Which risks does diversification remove, and which stay forever?
Every unit so far has pointed at a floor that diversification cannot get under. Time to name it — and to name the part it does remove, because the split is the organising idea of all professional risk work.
The split
Any holding's total risk breaks into two pieces.
Idiosyncratic risk (also called specific, unsystematic or diversifiable risk) — everything that can happen to that one thing: a failed product, a fraud, a lost lawsuit, a fire at the only factory, a CEO's resignation. These events are unrelated across companies. One firm's disaster is not another's.
Systematic risk (also called market risk, or undiversifiable risk) — everything that moves all holdings at once: recessions, interest-rate regimes, inflation, wars, liquidity crises, global panics. Every company in an economy stands in the same weather.
Diversification is a machine that removes the first kind and leaves the second untouched. That is not a flaw in the machine; it is the definition of what the machine does.
The arithmetic of the split
Risk components combine like the sides of a right triangle — in squares. Take a stock with 35% annual volatility whose correlation with the broad market is 0.6:
- Systematic portion: 0.6 × 35% = 21%
- Idiosyncratic portion: √(35² − 21²) = √784 = 28%
Check it: 21² + 28² = 441 + 784 = 1,225 = 35². The two pieces do not add to 49% — they add in squares to 35%, which is why risk arithmetic keeps surprising people.
Now hold hundreds of such stocks. The 28% pieces are unrelated to each other and largely cancel out. The 21% pieces all point the same direction and do not cancel at all. The portfolio converges toward its systematic component — 21% here — and stops there.
Two routes confirm the figure. If the broad market runs mid-teens volatility, say 15%, then a stock with 35% volatility and a 0.6 correlation to it moves about 1.4-for-1 with the market (0.6 × 35 ÷ 15 = 1.4), so a basket of such stocks floors at 1.4 × 15% = 21%. Unit 1's formula agrees: in a one-factor world two of these stocks share a pairwise correlation of 0.6 × 0.6 = 0.36, and σ × √ρ = 35% × √0.36 = 21%.
The mid-teens number often quoted as "the floor" belongs to a basket that moves roughly one-for-one with the market. These holdings ride it harder than that, so their floor sits higher. That floor is the market itself — scaled by how hard the holdings ride it.
Why the floor cannot be crossed by adding names
You cannot diversify away the market by buying more of the market. Owning every listed company on earth still leaves full exposure to "global equities have a bad decade." The tools for that exposure are different in kind — a longer horizon, other asset classes with different drivers, or simply holding less of it — and they belong to the allocation course, not this one. This course's job ends at a precise statement of what the free lunch buys: the removal of company-specific catastrophe, at no cost in expected return, and nothing beyond that.
Try it now
- One company and its index, the same year, below. Find a date where the first chart drops hard and the second barely registers it. That is the idiosyncratic 28% on display — the piece a basket cancels.
- Now find a date where both fall together. That is the systematic 21% — still inside the portfolio after all the diversification, and removed by no amount of counting.
- Sort a recent batch of headlines into the two buckets. Which kind dominates your reading, and which kind actually determines a diversified portfolio's year?