Guide · Part II — The mental tools

The problem with average returns

Issued 2026-07-05established frameworkConfidence high

DiagramTwo portfolios with the same average return: the steady path compounds; the volatile path hits a fall deep enough that the average never mattered.

An average return can describe an investment that loses money. That is not a paradox or a rounding error — it is the first fact of portfolio mathematics, and every tool in this part of the curriculum exists because of it.

The problem

Picture two ponds. In pond A, the water rises 50%, then falls 40%. In pond B, the water never moves. Most people, asked which pond holds more water at the end, pick A — it went up more than it came down. The arithmetic says B wins, decisively.

Run it with 100 dollars. Up 50% takes it to 150; down 40% takes it to 90. The portfolio is down 10%, even though the simple average of the two years is plus 5%. Returns multiply, they do not add, so a percentage gain and an equal percentage loss never cancel. Lose 40% and the climb back to even requires plus 67%, because the recovery starts from a smaller base.

This is the difference between the arithmetic mean (add the yearly returns, divide by the number of years — the number brochures quote) and the geometric mean (the rate wealth actually compounds at). Every marketing document leads with the first. Every account balance is built by the second.

The insight

The gap between the two numbers has a name and a formula. Over many periods, geometric mean is approximately the arithmetic mean minus half the variance — variance being volatility squared. That subtraction term is volatility drag: the permanent wealth bled away purely from bouncing around.

Two illustrative assets make the point. Asset X averages 15% with 20% volatility; its compound rate is roughly 15% minus 2%, so 13%. Asset Y averages 20% with 40% volatility; its compound rate is roughly 20% minus 8%, so 12%. Y carries the higher headline and compounds slower, because doubling the volatility quadruples the drag.

The extreme case is instructive. An asset that returns plus 300% one year and minus 75% the next has an “average” of plus 112% — and a compounded result of exactly zero. Every 100 dollars becomes 400 dollars, then becomes 100 dollars again, before fees, with a stomach-churning ride in between. The headline was a mirage; the drag was the truth. A hypothetical 2021 buyer of a token with exactly that profile was not unlucky — they were reading the wrong number.

The core of it:

Return is only half a number; the other half is what you risked to get it.

From here forward, no return figure in this curriculum stands alone; every one is paired with the risk taken to earn it. That pairing is what “risk-adjusted thinking” means, and it is the shared foundation of everything from Markowitz to Kelly.

In plain English

Smooth-and-slightly-lower beats volatile-and-higher over any real horizon. A big loss is a permanent tax on future compounding — a 50% drawdown demands a 100% gain to undo, which is why avoiding the catastrophic loss is mathematically more valuable than catching the extra upside. The question to carry into any pitch is not “what did it average?” but “what did it compound, and what was the worst stretch along the way?”

One more consequence hides in the safest corner of the portfolio. Cash is not zero-risk; it is a small, certain, invisible loss to inflation every year — a drag applied to the measuring ruler itself rather than to the asset. The allocator hiding entirely in cash has not escaped the compounding asymmetry; they have chosen the leak they cannot see over the one they can. “Safe” means matched to when the money is needed, not left in currency forever.

Where this breaks

The volatility-drag formula is an approximation, and it inherits the assumptions of the return series it is fed. It describes risks that live inside the historical distribution of prices — the bouncing. It says nothing about the risks that live outside it: an issuer defaulting, an asset being reclassified by a regulator, custody failing outright. An asset can have low measured volatility and a catastrophic hidden tail, and this chapter’s arithmetic will grade it as safe. Later chapters — particularly the move from VaR to CVaR — exist to patch exactly that blind spot.

Nor does risk-adjusted thinking pick assets by itself. It is a lens, not a menu. The regime map in Part I still decides which assets are in season; this chapter decides how their track records should be read.

The immediate action is a single substitution: wherever a product, a pitch, or a headline offers an average return, ask for the compound return and the worst peak-to-trough loss over the same period. Any seller who cannot or will not produce those two numbers has answered a different question — and that answer is also information.