Why Averages Hide The Risk In A Plan

A retirement projection built from average assumptions produces a single tidy line. The tidiness is the problem: the average outcome is not the outcome most households should plan around.

Averages discard the sequence

A given average return can be produced by many different orderings of annual returns. For a portfolio with no flows in or out, the ordering makes no difference to the final balance.

Once money is being withdrawn, ordering matters enormously. Poor returns early reduce the capital base that later good returns act upon.

An average-return projection assumes the ordering away entirely, which removes the single largest risk facing a retiree in the first decade.

Compounding is not symmetric

A decline requires a proportionally larger gain to recover from, because the gain applies to a smaller base. Losses and gains of equal size do not cancel.

This means a volatile series with a given arithmetic average produces a lower compounded result than a smooth series with the same average.

Projections that apply an average annual return therefore describe an outcome better than what an equally-averaged volatile path would actually deliver.

Average lifespan is the wrong planning horizon

Life expectancy is a midpoint. Roughly half of people at any given age will live beyond it, and planning to the midpoint means planning to run short in about half of cases.

The relevant figure is a percentile far out in the tail, which is materially longer, particularly for couples where the plan must last until the second death.

Expectancy also rises with age already attained, so a figure quoted from birth understates the horizon facing someone already retired.

What distributions show that averages cannot

Running many possible paths rather than one produces a range of outcomes, and the useful information sits at the unfavourable end of that range.

The question shifts from what will happen to how bad the plausible bad cases are and whether the household could absorb them.

That reframing tends to change decisions, because a plan that looks comfortable on average may have a poor tail that was never visible.

The limits of the alternative

Simulations are only as good as the assumptions behind them, and a distribution generated from the wrong inputs is confidently wrong rather than usefully uncertain.

Their real contribution is comparative: they show whether one plan has a worse tail than another, which is more reliable than any single probability figure.

Treating the output as a ranking tool rather than a forecast keeps it in the role it can actually fill.