Why portfolios beat picks — course checkpoint

Sixteen lessons ago the claim was that holding many things is not the same as holding one thing many times. Everything since has been the proof, the mechanism and the fine print. Here is the whole course in one place.

The one sentence

Expected return averages; risk does not. Every result in this course is a consequence of that asymmetry. Return is linear in the weights. Risk is not — it combines in squares, and carries a correlation term on top that pulls the total below the weighted average of the parts. That extra term is the only place in investing where something improves without anything being paid for it.

The four movements

Unit 1 — the one-stock problem. A single holding is a draw from a brutally skewed distribution: in the long US record, 57.4% of stocks failed to beat Treasury bills over their lifetimes while a small minority produced all the net wealth. Splitting money across near-identical holdings does not fix this. Risk combines in squares, and σ × √(ρ + (1 − ρ) ÷ N) shows exactly which part shrinks with count and which part never does.

Unit 2 — correlation is the engine. Correlation measures direction and consistency — not magnitude, not causation — and it is never a constant. Twenty holdings at ρ = 0.7 are about 1.4 independent bets; the same twenty at ρ = 0.2 are about 4.2. The two-asset formula puts numbers on it: at ρ = 0.2, a 20%-volatility holding mixed 50/50 with a 15% one produced a portfolio calmer than either — 13.6%.

Unit 3 — the free lunch and the frontier. Lower correlation reduces risk without reducing expected return, and lower volatility raises the compounded result from the same average (drag ≈ volatility² ÷ 2). Plot every possible mix and the best-in-class edge is the efficient frontier — a curve, not a line, and the bow in that curve is diversification drawn. The frontier's logic is durable; its exact coordinates rest on the least reliable input in finance.

Unit 4 — the limits. Total risk splits into idiosyncratic (removable, and in this framework uncompensated) and systematic (unremovable, and the part expected to carry a reward). Diversification takes the first and leaves the second — and correlations rise in crises, so even the part it does take gets handed back temporarily at the worst moment.

The three sentences worth keeping

1. Count is not diversification; difference is. Ask what would have to be true for a holding to fall 30%, and count the distinct answers.
2. The floor is set by correlation, not by holding count. Adding names to a tight cluster buys almost nothing.
3. The free lunch is free in the arithmetic, not in the world. Costs, shifting correlations and crisis behaviour each take a bite — and none of them makes the underlying asymmetry go away.

What this course did not do

It never said what to hold, in what proportion, or in what account. Every ticker, weight and percentage across these sixteen lessons was an illustration of arithmetic. Portfolio theory is a lens for describing what a set of holdings is — its structure, its concentrations, its floor. What anyone builds with that description depends on circumstances no course can see.

Before you sit it

Each of these is a minute at your desk. Any one that is not names the lesson to reopen first.

• Say roughly how many holdings it takes before another name stops reducing risk — How many holdings does it take before the benefit runs out?
• Say what happens to a two-asset portfolio's volatility when correlation falls from 1.0 to 0.5 — How much risk does a lower correlation actually remove?
• Describe the efficient frontier in one sentence without the word matrix — What is the efficient frontier, without the matrix algebra?
• Name the kind of risk diversification removes and the kind that stays whatever you do — Which risks does diversification remove, and which stay forever?

Try it now

1. Take your watchlist one final time and write its structure in three lines: how many holdings, how many genuinely distinct exposures, and what single event would hurt the largest share of it.
2. Verify one number against data rather than memory. Unit 4's 2022 claim is the easiest to check: navigate both charts below to 2022 and Measure the calendar year on each. Two sleeves that are supposed to disagree, falling together.

3. Next in this domain — Asset Allocation — takes the machinery you just built and applies it across asset classes, horizons and rebalancing. This course was the why; that one is the how much.

Checkpoint quiz next. Nothing here was a recommendation about any security or any allocation — you have learned how portfolios behave, which is a description, not a signal.