Information Ratio

The formula and what each piece is doing

The information ratio is structurally identical to Sharpe with two substitutions:

IR  =  (R_p − R_b)  /  TE

R_p − R_b = active return (excess of portfolio over benchmark)
TE        = tracking error = stdev(R_p − R_b)
            (standard deviation of the excess-return time series)

The numerator measures how much excess return the manager generated; the denominator measures how big a bet had to be taken to generate it. The ratio answers: per unit of "different-from-the-benchmark" risk that the manager loaded on, how much excess return did the investor receive? An IR of 0.5 means a fund running 4% tracking error has been generating 2% of annualised alpha; the manager is doing real work and the numbers are real.

Substituting r_f for R_b and σ for TE recovers the Sharpe ratio exactly. The two metrics answer different questions: Sharpe asks "is this portfolio worth holding instead of cash?" — information ratio asks "is this manager worth hiring instead of an index fund?"

Worked example: three managers, three IRs

Three managers run a US large-cap equity strategy against the S&P 500 over a 5-year window:

Manager	Excess return (ann.)	Tracking error (ann.)	IR	Verdict
A (closet indexer)	+0.4%	1.2%	0.33	Below "good" threshold; alpha barely visible above fees
B (standard active)	+2.0%	4.0%	0.50	"Good" — solid, hireable
C (concentrated)	+3.6%	8.0%	0.45	Higher excess return but worse skill-per-risk than B

Manager C looks superficially better with +3.6% excess return, but B is the more skilful manager — generating proportionally more alpha per unit of active bet taken. An institutional allocator who wants 2% of expected alpha can hire B with a 4% TE budget, or hire C with the same TE budget but expect only ~1.8% excess return on B's basis. Hiring decisions should optimise IR, not raw excess return — and then scale up the dollar TE budget separately.

Grinold's Fundamental Law of Active Management

Richard Grinold's 1989 Journal of Portfolio Management paper produced one of the most influential single equations in quantitative finance:

IR  ≈  IC  ·  sqrt(N)

IC  = Information Coefficient: correlation between forecast & realised return
N   = Breadth: number of independent active bets per year

The decomposition is exact under a simplifying assumption that all bets are independent and equally sized; in practice, both assumptions are violated and modern formulations (the "transfer coefficient" extension by Clarke, de Silva, and Thorley 2002) add a third multiplier for portfolio-construction efficiency. The conceptual implication is unchanged:

• Skill helps linearly. Doubling the information coefficient doubles IR. Skill is hard.
• Breadth helps as a square root. Quadrupling the number of independent bets only doubles IR. Breadth is much easier to scale than skill.
• Equivalent IRs via different paths. A discretionary stock-picker with IC = 0.20 running 25 bets a year has IR ≈ 1.0. A quant equity manager with IC = 0.05 running 400 bets a year also has IR ≈ 1.0. The first looks like Buffett; the second looks like AQR. Both are exceptional active managers.

The Law explains the rise of quantitative active management: by industrialising the bet-generation process across thousands of stocks, factor-style managers traded the difficulty of forecasting individual companies (high IC) for the scalability of mass production (high breadth). Renaissance Medallion's reported IR (~2-3 over decades, before fees) reflects a small army of breadth multiplied by uncommon IC.

Information ratio vs neighbouring metrics

Metric	Reference	Risk measure	Right question	Limitation
Sharpe ratio	Risk-free rate	Total stdev	Worth holding vs cash?	Indifferent to benchmark
Information ratio	Benchmark	Tracking error	Worth hiring vs index?	Benchmark choice matters
Sortino ratio	Min acceptable return	Downside dev	Asymmetric risk?	MAR is subjective
Treynor ratio	Risk-free rate	Beta to market	Per unit of systematic risk?	Ignores residual risk
Appraisal ratio (Treynor-Black)	CAPM model	Residual σ vs market model	Per unit of stock-specific risk?	Model-dependent (factor model)
Jensen's alpha	CAPM benchmark	Not normalised	Excess return per CAPM	Not a ratio — needs scaling
M² (Modigliani-Modigliani)	Risk-free + benchmark	Volatility-matched	Same risk as benchmark?	Same info content as Sharpe

Sharpe, Sortino, and IR form the standard three-ratio reporting package for institutional managers. Each captures a different facet of the same underlying performance, and a well-designed performance attribution shows all three with explicit assumptions about benchmark, MAR, and time window.

Origins: Treynor-Black to Grinold

The information ratio's intellectual roots run through Jack Treynor and Fischer Black's 1973 Journal of Business paper "How to Use Security Analysis to Improve Portfolio Selection." Treynor and Black introduced the "appraisal ratio" — α_i / σ_ε,i, the alpha-to-residual-risk ratio for an individual security — and showed that under simple assumptions the optimal portfolio weights each security proportionally to its appraisal ratio. The aggregate appraisal ratio of a portfolio becomes Grinold and Kahn's information ratio with a sign change in reference point.

Grinold's 1989 article and the subsequent textbook Active Portfolio Management (Grinold and Kahn, 1995; 2nd ed 1999) made IR the lingua franca of institutional active management. By the 2000s, Bloomberg terminals reported IR by default for any benchmark-aware fund, and consultants (Cambridge Associates, NEPC, Aksia) used IR thresholds as a first-pass screen. The information coefficient — originally a concept from analyst-forecast literature — became standard quant-equity language.

Implementation issues

• Benchmark choice is structural. A small-cap fund benchmarked to the S&P 500 will show inflated tracking error and potentially inflated IR. The right benchmark must be the closest passive alternative to the manager's mandate: Russell 2000 for US small-cap, MSCI EM for emerging markets, Bloomberg Aggregate for core bonds. SAA-aware allocators set benchmarks contractually; lazy reporting uses S&P 500 for everything.
• Sample size matters more than for Sharpe. Excess returns are typically smaller than total returns, so IR estimation is noisier. Five years of monthly data (60 observations) is the practical minimum; 10-year windows are standard. Quarterly IR comparisons are statistically meaningless.
• Fees: gross or net. Always state whether IR is computed gross or net of management fees. A 1% fee against a typical 4% TE budget represents a 0.25 IR deduction — the difference between "good" and "below threshold."
• Survivorship bias. Funds that close after persistent negative IR disappear from databases. Surviving funds' IRs are inflated. Adjustments (Carhart 1997; Brown-Goetzmann 1995) typically reduce the population-average IR by 0.5-1.5 percentage points.
• Concentration limits. Many institutional mandates impose tracking-error caps (often 6% or 8%). A high-conviction manager may be unable to scale their TE up to where IR-times-TE budget equals alpha target. The conceptual decomposition reveals such constraints as alpha-leaving-on-the-table.
• The breadth illusion. Grinold's Law assumes independent bets. Most actual portfolios have correlated active risks (sector tilts, factor exposures), so the effective breadth is far below the nominal bet count. The transfer coefficient (Clarke-de Silva-Thorley 2002) operationalises this discount.

How institutional allocators use it

• Manager hiring screens. 10-year net IR above 0.5 with consistent positive year-by-year IR is the standard institutional minimum. Above 0.8 attracts mandates.
• Tracking-error budgeting. Total fund TE = sqrt(Σ TE_i² for independent managers). With a 3% fund TE budget and IR = 0.7 across managers, expected alpha = 2.1% above benchmark.
• Performance fees. "Crystallise carry only above an IR of 0.5" or similar provisions tie incentive fees to skill demonstration, not just gross excess return.
• Multi-strategy combination. Combining managers with positive IR and uncorrelated tracking errors raises the fund-level IR via Grinold breadth. Bridgewater's All Weather, AQR's multi-strategy funds, and most fund-of-fund structures are explicit attempts at this.
• Active share complementarity. Cremers-Petajisto (2009) showed high active share is a necessary but not sufficient condition for positive IR. The two metrics together identify managers with both conviction and skill (high active share, positive IR) as the best hiring candidates.