Sharpe Ratio

The formula in one line

Sharpe ratio  =  (R_p − R_f)  /  σ_p

Where R_p is the portfolio's return, R_f the risk-free rate (typically the 3-month Treasury yield), and σ_p the standard deviation of the portfolio's returns. The numerator is the excess return — what you earned above the safe alternative. The denominator is the volatility — how much pain you endured to earn it. The ratio is reward per unit of risk.

Why dividing by volatility matters

Suppose two portfolios both delivered a 12% return last year. Manager A's monthly returns were +2%, +1%, +1%, +1% … steady wins. Manager B's were +15%, −10%, +20%, −18% … wild swings that happened to average out. Same end result, but very different experiences for the investor — and very different probabilities that next year will resemble last year.

Portfolio	Return	Volatility (σ)	R_f	Sharpe
Steady Eddie	12%	5%	3%	(12 − 3) / 5 = 1.80
Wild Wanda	12%	20%	3%	(12 − 3) / 20 = 0.45
S&P 500	10%	16%	3%	(10 − 3) / 16 = 0.44

Steady Eddie's Sharpe is four times Wild Wanda's. If you could double Eddie's leverage, you'd get 21% return at 10% vol — Sharpe still 1.80 — versus Wanda's 0.45 at any leverage. The Sharpe ratio is the slope of the line from the risk-free point to the portfolio's (vol, return) coordinate. Steeper slope = better risk-adjusted return. This is the geometric content of mean-variance analysis.

Worked example: Daily Sharpe of a momentum strategy

You backtest a momentum strategy over 2,520 trading days (10 years). Average daily return is 0.05% (12.6% annualized), daily volatility is 1.0% (15.8% annualized), and the average daily T-bill yield was 0.012% (3% annualized).

Daily excess return = 0.05% − 0.012% = 0.038%
Daily volatility = 1.0%
Daily Sharpe = 0.038 / 1.0 = 0.038

Annualized Sharpe = 0.038 × √252 = 0.038 × 15.87 = 0.60

Sharpe 0.60 over 10 years is meaningfully better than buy-and-hold S&P (≈ 0.44). The √252 multiplier converts daily Sharpe to annual — assuming returns are independent across days. For weekly data multiply by √52, for monthly by √12. The square-root scaling reflects that vol grows with √T while return grows linearly with T.

Origin — Sharpe, the LSE, and the reward-to-variability ratio

William F. Sharpe published "Mutual Fund Performance" in the Journal of Business in January 1966, two years after his CAPM paper. He called it the "reward-to-variability ratio." Sharpe ranked U.S. mutual funds by this measure and found, mortifyingly for the active-management industry, that the average active fund underperformed a passive market portfolio after fees. The Sharpe ratio was the first popular tool that quantified risk-adjusted performance — before it, managers simply trumpeted raw returns and hid leverage and vol behind aggregate numbers.

Sharpe shared the 1990 Nobel Prize in Economics with Harry Markowitz and Merton Miller for his foundational work in modern portfolio theory. The Sharpe ratio sits at the practical intersection of Markowitz's mean-variance optimization and Sharpe's CAPM — it identifies the portfolio with the steepest line from the risk-free rate, which in equilibrium is the market portfolio.

Sharpe vs alternative performance measures

Metric	Numerator	Denominator	Year	Best for
Sharpe ratio	R_p − R_f	σ_p (total vol)	1966 (Sharpe)	Stand-alone portfolios; balanced reward vs risk
Treynor ratio	R_p − R_f	β_p (market beta)	1965 (Treynor)	Diversified portfolios within a larger portfolio
Jensen's alpha	R_p − [R_f + β·ERP]	(absolute, not a ratio)	1968 (Jensen)	Manager skill above CAPM expectation
Sortino ratio	R_p − R_f	Downside deviation	1980s (Sortino)	Strategies with positive skew or asymmetric returns
Information ratio	R_p − R_b	σ(R_p − R_b)	1985 (Treynor-Black)	Active management vs benchmark
Calmar ratio	R_p (annualized)	Max drawdown	1991 (Young)	Trend-following CTAs; recovery emphasis
Omega ratio	P(R > τ) probability-weighted gains	Probability-weighted losses	2002 (Keating-Shadwick)	Non-normal returns; full distribution shape

The Sharpe ratio's longevity comes from its analytical simplicity and the fact that it is consistent with mean-variance optimization. The Sortino and Omega variants matter when returns deviate sharply from normality — selling out-of-the-money puts has a beautiful Sharpe ratio right up until the moment it doesn't.

Annualization in practice

Native-frequency Sharpe values are inconveniently small. The √T scaling produces the familiar annualized numbers you see in fund tearsheets:

Data frequency	Periods/year	Multiplier	Daily Sharpe → annual
Daily	252	√252 ≈ 15.87	0.04 daily ≈ 0.63 annual
Weekly	52	√52 ≈ 7.21	0.08 weekly ≈ 0.58 annual
Monthly	12	√12 ≈ 3.46	0.15 monthly ≈ 0.52 annual
Quarterly	4	√4 = 2	0.25 quarterly ≈ 0.50 annual

The square-root assumption breaks if returns are autocorrelated. Hedge funds reporting monthly, but with smoothed marks on illiquid private positions, have artificially low monthly vol — their reported Sharpe ratios are systematically biased upward. Andrew Lo's 2002 Financial Analysts Journal paper provides an autocorrelation-adjustment widely used by sophisticated fund-of-funds allocators.

Implementation in 10 lines of Python

import numpy as np

def sharpe_ratio(returns, rf_per_period, periods_per_year=252):
    """returns: array of period returns (e.g. daily)
       rf_per_period: scalar or array, same frequency
       Returns: annualized Sharpe ratio."""
    excess = np.asarray(returns) - rf_per_period
    if excess.std(ddof=1) == 0:
        return np.nan
    sharpe_period = excess.mean() / excess.std(ddof=1)
    return sharpe_period * np.sqrt(periods_per_year)

# Example
import pandas as pd
df = pd.read_csv('returns.csv', parse_dates=['date']).set_index('date')
sr = sharpe_ratio(df['return'], rf_per_period=0.0001)
print(f"Annualized Sharpe: {sr:.2f}")

Two implementation details bite traders. First, use ddof=1 for sample standard deviation — population σ underestimates true vol on finite samples. Second, the risk-free rate must match the return frequency. A common bug is to subtract annual R_f from daily returns; the resulting Sharpe is wildly wrong.

Where Sharpe fails

• Non-normal returns. Sharpe assumes σ captures all risk. Selling deep out-of-the-money puts produces a steady stream of small gains punctuated by occasional catastrophic losses — its Sharpe pre-crash is excellent. Long Term Capital Management ran a beautiful Sharpe right up to August 1998.
• Autocorrelated returns. Illiquid assets with smoothed marks show artificially low volatility. Real-estate funds, private equity, and unhedged credit consistently report Sharpe ratios 50-100% higher than autocorrelation-corrected estimates.
• Skew and kurtosis. The standard deviation makes no distinction between positively and negatively skewed distributions. Two portfolios with the same Sharpe can have wildly different tail risks.
• Short data windows. A 12-month Sharpe is noise. With 12 monthly observations the 95% confidence interval on a Sharpe of 1.0 is roughly (0.0, 2.0) — useless for ranking.
• Backfill and survivorship bias. Reported Sharpe ratios for hedge funds are systematically inflated because failed funds drop out of databases. Industry studies suggest the true cross-sectional Sharpe distribution is shifted downward by 0.2–0.3.
• Sharpe doesn't reward skill. A leveraged Treasury position can have an excellent Sharpe with zero skill. Jensen's alpha and Information Ratio measure benchmark-relative skill; Sharpe measures total reward-to-risk.

Real-world applications

• Fund tearsheets. Every mutual fund and hedge fund reports Sharpe (over 1, 3, 5, 10-year windows) on its monthly performance summary. Morningstar uses it as one of three star-rating inputs.
• Manager evaluation. Pension funds and endowments rank external managers by Sharpe over 5-year horizons. CalPERS, Yale, Norway sovereign wealth all publish Sharpe-based dashboards.
• Strategy backtesting. Quant shops gate-keep strategy approvals at Sharpe ≥ 1.5 (with sensible Lo adjustment). Below that line, the historic noise overwhelms the signal.
• Capital allocation across strategies. Within a multi-strategy hedge fund, capital is reallocated quarterly toward strategies with rising Sharpe — a Markowitz mean-variance approach in disguise.
• Risk parity construction. Bridgewater's All Weather and similar strategies scale asset class exposures so each contributes equally to portfolio vol — effectively equalizing Sharpe ratios within the portfolio.
• Crypto strategy gating. Renaissance, Two Sigma, and Citadel reportedly require minimum 2-year live Sharpe ≥ 1.0 before allocating significant capital to a new crypto sub-strategy.

Common pitfalls

• Comparing Sharpe across data frequencies. A "Sharpe of 1.0" means nothing without specifying the period. Always annualize before quoting.
• Cherry-picking the window. A strategy that ran negative for 2008 looks fine if you compute Sharpe starting 2009. Use rolling Sharpe and report distributions, not point estimates.
• Mismatched risk-free rate. Use the rate of the appropriate currency and frequency — Treasury yields for U.S. dollar portfolios, OIS for institutional benchmarks, the relevant SOFR for short-horizon strategies.
• Ignoring fees and slippage. Gross-of-fees Sharpe overstates investable return. Apply realistic transaction costs (5–20 bps per trade for equities, more for small caps) and management fees before reporting.
• Treating Sharpe as the only metric. Maximum drawdown matters for capital preservation; tail risk matters during crises. A high-Sharpe / high-drawdown strategy can be unholdable in practice — emotional capacity diverges from variance.
• Forgetting leverage scaling. Sharpe is invariant to scaling, but real-world leverage faces margin and funding constraints. A 5x leveraged Treasury portfolio has the same Sharpe as unlevered — but a 2008-style funding squeeze can wipe it out.