Up 50%, then down 50%, and you have lost a quarter of your money

Start with the arithmetic that everyone gets wrong on the first pass. Your portfolio gains 50% in year one and loses 50% in year two. The average return is zero, so you should be back where you started.

You are not. 100 grows to 150, then 150 falls to 75. Two years, average return of exactly zero, and a quarter of your money is gone.

Nothing exotic happened. No fees, no taxes, no bad timing. This is just what multiplication does: a gain and a loss of the same size do not cancel, because the loss applies to a bigger number. The gap between the average return and the money in your account has a name, volatility drag, and it quietly reshapes every long-term projection you have ever been shown.

The +50%/-50% example contains two different averages.

The arithmetic average adds the returns and divides by two: (50 - 50) / 2 = 0%. This is the number fund marketing, backtests, and most casual conversation mean by "average return."

The geometric average asks a different question: what single return, repeated every year, would have produced the same final balance? For 100 ending at 75 over two years, that is about -13.4% per year. Negative, both years, because your money actually shrank.

The arithmetic average describes the returns. The geometric average describes your wealth. They only agree when volatility is zero, and the gap between them grows with the square of volatility. The approximation is compact enough to keep: geometric return ≈ arithmetic return minus half the variance. A portfolio with 15% annual volatility pays roughly 0.15² / 2 ≈ 1.1 percentage points per year in drag. At 20% volatility, the toll is about 2 points.

Those are not hypothetical numbers. US large-cap stocks over the long run have an arithmetic average return around 12% with volatility around 20%, and the compound return investors actually earned is around 10%. The missing two points did not go to anyone. They are the arithmetic of volatility itself.

Take the most common planning sentence there is: "assume 7% average returns."

Compound 7% for ten years and you get 1.97x, the doubling that every simple calculator promises. But if that 7% is an arithmetic average and the portfolio has 15% volatility, the compound growth rate is closer to 5.9%, which after ten years is about 1.77x. Over thirty years the gap widens from 7.6x to 5.5x. Same "average return," same calculator, and the honest number is about 27% less money at retirement.

The uncomfortable part: neither input was wrong. 7% may be a perfectly reasonable average. The lie is structural, a calculator that accepts a return but not a volatility has silently assumed volatility is zero, and zero volatility is the one assumption guaranteed to be false.

If volatility drag feels abstract, you have probably already watched it operate in the wild: leveraged ETFs.

A 2x fund doubles the daily return, but drag scales with the square of volatility, so doubling the exposure quadruples the toll. One flat, choppy sequence makes it visible: the index gains 10% one day and loses 10% the next, ending down 1%. The 2x version gains 20% and loses 20%, ending down 4%. Repeat that chop for a year and the index is roughly flat while the leveraged fund has bled away a meaningful share of its value, which is why regulators publish plain warnings that these products can lose money over time even when the index they track ends up unchanged.

Leveraged funds are volatility drag with the volume turned up. An ordinary portfolio experiences the same force at lower volume, every year, forever.

Follow the logic to its end and the conclusion is awkward for every retirement calculator that returns a single number.

If the path of returns changes the outcome, then a given (return, volatility) pair does not produce one future. It produces a distribution of futures: thousands of possible paths, some where the bad years cluster early, some late, some barely at all. The honest answer to "when do I reach my number?" is not a date. It is a range, with a middle, and the middle of that range sits below what naive compounding promises, by roughly the drag.

This is what Monte Carlo simulation is for. Not sophistication for its own sake, but the only way to let volatility do to the projection what it will do to the money. And it is why a projection tool that does not ask for volatility cannot be honest, whatever else it does: the input it skipped is the input that separates the average from what you keep.

Opula's projection tool is built on exactly this arithmetic. When you ask Claude to project your net worth with Opula connected, the simulation requires both an expected return and a volatility for each scenario, runs a thousand paths per scenario, and returns the distribution: the median, the spread, and the probability of reaching a target by a date, rather than one flattering compound-interest line. Scenarios share the same sequence of random shocks, so when the pessimistic case lands below the optimistic one, the difference comes from your assumptions and not from luck of the draw.

One disclosure, because it is the difference between a model and a promise: the simulation is honest about your assumptions, not about the future. If the return and volatility you feed it are wrong, the fan of outcomes is wrong too. That is why Opula never supplies default assumptions and every projection states the ones you chose, where you can argue with them.

Why doesn't a 50% gain cancel a 50% loss? Because the two percentages apply to different amounts. The 50% loss acts on the larger, post-gain balance, so it removes more money than the gain added. In multiplicative terms, 1.5 × 0.5 = 0.75, a 25% loss.

What is the difference between arithmetic and geometric average returns? The arithmetic average sums yearly returns and divides by the number of years. The geometric average is the constant yearly return that would reproduce your actual final balance. Wealth compounds, so the geometric average is the one that describes what you earned, and it is always the lower of the two unless volatility is zero.

How large is volatility drag in practice? Roughly half the variance of returns, per year. At 15% volatility that is about 1.1 percentage points; at 20%, about 2. For leveraged products the volatility multiplies and the drag grows with its square, which is why they can decay in sideways markets.

Is a 7% average return assumption wrong? Not necessarily, it is in the range many long-run estimates land in. What is wrong is compounding it directly without a volatility input. The same 7% average with realistic volatility compounds like roughly 6%, and over decades that difference is a large share of the final balance.