The Sharpe Ratio Trap — Why Same Returns Get Different Scores
The same mean return can produce different Sharpe ratios when variability differs. With hypothetical annualized mean returns of 12% and a 3% risk-free return, standard deviations of 6% and 18% produce ratios of 1.5 and 0.5. This article explains that calculation and uses reproducible return sequences to show why the same Sharpe need not mean the same pattern of losses.
1. Definition — excess return and its variability
A historical Sharpe ratio starts with each period’s portfolio return minus its matching risk-free return. Divide the arithmetic mean of those excess returns by their standard deviation.
Sharpe = mean excess return (d̄) / standard deviation of excess returns (σd)
- Rₚ,ₜ: portfolio return in each period
- Rf,ₜ: risk-free return for the same currency and period
- σd: standard deviation of the same excess-return series. With Rf constant each period, this equals the standard deviation of portfolio returns.
The numerator is an arithmetic mean, not compound annual growth (CAGR). The hypothetical return sequences below use population standard deviation, dividing squared deviations by n. If analysis uses sample standard deviation with n−1 instead, state that convention and compare like with like. A zero denominator does not produce a finite Sharpe ratio.
For m periods per year, annualizing by multiplying the mean by m, the standard deviation by √m and Sharpe by √m assumes stable, uncorrelated period excess returns and an additive-return approximation. Monthly data use m=12. Compounding and serial correlation can invalidate this scaling. See William F. Sharpe, “The Sharpe Ratio” (1994) for the definition and time dependence.
2. Same 12% mean — different scores
These are hypothetical inputs: annualized arithmetic mean return 12%, a risk-free return constant each period and annualized to 3%, and consistently annualized standard deviations of 6% and 18%. The inputs are not calculated from a particular monthly return path or an actual fund.
| Metric | Lower-volatility example | Higher-volatility example |
|---|---|---|
| Arithmetic mean return (annualized Rp) | +12% | +12% |
| Excess-return standard deviation (annualized σd) | 6% | 18% |
| Sharpe = (12 − 3) / σ | 1.5 | 0.5 |
With the same positive mean excess return of 9 percentage points, one-third the standard deviation gives three times the Sharpe ratio. This describes the ratio, not every investment risk or whether the result came from skill.
A separate hypothetical series earning exactly +1% each month compounds to (1.01¹² − 1) × 100 ≈ 12.68% over 12 months. Its observed monthly standard deviation is zero. If Rf is also constant each month, Sharpe’s denominator is zero and no finite ratio can be calculated. That series cannot represent the table’s 6% standard deviation.
3. What the numbers mean
For a positive standard deviation, the following are mathematical interpretations, not universal fund ratings or percentile rankings.
| Sharpe | Interpretation |
|---|---|
| < 0 | Mean excess return is negative; this need not mean a loss every period or a loss of principal. |
| 0 | Mean excess return is zero. |
| 0.5 | Mean excess return is half its standard deviation. |
| 1.0 / 2.0 | Mean excess return is one / two times its standard deviation. |
For a sourced historical example, Frazzini, Kabiller and Pedersen’s “Buffett’s Alpha” (November 2013 draft, Table 2) reports Berkshire Hathaway’s annualized Sharpe as 0.76 for 1976–2011. That figure belongs to the paper’s period and methodology; it does not cover returns through 2023 or mean that every year achieved 0.76.
4. The trap — volatility ≠ risk
Standard deviation includes observations above and below the mean. Two sequences can have equal means and standard deviations while showing different patterns of losses.
These are hypothetical returns over three equal-length periods. Rf is 0% each period, and the results are not annualized.
| Series | Period returns | Arithmetic mean | Population standard deviation | Sharpe |
|---|---|---|---|---|
| A | +1%, +1%, +10% | 4% | 4.24264% | 0.94281 |
| B | −2%, +7%, +7% | 4% | 4.24264% | 0.94281 |
B: σ = √[(36 + 9 + 9) / 3] = √18
Sharpe = 4 / √18 ≈ 0.94281
The ratio uses the unrounded √18; only displayed results are rounded. Series A has no negative period, while B does. Equal means and standard deviations do not imply equal compound returns or loss patterns.
Sortino Ratio measures shortfalls below a minimum acceptable return (MAR). Calmar Ratio compares compound annual return with the absolute maximum drawdown over the same measurement period.
| Ratio | Denominator | Use |
|---|---|---|
| Sharpe | Excess-return standard deviation | Mean excess return relative to overall variability |
| Sortino | Downside deviation relative to MAR | Returns falling below a chosen target |
| Calmar | Absolute maximum drawdown over the same period | Compound annual return relative to the largest decline |
Using downside deviation = √[Σmin(0, Rₜ − MAR)² / n] and MAR=0%, series A has zero downside deviation, so a finite Sortino ratio cannot be calculated. Its numerator under this convention is mean return minus MAR. No observed loss does not mean no future loss. Calmar likewise has no finite ratio when maximum drawdown is zero. State the target, measurement period and calculation convention.
5. Other practical pitfalls
① Short measurement window
With few observations, both the mean and standard deviation can be sensitive to individual returns. A longer history can span different market conditions or changes in strategy, so no universal minimum number of years establishes reliability. A historical ratio is not a forecast.
② Risk-free rate choice
Changing Rf changes excess returns and Sharpe; the size of the effect depends on volatility and the other inputs. Use matching currencies and holding periods. A long-term bond yield or policy rate is not automatically the realized risk-free return for a short period.
③ Return distribution
Normal returns are not required to calculate Sharpe. However, mean and standard deviation do not fully describe asymmetry or rare large losses. A high ratio during a period with no observed tail loss does not establish that such losses are unlikely.
④ Gross vs net
Fees and trading costs affect returns. Check which costs a reported series includes and use consistent assumptions when comparing ratios.
Summary
- Sharpe = mean excess return / its standard deviation. State the return frequency and annualization assumptions.
- With the same positive mean, greater variability lowers Sharpe. A zero denominator produces no finite ratio.
- The same Sharpe can accompany different losses. It does not guarantee future safety.
- Sortino and Calmar measure different aspects of risk. Their conventions and zero denominators also matter.
- Watch the 4 pitfalls: window, Rf, distribution, fees.