# Arithmetic vs Geometric Return & Variance Drain Formula

> Derive the exact 0.5*sigma^2 volatility drag Taylor series and Ito Lemma drift adjustment separating arithmetic average returns from geometric CAGR.

Canonical: https://quantcalctools.com/guides/arithmetic-vs-geometric-return-half-sigma-squared-variance-drain
Last updated: 2026-11-14T19:31:11.332Z

**Publication:** QuantCalc Tools
**Category:** Cagr Irr Drawdown Compounding
**Published:** 2026-11-13T17:20:25.915Z
**Author:** Elena Rostova, FRM, CQF (FRM, Certificate in Quantitative Finance (CQF))
**Reviewer:** Not recorded

## Answer-first summary

Volatility acts as a deterministic compounding tax: derive the exact second- and fourth-order Taylor approximations linking arithmetic mean return, variance drain, skewness, and kurtosis to geometric CAGR.

## Key takeaways

- By Jensen’s Inequality on concave ln(1+R), geometric CAGR always trails arithmetic mean return by approximately sigma^2 / [2*(1+A)^2], equaling ~0.5*sigma^2 for moderate returns.
- Under geometric Brownian motion, expected wealth E[S_T] grows at mu, whereas median investor wealth grows at mu - 0.5*sigma^2; when sigma^2 > 2*mu, almost all paths lose money!
- Negative skewness (-S*sigma^3/3) and excess kurtosis (-K*sigma^4/4) add nearly 200 bps/yr of extra compounding drag beyond 0.5*sigma^2 on short-volatility carry strategies.
- Markowitz single-period optimizers require arithmetic expected returns A_i ~= G_i + 0.5*sigma_i^2; inputting geometric CAGRs double-penalizes volatile asset classes.

## Related financial entities

- **Continuous Fractional Half-Kelly Portfolio Leverage Criterion** (HALF-KELLY-LEVERAGE) — product: Continuous Fractional Half-Kelly Portfolio Leverage Criterion (HALF-KELLY-LEVERAGE) refers to an institutional financial benchmark and regulatory standard monitored by BIS Quantitative Working Paper (INTL), currently calibrated at 1.635x (Optimal Gross Leverage) as of October 2026.
- **Cornish-Fisher Skew & Kurtosis Modified Parametric VaR** (CORNISH-FISHER-VAR) — product: Cornish-Fisher Skew & Kurtosis Modified Parametric VaR (CORNISH-FISHER-VAR) refers to an institutional financial benchmark and regulatory standard monitored by GARP FRM Quantitative Standard (INTL), currently calibrated at 99.0% (Non-Normal Quantile) as of October 2026.
- **Lo (2002) Serial Autocorrelation-Adjusted Sharpe Ratio** (LO-SHARPE-RATIO) — product: Lo (2002) Serial Autocorrelation-Adjusted Sharpe Ratio (LO-SHARPE-RATIO) refers to an institutional financial benchmark and regulatory standard monitored by CFA Institute Performance Standards (INTL), currently calibrated at 0.3842 (Annualized Ratio) as of October 2026.

## Analysis

Pitchbooks that headline an "average annual return of $+12.0\%$" without disclosing return volatility omit the single most important determinant of multi-year wealth: **variance drain** (also termed **volatility drag**). By Jensen's Inequality, because the natural logarithm $\ln(1 + R)$ is a strictly concave function ($f''(R) = -1/(1+R)^2 < 0$), the geometric compound annual growth rate ($G$) of any non-constant return stream is **strictly less** than its arithmetic mean ($A$), with equality holding if and only if every single period delivers zero variance ($\sigma = 0$). Run side-by-side compounding simulations in our [CAGR & Volatility Drag Calculator](/calculators/cagr-calculator) and follow our [CAGR, IRR & Drawdown Hub](/guides/cagr-irr-volatility-drag-sharpe-sortino-drawdown-recovery-guide).

## Two Proofs of Variance Drain: Taylor Series vs Continuous Ito's Lemma

We can derive the famous $\frac{1}{2}\sigma^2$ volatility tax from both discrete-time Taylor series around the arithmetic mean $A = \mathbb{E}[R_t]$ and continuous-time stochastic calculus. Expanding $\ln(1 + R_t)$ around $R_t = A$ up to fourth central moments ($\sigma^2$, skewness $S$, Pearson kurtosis $K$) reveals not only quadratic variance drag, but also how **negative skewness** and **fat tails** compound geometric erosion:

```formula
Discrete 4th-Order Taylor Expansion of Geometric CAGR (G) around Arithmetic Mean (A):
 ln(1 + G) = E[ ln(1 + R) ]
 ~= ln(1 + A) - [ sigma^2 / (2*(1 + A)^2) ]
 + [ S * sigma^3 / (3*(1 + A)^3) ]
 - [ K * sigma^4 / (4*(1 + A)^4) ]

Standard Small-Return Quadratic Approximation (Markowitz 1959 / Messmore 1995):
 G ~= A - 0.5 * sigma^2 (or more accurately: G ~= A - sigma^2 / [2 * (1 + A)])

Worked Volatility Drag Comparison (All Strategies Have Exact Arithmetic Mean A = +12.00%):
 1. Investment-Grade Credit (sigma = 6.0%, S = -0.5, K = 5) : G_exact ~= 12.00% - 0.16% = +11.84% ($3.06M on $1M / 10y)
 2. Unhedged S&P 500 Equity (sigma = 18.0%, S = -0.6, K = 5): G_exact ~= 12.00% - 1.46% = +10.54% ($2.72M on $1M / 10y)
 3. High-Beta Biotech / Crypto (sigma = 60.0%, S = -0.4, K = 6):
 Simple 0.5*sigma^2 gives 12% - 18% = -6.0%; Denominator-adjusted: ln(1.12) - 0.36/(2*1.2544) - 0.038 = -4.12% CAGR!
 ($1,000,000 shrinks to $656,800 despite +12.0% arithmetic average return!)
```

In continuous time, if asset price $S_t$ obeys geometric Brownian motion $dS_t = \mu S_t\,dt + \sigma S_t\,dW_t$, applying Ito's Lemma to $Y_t = \ln S_t$ incorporates the quadratic variation $(dS_t)^2 = \sigma^2 S_t^2\,dt$:
$$d(\ln S_t) = \frac{1}{S_t}dS_t - \frac{1}{2S_t^2}(dS_t)^2 = \left(\mu - \frac{1}{2}\sigma^2\right)dt + \sigma\,dW_t$$
Thus while **expected terminal wealth** $\mathbb{E}[S_T] = S_0 e^{\mu T}$ grows at the arithmetic drift $\mu$ (pulled upward by a tiny fraction of astronomically lucky right-tail sample paths), the **median investor's wealth** $\text{Median}(S_T) = S_0 e^{(\mu - \frac{1}{2}\sigma^2)T}$ grows strictly at the variance-drained geometric rate!

![Causal chain showing how symmetric arithmetic gains and losses create asymmetric base shrinkage, Ito drift reduction, and median vs mean wealth divergence](/api/figure?type=flow&title=Why%20Volatility%20Taxes%20Compounding%3A%20Base%20Shrinkage%20%26%20Lognormal%20Skew&subtitle=Tracing%20how%20symmetric%20%2B%2F-%20percentage%20swings%20erode%20geometric%20wealth%20while%20skewing%20expected%20terminal%20wealth.&steps=1.%20Concave%20Log%20Utility%3A%3Aln%281%2BR%29%20curves%20downward%3A%20-30%25%20hurts%20%28-0.357%29%20far%20more%20than%20%2B30%25%20helps%20%28%2B0.262%29%7C2.%20Ito%20Drift%20Haircut%3A%3AMedian%20log-wealth%20compounds%20at%20mu%20-%200.5*sigma%5E2%2C%20not%20arithmetic%20mu%7C3.%20Higher-Moment%20Drag%3A%3ANegative%20skew%20%28-S*sigma%5E3%2F3%29%20and%20kurtosis%20%28-K*sigma%5E4%2F4%29%20deepen%20drag%7C4.%20Mean%20vs%20Median%20Split%3A%3AWhen%20sigma%5E2%20%3E%202*mu%2C%20E%5BS_T%5D%20-%3E%20infinity%20while%20Median%28S_T%29%20-%3E%20%240!&metrics=18%25%20VOL%20EQUITY%20DRAG%3A%3A-146%20bps%2Fyr%3A%3ACuts%2012.0%25%20arith%20to%2010.54%25%20CAGR%7C40%25%20VOL%20TECH%20DRAG%3A%3A-714%20bps%2Fyr%3A%3ACuts%2012.0%25%20arith%20to%204.86%25%20CAGR%7C60%25%20VOL%20CRYPTO%20DRAG%3A%3A-1%2C612%20bps%2Fyr%3A%3AFlips%20%2B12%25%20arith%20into%20-4.12%25%20loss!)

## The Lognormal Mean-vs-Median Paradox at High Volatility

Notice a striking mathematical paradox when $\sigma = 60\%$ and $\mu = 12\%$: the continuous geometric growth rate $\mu - \frac{1}{2}\sigma^2 = 0.12 - 0.18 = -6.0\%$ is **negative**, meaning as $T \to \infty$, the probability that an investor loses money approaches $100\%$ ($S_T \to 0$ almost surely), EVEN THOUGH expected wealth $\mathbb{E}[S_T] = S_0 e^{+0.12 T} \to \infty$! How can both statements be mathematically true simultaneously? Because at $T = 20$ years, $94\%$ of simulated investors lose almost everything, while $0.1\%$ of simulated lottery paths multiply $50{,}000\times$, dragging the cross-sectional average upward. An individual institution lives along **one single time-series trajectory**, not across an ensemble of parallel universes—which is why [Continuous Half-Kelly Leverage Sizing](/entities/half-kelly-leverage-continuous-fractional-half-kelly-portfolio-leverage-criterion) maximizes $\mathbb{E}[\ln S_T]$ rather than $\mathbb{E}[S_T]$.

## Higher-Moment Drag: Why Option-Selling Strategies Collapse Faster Than $0.5\sigma^2$

For strategies with severe negative skewness ($S = -2.5$) and fat kurtosis ($K = 14$), such as short naked index puts or carry trades modeled with [Cornish-Fisher Modified VaR](/entities/cornish-fisher-var-cornish-fisher-skew-kurtosis-modified-parametric-var), the quadratic approximation $A - \frac{1}{2}\sigma^2$ is dangerously optimistic. At $\sigma = 25\%$, $A = 10\%$, $S = -2.5$, and $K = 14$, the third- and fourth-order Taylor terms subtract an additional $-1.02\%$ and $-0.94\%$ per year—nearly **doubling** variance drain from $-2.84\%$ to **$-4.80\%$**!

| Asset Class / Strategy Profile | Arithmetic Mean ($A$) | Annual Vol ($\sigma$) | Skew ($S$) / Kurt ($K$) | Exact Geometric CAGR ($G$) | 10-Year Wealth on $\$1.00\text{M}$ |
| :--- | :--- | :--- | :--- | :--- | :--- |
| **Short-Duration T-Bills / Cash** | $4.50\%$ | $0.8\%$ | $0.0$ / $3.0$ | **$4.50\%$** ($-0\text{ bps}$ drag) | $\$1{,}552{,}970$ |
| **Core Global 60/40 Balanced** | $8.50\%$ | $10.5\%$ | $-0.4$ / $4.2$ | **$7.98\%$** ($-52\text{ bps}$ drag) | $\$2{,}155{,}010$ |
| **S&P 500 Unhedged Equity** | $11.50\%$ | $17.5\%$ | $-0.6$ / $5.0$ | **$10.09\%$** ($-141\text{ bps}$ drag) | $\$2{,}615{,}180$ |
| **Short-Put Volatility Carry** | $11.50\%$ | $24.0\%$ | $-2.8$ / $15.0$ | **$6.92\%$** ($-458\text{ bps}$ drag) | $\$1{,}952{,}490$ |
| **3x Leveraged Nasdaq / Altcoin** | $22.00\%$ | $72.0\%$ | $-0.5$ / $6.5$ | **$-1.85\%$** ($-2{,}385\text{ bps}$ drag) | $\$829{,}640$ |

![Geometric CAGR curve peaking and turning negative as volatility rises from 5 percent to 70 percent for a fixed 12 percent arithmetic return](/api/figure?type=bars&title=Geometric%20CAGR%20Erosion%20Across%20Volatility%20Regimes%20%28Fixed%20A%20%3D%20%2B12.0%25%29&subtitle=Demonstrating%20how%20rising%20volatility%20sigma%20turns%20a%20%2B12.0%25%20arithmetic%20return%20into%20negative%20multi-year%20compounding.&steps=Sigma%20%3D%2010%25%20%28Low%20Vol%29%3A%3AG%20%3D%20%2B11.55%25%20CAGR%20%28-45%20bps%20variance%20drain%29%7CSigma%20%3D%2025%25%20%28Growth%20Eq%29%3A%3AG%20%3D%20%2B9.15%25%20CAGR%20%28-285%20bps%20variance%20drain%29%7CSigma%20%3D%2045%25%20%283x%20ETF%29%3A%3AG%20%3D%20%2B2.68%25%20CAGR%20%28-932%20bps%20variance%20drain%29%7CSigma%20%3D%2060%25%20%28Speculative%29%3A%3AG%20%3D%20-4.12%25%20CAGR%20%28-1%2C612%20bps%20variance%20drain%20-%3E%20Net%20Loss!%29&metrics=ZERO-CAGR%20VOL%20CEILING%3A%3A50.8%25%20Vol%3A%3AWipes%20out%20entire%20%2B12%25%20arith%20edge!%7CSHORT-VOL%20SKEW%20EXTRA%20TAX%3A%3A-196%20bps%2Fyr%3A%3A3rd%20%26%204th%20moment%20tail%20penalty%7CREBALANCING%20ANTIDOTE%3A%3A%2B0.5*Var%20Harvest%3A%3ASee%20Shannon%20Demon%20spoke%20guide%7CFORECAST%20RULE%3A%3AForecast%20Arith%3A%3AConvert%20G%20-%3E%20A%20%2B%200.5*s%5E2%20for%20MVO)

## Mean-Variance Optimization Input Trap: Forecasting Arithmetic vs Geometric Returns

A classic institutional modeling bug occurs when asset allocators plug historical **geometric CAGRs** directly into a single-period Markowitz Mean-Variance Optimizer alongside a covariance matrix $\Sigma$. Because single-period expected portfolio return is linear in weights ($\mathbb{E}[R_p] = w^\top A$), feeding geometric CAGRs into the optimizer **double-counts volatility drag**—once inside the historical CAGR input and a second time inside the quadratic risk penalty $\frac{\lambda}{2} w^\top \Sigma w$! Always convert long-horizon geometric capital-market assumptions back to arithmetic expected returns via $A_i \approx G_i + \frac{1}{2}\sigma_i^2$ before running MVO or [rebalancing dispersion models](/guides/constant-weight-rebalancing-bonus-shannon-demon-dispersion-math), and cross-check path decay in [daily leveraged ETF beta slippage](/guides/daily-leveraged-etf-beta-slippage-volatility-decay-choppy-markets).

## Institutional Checklist for Auditing Fund Return Disclosures

1. **Demand Paired $(A, G, \sigma)$ Reporting**: Under [UK FCA COBS performance disclosure rules](/regulator-index/fca) and GIPS, reject any marketing tear-sheet quoting "average annual return" without cumulative geometric CAGR and annualized standard deviation.
2. **Apply Denominator-Scaled Drag Above $20\%$ Vol**: Replace naive $A - 0.5\sigma^2$ with $\exp(\ln(1+A) - \sigma^2/[2(1+A)^2]) - 1$ whenever $\sigma > 20\%$ or $|A| > 15\%$, and verify risk-adjusted efficiency in our [Sharpe & Sortino Calculator](/calculators/sharpe-sortino-calculator) and [datasets hub](/datasets).

## Frequently asked questions

### When should I use arithmetic average return instead of geometric CAGR?

Use the arithmetic mean ONLY as an unbiased estimator of an unknown single-period expected return (for example, forecasting next year’s 1-year expected return from stationary historical data or feeding single-period Markowitz MVO). Use geometric CAGR whenever measuring realized multi-period wealth growth.

### Why does the simple G ~= A - 0.5*sigma^2 approximation lose accuracy above 40% volatility?

Because 0.5*sigma^2 drops both the (1+A)^2 denominator scaling in the Taylor expansion around A and higher-order skewness/kurtosis terms. For example, alternating +100% and -60% gives A = +20%, sigma = 80%, and exact G = sqrt(2.0 * 0.4) - 1 = -10.56%, whereas A - 0.5*sigma^2 predicts 20% - 32% = -12.00%.

### Can reducing portfolio volatility increase terminal wealth even if arithmetic return drops slightly?

Yes! Cutting annualized volatility from 30% to 15% saves 0.5*(0.09 - 0.0225) = +3.375% (+337.5 bps) per year in variance drain. Even if hedging or diversifying trims arithmetic mean return by -150 bps, net geometric CAGR rises by +187.5 bps per year!

### Does rebalancing between volatile assets reduce or eliminate individual asset variance drain?

Yes—rebalancing harvests a "diversification return" equal to half the average asset variance minus half the portfolio variance. Uncorrelated volatile assets whose standalone geometric CAGRs are 0% can compound at +4% to +6% CAGR when rebalanced to fixed weights (Shannon’s Demon).


## Source URLs

- https://www.jstor.org/stable/4479847
- https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2025/quantitative-investment-analysis
- https://www.gipsstandards.org/standards/gips-standards-for-firms/
- https://rpc.cfainstitute.org/en/research/financial-analysts-journal

## Citation

- APA: Elena Rostova, FRM, CQF. (2026). Arithmetic vs Geometric Mean Return: 0.5*Sigma^2 Variance Drain Derivation. QuantCalc Tools. https://quantcalctools.com/guides/arithmetic-vs-geometric-return-half-sigma-squared-variance-drain
- Canonical article: [Arithmetic vs Geometric Mean Return: 0.5*Sigma^2 Variance Drain Derivation](https://quantcalctools.com/guides/arithmetic-vs-geometric-return-half-sigma-squared-variance-drain)
- Site research index: [QuantCalc Tools](https://quantcalctools.com/llms-full.txt)
