The Arithmetic–Geometric Mean Inequality: Meaning, Proof, Example, and Why It Matters in Finance

There’s a question that sounds almost too simple to deserve a name: if you average two numbers two different ways, will you always get the same answer? Add them and divide by two, or multiply them and take the square root, you’d think these should land somewhere close to each other most of the time. They don’t, not exactly, and the gap between them turns out to matter a lot more in finance than the question first suggests.

That gap has a name. It’s called the arithmetic-geometric mean inequality, usually shortened to AM-GM, and once you’ve internalized it, you start noticing it quietly running underneath a surprising number of finance concepts, including ones that never explicitly mention it.

What is the Arithmetic-Geometric Mean Inequality?

The AM-GM inequality states that for any set of non-negative numbers, the arithmetic mean of those numbers is always greater than or equal to the geometric mean of the same numbers.

For two numbers, written simply:

(a + b) / 2 ≥ √(ab)

Equality holds only in one specific case, when a equals b. The moment the numbers differ even slightly, the arithmetic mean pulls strictly ahead of the geometric mean, and the gap between them grows as the numbers become more spread apart.

A Quick Numerical Check

Take two numbers, 4 and 9.

Arithmetic mean = (4 + 9) / 2 = 6.5

Geometric mean = √(4 × 9) = √36 = 6

6.5 is greater than 6, exactly as the inequality predicts.

Now widen the gap. Try 1 and 100.

Arithmetic mean = (1 + 100) / 2 = 50.5

Geometric mean = √(1 × 100) = √100 = 10

Look at that difference. 50.5 versus 10. The arithmetic mean barely budges off the average of the two numbers, while the geometric mean is dragged way down toward the smaller number. This isn’t a coincidence or a quirk of these particular numbers. It’s the whole point of the inequality. The more dispersed your numbers are, the more the arithmetic mean overstates what the geometric mean would tell you.

Why Does This Happen At All?

The intuitive explanation comes down to what each mean is actually sensitive to.

The arithmetic mean treats every number as additive and independent. It doesn’t care about the relationship between the numbers, just their sum.

The geometric mean is multiplicative, and multiplication punishes extreme values asymmetrically in a way addition never does. If one number in a series is very small, even close to zero, it drags the entire geometric mean down hard, because you’re multiplying everything together and a small number shrinks the whole product. The arithmetic mean has no equivalent vulnerability. A very small number just gets averaged in proportionally with everything else.

This is precisely why the geometric mean is always equal to or smaller than the arithmetic mean, never the other way around. It’s structurally more sensitive to variability and to any single low value sitting in the dataset.

Where This Shows Up in Finance: Investment Returns

This is the part that actually matters for CFA candidates and working analysts, and it’s also the part most likely to trip people up if they don’t understand AM-GM at an intuitive level first.

Suppose a stock returns 50% in year one and then loses 50% in year two.

The arithmetic mean return looks deceptively fine:

Arithmetic mean = (50% + (-50%)) / 2 = 0%

Looks like you broke even, right? Now actually trace through what happened to your money.

Start with ₹100.

After year one (+50%): ₹100 × 1.50 = ₹150

After year two (-50%): ₹150 × 0.50 = ₹75

You started with ₹100 and ended with ₹75. You did not break even. You lost 25% of your capital. The arithmetic mean told you 0%, and reality told you a 25% loss. That gap is not a rounding error or a quirk of this specific example. It is the AM-GM inequality showing up directly in your portfolio statement.

The geometric mean return captures this correctly. For two periods:

Geometric mean = √(1.50 × 0.50) – 1 = √0.75 – 1 = 0.866 – 1 = -13.4%

A -13.4% annualized return compounded over two years gets you to that same ₹75 ending value, which is exactly what actually happened to your money. The arithmetic mean of 0% never could have told you that, because it doesn’t account for the order and compounding effect of returns, only their simple average.

Why This Matters So Much for Volatile Assets

The bigger the swings in returns, the bigger the gap between arithmetic and geometric mean, and the more dangerously misleading the arithmetic figure becomes if you mistake it for the actual return an investor experienced.

Consider a slightly less extreme but more realistic example. A volatile small-cap fund returns the following over four years: +40%, -30%, +25%, -15%.

Arithmetic mean = (40% – 30% + 25% – 15%) / 4 = 20% / 4 = 5%

Now trace the actual compounding, starting with ₹1,00,000.

Year 1: ₹1,00,000 × 1.40 = ₹1,40,000
Year 2: ₹1,40,000 × 0.70 = ₹98,000
Year 3: ₹98,000 × 1.25 = ₹1,22,500
Year 4: ₹1,22,500 × 0.85 = ₹1,04,125

So ₹1,00,000 actually grew to ₹1,04,125 over four years. What annualized return gets you there?

Geometric mean = (1,04,125 / 1,00,000)^(1/4) – 1 = (1.04125)^0.25 – 1 ≈ 1.0%

The arithmetic mean said 5% per year. The actual compounded experience was closer to 1% per year. If a fund’s marketing material quoted the arithmetic average of annual returns rather than the geometric one, an investor would walk away with a return expectation roughly five times too optimistic, and this is exactly the kind of gap regulators and the CFA Institute’s GIPS standards are trying to close by mandating specific, standardized return calculation methods.

The General Formula for More Than Two Numbers

The two-number version is the easiest to build intuition with, but AM-GM extends to any number of non-negative values.

For n numbers a₁, a₂, …, aₙ:

(a₁ + a₂ + … + aₙ) / n ≥ ⁿ√(a₁ × a₂ × … × aₙ)

For investment returns specifically, this becomes the geometric mean return formula used to calculate compound annual growth rate, or CAGR, over multiple periods:

Geometric mean return = [(1+R₁)(1+R₂)…(1+Rₙ)]^(1/n) – 1

This is the formula behind essentially every CAGR figure quoted in a mutual fund factsheet, an annual report, or a company’s revenue growth disclosure. Whenever you see a single “average annual growth rate” figure spanning multiple years, it is almost always the geometric mean doing the work, not the arithmetic mean, precisely because the geometric version correctly accounts for compounding.

A Quick Proof, For Anyone Curious

It’s worth seeing why the inequality has to be true, rather than just accepting it as a rule.

Start with the fact that for any real numbers a and b, the square of their difference can never be negative:

(a – b)² ≥ 0

Expand this:

a² – 2ab + b² ≥ 0

Add 4ab to both sides:

a² + 2ab + b² ≥ 4ab

The left side is a perfect square:

(a + b)² ≥ 4ab

Take the square root of both sides, assuming a and b are non-negative:

a + b ≥ 2√(ab)

Divide both sides by 2:

(a + b) / 2 ≥ √(ab)

And there it is. The entire inequality falls directly out of the simple, undeniable fact that a squared number can never be negative. There’s something genuinely satisfying about how a deep, widely-applicable inequality in finance ultimately traces back to something this elementary.

Why CFA Candidates Specifically Need to Internalize This

The CFA curriculum tests arithmetic and geometric mean return calculations directly, often within Quantitative Methods and Portfolio Management, but the deeper reason to actually understand the inequality, rather than just memorize the two formulas, is that exam questions frequently test whether you know which mean is appropriate for which purpose.

Arithmetic mean return is the better estimate for expected return over a single future period, because it represents the simple average of what happened across independent periods and is the unbiased estimator looking forward one period at a time.

Geometric mean return is the correct measure for describing actual historical performance over multiple periods, because it captures the real compounding effect and tells you the single constant rate that would have produced the same ending wealth.

Using the wrong one in the wrong context is a very common exam trap, and it’s also a very common real-world mistake. A financial advisor projecting future wealth using an arithmetic mean of historical returns, without adjusting for volatility drag, will systematically overstate expected future portfolio value, sometimes by a significant margin if the underlying asset is volatile.

Volatility and the Size of the Gap

There’s a useful approximation worth knowing: the gap between arithmetic mean and geometric mean return is roughly half the variance of returns.

Geometric mean ≈ Arithmetic mean – (Variance / 2)

This is sometimes called the volatility drag or variance drag. It explains directly why higher-volatility assets show a bigger divergence between their arithmetic and geometric average returns, even when the arithmetic mean is identical across two different assets.

Two funds could both report an arithmetic mean annual return of 12%. If Fund A has low volatility and Fund B has high volatility, Fund B’s geometric mean, the one that actually matches what an investor experienced in their account, will be meaningfully lower than Fund A’s, purely because of this drag. The headline number can look identical while the lived experience is genuinely different.

Exam Perspective

For CFA and finance students, keep these points in mind.

The AM-GM inequality states arithmetic mean is always greater than or equal to geometric mean for non-negative numbers, with equality only when all the numbers are identical.

Arithmetic mean return is appropriate for estimating expected return over a single future period. Geometric mean return is appropriate for measuring actual historical compounded performance over multiple periods.

The gap between the two means widens as volatility increases, a relationship sometimes approximated as geometric mean equaling arithmetic mean minus half the variance.

The geometric mean is the basis for CAGR calculations and is the standard required under GIPS for reporting historical investment performance.

Mistaking arithmetic mean for actual compounded performance, especially with volatile assets, leads to systematically overstated expectations of wealth accumulation.

Final Thoughts

The arithmetic-geometric mean inequality looks, at first glance, like the kind of abstract mathematical curiosity that belongs in a textbook chapter nobody reads twice. It isn’t. It’s quietly responsible for one of the most common and most expensive misunderstandings in investing: believing that an average return number tells you what actually happened to your money.

It doesn’t, not unless you’re careful about which average you’re looking at. A 50% gain followed by a 50% loss feels, on paper, like it should cancel out. It doesn’t, and the inequality explains exactly why it can’t, mathematically, ever cancel out except in the one special case where nothing actually changed at all.

Every time volatility enters a return stream, the gap between what arithmetic averaging suggests and what geometric averaging confirms actually happened starts to open up. Understanding why that gap exists, rather than just memorizing two separate formulas, is what actually lets you catch a misleading return figure the moment you see one, instead of discovering the difference the hard way, in your own portfolio statement, years later.