Information Ratio: Active Return, Tracking Error, and Python Implementation

The Information Ratio (IR) measures how consistently a strategy or portfolio manager generates alpha above a benchmark, relative to the volatility of that alpha. Where the Sharpe Ratio penalizes total volatility against a risk-free rate, the Information Ratio uses tracking error — the standard deviation of active returns — as its denominator. This makes IR the natural metric for any strategy evaluated relative to a benchmark: long-only equity funds, factor strategies, sector-rotation models, and tactical overlay programs.

If Sharpe asks "how much return per unit of total risk?", Information Ratio asks "how reliably do you beat your benchmark per unit of deviation from it?" A manager who consistently beats the S&P 500 by 2% with 1% tracking error (IR 2.0) is far more valuable than one who beats it by 5% with 20% tracking error (IR 0.25) — the first is a consistent alpha generator; the second is largely noise.

The Information Ratio Formula

The Information Ratio is the annualized active return divided by the annualized tracking error:

Where r_p is the portfolio return each period, r_b is the benchmark return each period, e_i = r_p,i − r_b,i is the active return each period, and N is the number of periods per year (252 daily, 52 weekly, 12 monthly). The annualized active return is mean(e) × N; the annualized tracking error is std(e) × √N. Dividing these cancels one √N factor, giving the compact form: mean(e) × √N / std(e).

This formula is structurally identical to the Sharpe Ratio, but applied to active returns versus a benchmark instead of excess returns versus the risk-free rate. If the benchmark is the risk-free rate, IR and Sharpe are mathematically equivalent.

Tracking Error

Tracking error (TE) is the annualized standard deviation of the period-by-period difference between portfolio and benchmark returns:

Low tracking error (0–2%) indicates a strategy that hugs its benchmark closely — a highly index-like portfolio with small active bets. High tracking error (10%+) indicates large concentrated positions that deviate substantially from the benchmark each period. Tracking error is the currency of active risk budget: a 5% annual TE means the manager is taking enough active risk to deviate from the benchmark by roughly 5 percentage points per year on average.

Python Implementation

import math

ANNUALIZATION = {
    "daily":   math.sqrt(252),
    "weekly":  math.sqrt(52),
    "monthly": math.sqrt(12),
    "annual":  1.0,
}

def information_ratio(
    returns: list[float],
    benchmark: list[float],
    periodicity: str = "daily",
) -> float:
    """
    Annualized active return divided by annualized tracking error.

    Args:
        returns:     portfolio period returns as decimals (e.g., 0.01 for 1%)
        benchmark:   benchmark period returns aligned to same dates
        periodicity: "daily" | "weekly" | "monthly" | "annual"

    Returns Information Ratio. Returns infinity when tracking error is zero
    and active return is positive; 0 when tracking error is zero and active
    return is non-positive.
    """
    n = min(len(returns), len(benchmark))
    if n < 2:
        return 0.0

    excess = [r - b for r, b in zip(returns[:n], benchmark[:n])]
    mean_e = sum(excess) / n
    variance = sum((e - mean_e) ** 2 for e in excess) / (n - 1)
    tracking_error = math.sqrt(variance)

    if tracking_error == 0:
        return float("inf") if mean_e > 0 else 0.0

    factor = ANNUALIZATION.get(periodicity, ANNUALIZATION["daily"])
    # IR = mean(excess) * sqrt(N) / std(excess) — identical to Sharpe on active returns
    return (mean_e / tracking_error) * factor

Example: Evaluating a Strategy Against the S&P 500

# Monthly returns (as decimals) from a long-only equity strategy and S&P 500
portfolio_monthly = [0.015, -0.014, 0.031, -0.001, 0.003, 0.003,
                     0.018, 0.000, 0.008, 0.014, -0.014, 0.018]
sp500_monthly     = [0.010, -0.007, 0.019, 0.005, -0.006, 0.011,
                     0.007, -0.003, 0.015, 0.004, -0.010, 0.012]

ir = information_ratio(portfolio_monthly, sp500_monthly, periodicity="monthly")
print(f"Information Ratio: {ir:.3f}")  # prints 0.880
# Average active return = 0.2% per month → 2.4% annualized
# Tracking error ≈ 0.79% per month → ~2.7% annualized
# IR ≈ 2.4 / 2.7 ≈ 0.88 — top decile (small sample; verify with full history)

Interpretation Benchmarks

Information Ratio benchmarks reflect the empirical distribution of active manager performance across the industry. The Fundamental Law (IR ≈ IC × √BR) implies that benchmarks scale with both skill and breadth — systematic strategies with hundreds of independent daily signals can legitimately achieve higher IR than discretionary managers making a few bets per year:

The Fundamental Law of Active Management

Grinold and Kahn's Fundamental Law formalizes the two drivers of Information Ratio:

The law has a critical implication for systematic strategy design: you can compensate for modest IC with high breadth. A discretionary macro manager making 10 independent calls per year with IC 0.15 achieves IR ≈ 0.15 × √10 ≈ 0.47. A systematic equity model making 1,000 independent daily bets with IC 0.015 achieves IR ≈ 0.015 × √1000 ≈ 0.47. Same expected IR, radically different mechanisms.

• IC (Information Coefficient): The Pearson correlation between your model's predicted return rank and the realized return rank across bets. IC of 0.05 is considered good for discretionary stock-picking; 0.02–0.03 is realistic for systematic models. IC decays quickly: a model that worked on training data often has IC near zero out-of-sample. Use rolling IC to monitor model health.
• BR (Breadth): The number of independent bets per year — not the number of trades. Trading 10 correlated tech stocks daily does not give BR = 2,520; the effective BR is roughly the number of independent signals in the universe. Grinold defines a bet as independent only if the alpha for one asset tells you nothing about the alpha for another. Adding correlated signals does not increase BR.
• IC × √BR decomposition: The formula reveals that improving IC from 0.03 to 0.06 (doubling skill) has the same effect on IR as increasing BR from 100 to 400 (quadrupling independent coverage). For most systematic strategies, the cheaper path to better IR is usually expanding the signal universe (higher BR), not improving the model (higher IC).

Information Ratio vs. Other Risk Metrics

The four ratios are complementary, not substitutes. A long-only equity fund should report Sharpe (for absolute risk-adjusted return), IR (for benchmark-relative consistency), and Calmar (for drawdown perspective). A CTA trend-follower with no benchmark should report Sharpe, Sortino, and Calmar. A fully market-neutral long-short fund with a cash benchmark can use IR and Sharpe interchangeably.

Failure Modes: When Information Ratio Misleads

• Benchmark selection bias: The single most impactful variable in the IR calculation is the choice of benchmark. A strategy benchmarked to cash will show very high IR if markets trend up; the same strategy benchmarked to the S&P 500 may show IR below zero. There is no universal correct benchmark — it must reflect the strategy's actual opportunity set and the mandate the manager was hired to fulfill.
• Short sample: Like the Sharpe Ratio, IR has high estimation error with fewer than 36 monthly observations. The standard error of IR is approximately 1/√T (for a Gaussian process), so a 12-month IR of 1.0 has a standard error of ~0.29 — it is statistically indistinguishable from 0.5. Always report the sample length alongside the IR.
• Non-stationary IC: The Fundamental Law assumes IC is stable over time. In practice, model IC decays as the signal's alpha erodes (capacity, crowding, changing regimes). A 3-year IR of 0.8 built on a signal that had IC 0.05 in years 1–2 and IC < 0.01 in year 3 will overstate the model's current expected IR substantially. Monitor rolling 12-month IC as a model health indicator.
• Ex-ante vs. ex-post TE divergence: Risk models predict ex-ante tracking error; what you actually experience is ex-post. In stress scenarios — 2008, 2020 — correlations spiked, and ex-post TE exceeded ex-ante TE by 2–5× at many funds. An IR computed on a calm-period backtest where ex-ante TE was 3% is not reliable when the realized TE was 10%.
• Survivorship bias in the benchmark: If the benchmark index excludes delisted stocks but the strategy held some, the strategy's active returns are downward-biased relative to the index. Use point-in-time constituent data for both the strategy and the benchmark when constructing a long history.

Information Ratio and Position Sizing

Information Ratio does not directly determine position size — that remains the Kelly Criterion's domain. But IR provides context for Kelly sizing: a strategy with IR 0.5 has meaningful but modest alpha above its benchmark; a strategy with IR 1.5 has strongly reliable alpha. Higher IR justifies larger active bets (higher active weight relative to the benchmark). Grinold's Optimal Active Weight formula formalizes this: the target active weight in each position is proportional to IC × √(TE budget), and the aggregate IR emerges from the IC and breadth of all positions combined.

The practical sequence for a benchmarked strategy: compute Sharpe (absolute risk-adjusted return), Calmar (drawdown survivability), and IR (benchmark-relative consistency) together. A strategy with IR > 0.5, Sharpe > 1.0, and Calmar > 1.0 across a multi-year sample with diverse regimes has a sound risk profile. Apply Kelly to size the total active risk budget; distribute it across positions using the Grinold optimal weight formula.