What a Portfolio Is and Why the Collection Behaves Differently

Scope. This article explains why a group of holdings has properties none of its members has, and what follows arithmetically. It recommends no allocation, suggests no weights, gives no rule of thumb and produces no number a reader could apply to their own money. All figures are the illustrative teaching parameters fixed on this pillar's hub and are not forecasts of any market.

The return is an average — that part is intuitive. A portfolio's expected return is the weighted average of the expected returns of what it holds, and nothing surprising happens. The risk is not an average, and everything interesting in this pillar follows from that one asymmetry.

Why risk does not average

Two holdings that both fall on the same days are a different proposition from two that fall on different days, even if each is individually just as variable. When holdings move independently, their fluctuations partly cancel; when they move together, they compound.

So a portfolio's variability depends on three things rather than one: how variable each holding is, how much of each is held, and how the holdings move relative to one another. The third has no counterpart in the return calculation, and it is the one people ignore.

The result that intuition gets wrong

Using the hub parameters — equities at a 9.0% expected return with a 16.0% standard deviation, bonds at 5.0% and 6.0%, and a correlation of 0.20 between them.

Worked example — adding the riskier asset made the portfolio safer. Bonds alone carry a standard deviation of 6.00%. Adding equities — an asset almost three times as variable — brings it down to 5.91% at a 6.6% holding, while raising the expected return from 5.00% to 5.26%. Lower risk and higher return simultaneously, from adding the riskier thing. This is not a trick and it is not a rounding artefact. It happens because at a correlation of 0.20 the two assets frequently move differently, so a small equity holding cancels part of the bonds' fluctuation more than it adds of its own. Beyond about 6.6% the effect reverses and risk rises with every further addition, which the rest of the table shows. The point is not that 6.6% is a good number — it is an artefact of three assumed inputs and this portal recommends no allocation whatsoever. The point is that the relationship between a holding's own risk and its effect on a portfolio is not monotonic, so no intuition about individual holdings can be trusted to describe the collection.

What diversification cannot do

The same mathematics that produces that result also imposes a hard limit, and the limit is the more important half.

Consider a portfolio of equally weighted individual stocks, each with a 30.0% standard deviation and an average correlation of 0.20 with the others.

Worked example — the floor, and why it exists. The irreducible standard deviation is the single-holding figure multiplied by the square root of the average correlation: 30.0% × √0.20 = 13.42%. Adding holdings removes the part of the variability that is specific to each one and removes none of the part they share. A portfolio of five hundred stocks still carries 13.42% of variability, and a portfolio of five thousand carries the same. Two consequences follow and neither is a recommendation. First, the returns to breadth diminish sharply — the move from one holding to twenty removes 92% of what can be removed, and everything after that competes for the remaining 8%. Second, and more consequential: the floor rises with correlation. If the average correlation were 0.60 rather than 0.20, the floor would be 23.24% rather than 13.42% — so the amount of risk diversification can remove is set by how much the holdings have in common, not by how many there are. That is why the next article in this pillar is about correlation rather than about counting.

Three things this changes

A holding cannot be assessed alone. Whether something adds or removes risk depends entirely on what else is held, so an assessment of a security in isolation is incomplete by construction.

Counting holdings measures very little. Twenty holdings that share an exposure are less diversified than eight that do not, and the count says nothing about which case applies.

The inputs are estimates and the outputs inherit that. Every figure above rests on assumed returns, variabilities and correlations. The arithmetic is exact; the inputs are not, and Pillar 25's article on quantitative analysis sets out what happens when exact machinery is fed uncertain estimates.