Benchmark-Relative Portfolio Evaluation
Separating absolute portfolio quality from active outcomes against a benchmark
QM013 · Portfolio Methods · Intermediate
Core idea. Benchmark-relative evaluation measures the portfolio in active space—what it earned, risked, and lost relative to a specified benchmark—without confusing relative underperformance with an absolute capital loss.
Use it for. Active return, tracking error, information ratio, relative wealth, benchmark-relative drawdown, and comparisons to same-universe or external benchmarks.
It does not establish. That one benchmark is uniquely correct, that positive absolute Sharpe implies successful active management, or that benchmark-relative drawdown means the portfolio itself lost money.
The Question
A portfolio can make money and still disappoint relative to its benchmark. It can also have an attractive absolute Sharpe ratio while taking persistent benchmark-relative risk that investors did not intend.
The central question is:
How should performance and risk be measured when the investment objective is explicitly relative to a benchmark?
Why It Matters
Absolute and benchmark-relative metrics answer different questions.
- Absolute volatility asks how variable the portfolio return is.
- Tracking error asks how variable the active return is.
- Sharpe ratio asks how much excess return over a cash or risk-free reference was earned per unit of total volatility.
- Information ratio asks how much average active return was earned per unit of active risk.
MSCI defines tracking error as the annualized standard deviation of active returns and the information ratio as active return divided by tracking error (MSCI 2013). These are standard active-management concepts, but their interpretation still depends on a correctly specified benchmark and return frequency.
Intuition
Absolute evaluation asks whether the portfolio itself was rewarding relative to its total risk. Benchmark-relative evaluation first subtracts or divides by the benchmark so that the analysis focuses on the active decision. The benchmark is therefore part of the estimand: changing it changes what “success” means.
The Method
Active Return
Let portfolio return be \(r_{p,t}\) and benchmark return be \(r_{b,t}\). The periodic active return is
\[ a_t=r_{p,t}-r_{b,t}. \]
For monthly observations, an arithmetic annualized mean active return can be written as
\[ \overline a_{ann}=12\,\overline a. \]
This is not the same object as the annualized compound growth rate of the relative-wealth ratio. Arithmetic active return is convenient for tracking-error and information-ratio calculations; geometric relative wealth answers a path-compounding question.
Tracking Error
With monthly active returns and sample standard deviation \(s(a)\),
\[ TE_{ann}=\sqrt{12}\,s(a). \]
Tracking error is often called active risk. It measures the dispersion of portfolio returns around benchmark returns, not the volatility of the portfolio in isolation (MSCI 2013).
Information Ratio
Using the arithmetic convention above,
\[ IR=\frac{\overline a_{ann}}{TE_{ann}} =\sqrt{12}\frac{\overline a}{s(a)}. \]
If tracking error is zero, the information ratio is undefined. A portfolio that exactly replicates the benchmark has no active-risk denominator; assigning it an infinite or zero information ratio would be misleading.
Benchmark-Relative Wealth
Active returns do not compound by simple summation. To follow the economic value of one dollar invested in the portfolio relative to one dollar invested in the benchmark, define
\[ W^p_t=\prod_{j=1}^{t}(1+r_{p,j}),\qquad W^b_t=\prod_{j=1}^{t}(1+r_{b,j}), \]
and the relative-wealth ratio
\[ Q_t=\frac{W^p_t}{W^b_t} =\prod_{j=1}^{t}\frac{1+r_{p,j}}{1+r_{b,j}}. \]
Then:
- \(Q_t>1\): the portfolio has compounded to more wealth than the benchmark since the common start;
- \(Q_t<1\): it has compounded to less.
This ratio is different from \(1+\sum a_t\).
Benchmark-Relative Drawdown
Define the running peak of relative wealth as
\[ H^Q_t=\max_{u\le t} Q_u. \]
The benchmark-relative drawdown is
\[ D^{rel}_t=\frac{Q_t}{H^Q_t}-1. \]
The maximum benchmark-relative drawdown is the minimum of this path.
A portfolio can have a negative benchmark-relative drawdown while its own wealth is rising. The statistic means the portfolio has fallen from a previous relative-performance peak versus the benchmark, not necessarily from an absolute wealth peak.
For clarity, avoid the phrase “active drawdown” unless the exact construction is defined. A drawdown of the relative-wealth ratio is economically different from taking a running peak of a cumulative arithmetic active-return series.
Same-Universe versus External Benchmarks
Benchmark choice changes the estimand.
Same-universe benchmark
Suppose a strategy ranks and selects among 11 ETFs. An equal-weight portfolio of those same 11 assets can be a useful benchmark because it holds the opportunity set largely fixed. Relative performance then speaks more directly to the value of selection, ranking, or weighting inside that universe.
External benchmark
A 60/40 portfolio, broad equity index, or institutional policy benchmark can be useful for practitioner relevance. But it changes the exposure mix and possibly the asset universe. Outperformance versus that benchmark is therefore a broader portfolio comparison, not a clean attribution of the selection rule alone.
A same-universe benchmark often sharpens method attribution. An external benchmark often sharpens economic relevance. Neither is automatically superior; the article should state what each comparison can and cannot isolate.
Financial / Economic Example
Suppose four monthly portfolio returns are
\[ (2\%,1\%,-1\%,3\%) \]
and benchmark returns are
\[ (1\%,1.5\%,-0.5\%,2\%). \]
The active returns are
\[ (1\%,-0.5\%,-0.5\%,1\%). \]
Their arithmetic mean is positive, but relative wealth must still be computed multiplicatively. The included Python example calculates active return, annualized tracking error, information ratio, relative wealth, and benchmark-relative drawdown from the same four periods.
Implementation
import numpy as np
def relative_wealth(portfolio, benchmark):
wp = np.cumprod(1.0 + np.asarray(portfolio, float))
wb = np.cumprod(1.0 + np.asarray(benchmark, float))
return wp / wbThe benchmark series must use the same return interval, currency convention, and net/gross treatment as the portfolio. If portfolio returns are net of modeled transaction costs while benchmark returns are costless, state that asymmetry rather than hiding it.
Common Mistakes
Calling tracking error portfolio volatility. TE is the volatility of active returns.
Using Sharpe as a substitute for IR. Sharpe is absolute; IR is benchmark-relative.
Adding active returns to create relative wealth. Relative wealth compounds the portfolio-to-benchmark gross-return ratio.
Calling relative drawdown a capital loss. It is a drawdown of the relative-wealth path.
Treating every benchmark as attributionally equivalent. Same-universe and external benchmarks answer different questions.
Mixing return conventions. Net portfolio returns versus gross benchmark returns can be a valid practitioner comparison, but it must be explicit.
When Not to Use It
Benchmark-relative metrics should not dominate when there is no economically meaningful benchmark. A market-neutral absolute-return mandate, for example, may be better evaluated against cash and explicit risk limits.
Likewise, a benchmark can become stale if the investable opportunity set or mandate changes. Relative metrics are only as meaningful as the benchmark contract they encode.
Used in SlackQuant Research
- The Decision Value of Price and Macro Information: uses same-universe and external benchmarks, active return, tracking error, information ratio, and relative wealth to distinguish ranking value from broader portfolio relevance.
- When Protection Works but the Portfolio Still Lags: uses benchmark-relative wealth paths and relative drawdown to show how protection can coexist with persistent benchmark lag.
Reproducibility
The deterministic Python example verifies active-return arithmetic, annualization, relative-wealth compounding, and benchmark-relative drawdown. The validator also checks a case in which portfolio wealth rises while relative wealth falls, demonstrating why relative drawdown is not an absolute loss statistic.