Compounding Isn’t One Thing: The Real Difference Between Arithmetic and Geometric Returns

Compounding looks simple—until you average returns the wrong way.

The two “averages” investors casually mix up

In everyday investing talk, “average return” can mean at least two different things:

Arithmetic average return: the plain average of periodic returns.
Geometric average return: the compounded average rate that actually matches the ending wealth.

They often look close when returns are steady. They can diverge sharply when returns are volatile, and that divergence is not a small technicality—it’s the math of why investors with the same arithmetic average can end up with different outcomes.

Let’s define both precisely.

Arithmetic return: the average of the path’s steps

Suppose you have returns ( r_1, r_2, \dots, r_n ) over ( n ) periods (monthly, yearly, etc.). The arithmetic mean is:

[ \bar{r}A = \frac{1}{n}\sum{t=1}^{n} r_t ]

This is the “expected one-period return” estimator you’ll see in many textbooks and models. If you pick a random period from your history, ( \bar{r}_A ) describes the average step.

Geometric return: the average rate that reproduces compounding

Your wealth grows multiplicatively:

[ \frac{W_n}{W_0} = \prod_{t=1}^{n} (1+r_t) ]

The geometric mean return ( \bar{r}_G ) is the single constant rate that would produce the same final wealth:

[ (1+\bar{r}G)^n = \prod{t=1}^{n} (1+r_t) ]

So:

[ \bar{r}G = \left(\prod{t=1}^{n} (1+r_t)\right)^{1/n} - 1 ]

This is essentially the engine behind CAGR (compound annual growth rate). When people ask “what did I actually compound at?”, they are asking for the geometric return.

A simple example where the arithmetic mean lies to your intuition

Consider a two-year investment:

Year 1: +50%
Year 2: -50%

Arithmetic mean:

[ \bar{r}_A = \frac{0.50 + (-0.50)}{2} = 0 ]

That seems to suggest “flat.” But the wealth path says otherwise:

[ W_2 = W_0 (1.5)(0.5) = 0.75 W_0 ]

That’s a 25% loss over two years. The geometric mean shows it:

[ \bar{r}_G = \sqrt{1.5 \cdot 0.5} - 1 = \sqrt{0.75} - 1 \approx -0.13397 ]

So the compound rate is about -13.4% per year, not 0%.

This is the first key difference:

Arithmetic mean averages the returns.
Geometric mean averages the growth factors ((1+r)).

And because growth factors multiply, volatility reshapes the end result.

Why volatility creates a gap: the “volatility drag” in plain math

A useful way to see the difference is to move into log space.

Define log returns ( g_t = \ln(1+r_t) ). Then:

[ \ln\left(\frac{W_n}{W_0}\right) = \sum_{t=1}^{n} \ln(1+r_t) = \sum_{t=1}^{n} g_t ]

The geometric mean is tied to the average log return:

[ \ln(1+\bar{r}G) = \frac{1}{n}\sum{t=1}^{n} \ln(1+r_t) ]

Now compare that to the arithmetic mean. For small-ish returns, we can use the approximation:

[ \ln(1+r) \approx r - \frac{r^2}{2} ]

Take the average:

[ \overline{\ln(1+r)} \approx \bar{r}_A - \frac{1}{2}\overline{r^2} ]

And since ( \overline{r^2} = \sigma^2 + \bar{r}_A^2 ), a common rule-of-thumb appears:

[ \bar{r}_G \approx \bar{r}_A - \frac{1}{2}\sigma^2 ]

(where ( \sigma^2 ) is variance of periodic returns, in decimal form)

This isn’t perfect in all regimes, but it captures the central reality:

Higher volatility lowers geometric returns even if arithmetic average stays the same.

That downward adjustment is often called volatility drag. It’s not a fee and not a behavioral mistake. It’s built into multiplication.

Same arithmetic average, different compounding: a side-by-side comparison

Imagine two portfolios over four years:

Portfolio Smooth

+8%, +8%, +8%, +8%

Portfolio Wild

+28%, -12%, +28%, -12%

Arithmetic mean for both:

Smooth: ( (8+8+8+8)/4 = 8% )
Wild: ( (28-12+28-12)/4 = 8% )

Now compute ending wealth with (W_0 = 1):

Smooth: ( 1.08^4 \approx 1.3605 )
Wild: ( 1.28 \cdot 0.88 \cdot 1.28 \cdot 0.88 )

Notice ( 1.28 \cdot 0.88 = 1.1264 ). So:

Wild: ( 1.1264^2 \approx 1.2688 )

Both averaged 8% arithmetically, but the smoother path compounds to a higher ending value. The geometric means:

Smooth: ( \bar{r}_G = 8% ) (because it’s constant)
Wild: ( \bar{r}_G = 1.2688^{1/4} - 1 \approx 6.1% )

The “extra” variability shaved off roughly 1.9% per year of compound growth, even though the arithmetic average looked identical.

Compounding is path-dependent; arithmetic averages are not

A crucial conceptual split:

Arithmetic mean is path-agnostic (it doesn’t care about the order).
Wealth outcomes are path-dependent (the order and size of gains/losses matters through multiplication).

To see order effects, use the same two returns in different sequences when adding cash flows (like retirement withdrawals). Even if the investment’s returns are identical as a set, the timing of negative years can be devastating when you’re pulling money out—a phenomenon tied to sequence of returns risk. This sits on top of the basic arithmetic vs geometric gap, amplifying it in real life.

Where arithmetic returns belong (and where they don’t)

Arithmetic averages are not “wrong.” They are just answers to a different question.

Arithmetic return is useful when:

You’re modeling a single-period expectation, like “What’s the expected return next year?”
You’re comparing short-horizon payoffs or forecasting one step ahead.
You’re working in classic mean-variance frameworks where “expected return” is a one-period concept.

But it becomes misleading when you ask long-horizon questions like:

“What will my portfolio be worth in 20 years?”
“What’s my long-term growth rate?”
“How fast am I compounding?”

Those questions live in geometric space.

Geometric return is useful when:

You’re summarizing historical performance as a single growth rate (CAGR).
You’re comparing managers over multi-year windows.
You care about wealth accumulation and compounding, not just average periodic behavior.

The deeper point is that investors routinely take a one-period statistic (arithmetic mean) and mentally compound it. That mental move is exactly where the trouble starts.

The compounding difference in one inequality

For nonnegative wealth and returns above -100%, the geometric mean is never above the arithmetic mean:

[ \bar{r}_G \le \bar{r}_A ]

Equality holds only when returns are constant (no volatility) or degenerate in special ways.

This inequality is just the classic fact that the geometric mean is less than or equal to the arithmetic mean. In investing it becomes a statement about performance reality:

Your “average return” cannot be compounded as if it were a constant return unless it’s geometric.

How drawdowns mathematically punish compounding

Losses require larger gains to recover. This is not a motivational poster; it’s a direct consequence of multiplication.

If you lose (L%), your wealth becomes (1-L). To get back to 1, you need a gain (G) such that:

[ (1-L)(1+G) = 1 \Rightarrow G = \frac{L}{1-L} ]

So:

Down 10% requires +11.11% to recover.
Down 20% requires +25%.
Down 50% requires +100%.

Arithmetic averages don’t “see” this curvature. Geometric compounding does, because it is built from the product of growth factors.

_{Photo by Ibrahim Rifath on Unsplash}

A practical investing-math lens: expected wealth vs expected return

There’s another subtlety that trips up even careful readers: the expectation of a product is not the product of expectations.

If returns are random, then terminal wealth after (n) periods is:

[ W_n = W_0 \prod_{t=1}^{n} (1+R_t) ]

The expected wealth is:

[ \mathbb{E}[W_n] = W_0 , \mathbb{E}\left[\prod_{t=1}^{n} (1+R_t)\right] ]

Investors often approximate this by compounding the arithmetic mean:

[ W_0 (1+\mathbb{E}[R])^n ]

But this shortcut ignores dispersion and correlations through time. The geometric mean aligns more naturally with typical realized growth because it is tied to average log growth, which respects multiplication.

This distinction matters when you read performance projections. A brochure might say “expected return 8%,” then show ( (1.08)^{30} ) to draw a rosy curve. Unless that 8% is a geometric estimate or the path is assumed nearly deterministic, the picture is optimistic relative to median outcomes.

A two-coin-flip thought experiment that makes it click

Imagine an investment that each year does one of two things with equal probability:

+20% or -10%

Arithmetic mean per year:

[ \bar{r}_A = 0.5(0.20) + 0.5(-0.10) = 0.05 = 5% ]

Geometric mean per year:

[ \bar{r}_G = \sqrt{1.2 \cdot 0.9} - 1 = \sqrt{1.08} - 1 \approx 3.92% ]

Even though the expected one-year return is 5%, the compound rate is closer to 3.9%. Over decades, that gap is enormous.

And notice what caused it: not “bad luck,” not fees, not market manipulation—just the mathematical effect of having some down years.

Converting between them (carefully) in real analysis

In professional settings, analysts sometimes need to go from arithmetic inputs (like forecasted expected return and volatility) to an estimate of geometric growth.

A commonly used approximation (for annual returns, moderate volatility) is:

[ \text{Geometric} \approx \text{Arithmetic} - \frac{1}{2}\sigma^2 ]

Example: if expected arithmetic return is 8% and volatility is 20%:

( \sigma^2 = 0.20^2 = 0.04 )
Half variance = 0.02
Estimated geometric ≈ 8% − 2% = 6%

That’s not a minor haircut. It changes the story.

Two cautions matter:

This approximation relies on log-return reasoning and behaves best when returns aren’t extreme.
Volatility is not constant in reality, and returns aren’t always close to normal; fat tails and crashes make compounding harsher than clean formulas suggest.

Still, it gives a disciplined way to understand why “high expected return” and “high volatility” can coexist with disappointing compounded wealth.

The rebalancing angle: when volatility can help and when it can’t

Investors sometimes hear that volatility “helps” through rebalancing. There’s truth there, but it lives in a narrower box than many assume.

Within a diversified portfolio, if assets move differently, periodic rebalancing can capture a “rebalancing premium” (some call it volatility harvesting). You’re systematically trimming what went up and adding to what went down.
This effect depends on imperfect correlations and the presence of multiple assets. It is not magic; it’s a disciplined trading rule inside a portfolio.

But for a single asset, volatility almost always reduces geometric growth relative to arithmetic, because there’s nothing to rebalance against. You just ride the multiplicative roller coaster.

So you can hold two ideas at once without contradiction:

Volatility in a single return stream drags on compounding.
Volatility across multiple imperfectly correlated assets can create opportunities when coupled with rebalancing—even then, costs, taxes, and trend persistence can reduce or erase the benefit.

How this shows up in performance reporting and fund marketing

Performance tables often report annual returns and then show an “average annual return.” You have to check which one it is.

If it’s an arithmetic average, it tells you about the average year.
If it’s a geometric average / CAGR, it tells you what you compounded at.

A common marketing trick is unintentional rather than malicious: people see a set of good years, compute an arithmetic average, and treat it as a sustainable compound growth rate. In volatile assets—growth stocks, crypto, concentrated thematic funds—that mistake is amplified.

When comparing two managers, geometric return is typically the fairer summary of what happened to a dollar invested, but you still need context:

Did the manager take more risk?
Were drawdowns deeper?
Was the track record short (making the geometric estimate noisy)?
Did cash flows and timing affect investor experience (dollar-weighted returns)?

Arithmetic vs geometric is foundational, but it doesn’t replace risk analysis; it clarifies the compounding part of risk.

The investor’s rule of thumb: match the metric to the question

If your question is framed like a single period:

“What is the expected return next year?”
“What is the average quarterly return?”

Then arithmetic returns are appropriate.

If your question is framed like a multi-period wealth outcome:

“What did I compound at over 10 years?”
“What constant rate would match my ending value?”
“How should I think about long-run growth?”

Then geometric returns belong at the center.

The trap is using arithmetic language for geometric realities. People do it because arithmetic averages feel intuitive and because “8% average return” is easy to say out loud. But your portfolio doesn’t grow by averaging; it grows by multiplying.

A concrete checklist for analyzing any return series

When you look at returns—your own, a fund’s, or an asset class—walk through these steps:

Compute arithmetic mean ( \bar{r}_A ) to understand typical period performance.
Compute geometric mean/CAGR ( \bar{r}_G ) to understand compounded growth.
Measure volatility (standard deviation) to anticipate the gap between the two.
Inspect drawdowns because large losses dominate compounding math.
Ask about horizon and cash flows because contributions/withdrawals make sequence matter.

In practice, an investor who internalizes this framework stops being surprised by outcomes like “the fund averaged 10% but my money didn’t double when I expected.” The answer is usually sitting in the arithmetic–geometric gap plus the timing of gains and losses.

Why this distinction keeps paying dividends in decision-making

Many investing decisions are really decisions about compounding under uncertainty:

Should you concentrate or diversify?
How much leverage is too much?
How aggressive should you be near retirement?
Is a high-volatility strategy worth it if the arithmetic average looks attractive?

Arithmetic returns can seduce you into thinking only about the center of the distribution. Geometric returns force you to respect the path your wealth must travel. And since wealth can’t go below zero (and since big losses require outsized gains to recover), compounding is inherently asymmetric.

Once you see that, you start interpreting “average return” claims with a sharper eye: not as a single number that magically compounds, but as a statistic that needs its correct partner—geometric growth—before it becomes a story about real money over time.

External Links

Geometric vs. Arithmetic Returns | Explained with an Investing … Arithmetic Returns vs. Geometric Returns - Bogleheads.org The difference between arithmetic and geometric investment returns Arithmetic, Geometric, and Dollar-Weighted Returns and Indices Geometric vs Arithmetic Mean In The Wild