than any investment change and both need to be identified as options while they still exist.

Then, and only then, larger structural changes — housing, or a legacy intention that gives way.

Adjusting the allocation is deliberately last on that list, and it is usually first on everybody else's. Reaching for a higher expected return to repair a shortfall adds volatility exactly where volatility does the most damage, which makes the plan more fragile in the name of fixing it.

Redo it, because it is a snapshot

The final point is that a stress test ages.

It rests on today's balance, today's spending and today's assumptions, and all three move. A test run at retirement and never repeated is answering a question about a household that no longer exists. Repeated annually, it becomes something better: a comparison against last year's version, which shows whether the plan is drifting toward one of its failure modes while there is still time to act.

That drift is the thing to watch for, and it is invisible from any single year's test.

What to take away

Vary one assumption at a time — inflation, medical costs, longevity, return, a bad first few years, lost income, a large expense, an early death — because a plan that fails needs a diagnosis rather than a verdict.

For each, record when the shortfall first shows, how deep the fall goes, whether essential spending survives, and what size of correction repairs it. Test the plausible combinations and skip the everything-at-once disaster, which fails for every plan and teaches nothing. Rank the responses in advance, with essential spending protected first and the allocation changed last. Then run it again next year and compare, because the drift between two tests is more informative than either one.