Look at what the single data set just produced. The portfolio beat its benchmark by 1.67 points a year. And:
| Measure | Portfolio | Benchmark | Verdict |
|---|---|---|---|
| Raw annualised return | 10.80% | 9.13% | beat |
| Sharpe | 0.512 | 0.549 | lost |
| Treynor | 4.89% | 5.13% | lost |
| Jensen's alpha | −0.35% | 0 by construction | lost |
| M² | 8.79% | 9.13% | lost |
| Information ratio | 0.369 | N/A | positive |
One manager, one series, four "lost", one "beat", one "positive". None of these numbers is wrong; they are answers to different questions, and the pattern is diagnostic. Total-risk measures (Sharpe, M²) and beta-risk measures (Treynor, Jensen) all say the same thing here: the excess return was bought with more risk than it was worth. The portfolio ran a beta of 1.392, so the security market line demanded 11.15% and it delivered 10.80%. Meanwhile the IR is positive because the manager did generate active return, reasonably consistently, per unit of active risk.
How do you read a disagreement between the ratios?
The professional reading is not "which number is right" but "what does the pattern say?" Four canonical patterns:
- Sharpe low, IR high. She adds value against her benchmark but the benchmark is a volatile place to be. A policy problem, not a manager problem: re-examine the allocation (AA1.01).
- Sharpe high, IR low. She hugs the index and the index did well. You may be paying active fees for beta; compute active share before renewing, which AA1.07 defines and computes on a ten-name book.
- Jensen positive, Sharpe below benchmark. Genuine alpha on top of high idiosyncratic risk: fine as one sleeve of a diversified whole (Treynor's frame), dangerous as someone's entire portfolio.
- All five positive, IR above 1.0 for years. Check the benchmark (see characteristics of a valid benchmark), check for return smoothing from stale pricing of illiquid holdings, check whether the record is a survivor. Sustained excellence is real but rare.
Advertisement
What traps catch the risk-adjusted ratios?
Two structural traps. Leverage is invisible to the ratios. Double every position at the risk-free rate: excess return doubles to 13.60% and volatility to 26.57%, so Sharpe is unchanged at 0.512 and Treynor at 4.89%, while Jensen's alpha doubles to −0.69%. Never rank on alpha without controlling for leverage. And all five assume roughly normal, independent returns. Real series are fat-tailed and negatively skewed, and any strategy that sells insurance (short volatility, carry, illiquid credit) reports a beautiful Sharpe right up until the tail arrives. Here Phase 9 re-enters and deserves a hearing: a ratio built on standard deviation defines risk as wobble, M9.01 defines it as permanent loss of capital. This grammar is worth fluency; it is not worth belief.
Five is also not the whole set, and the missing three are the ones that speak to the traps just named. AA1.07 adds three. The Sortino ratio replaces total volatility with downside deviation against a stated minimum acceptable return. Maximum drawdown and drawdown duration measure the loss a client actually experienced rather than the wobble around it. And the upside and downside capture ratios, read as a spread, say plainly whether a fund bought its excess return with leverage. It computes all three on this node's own twelve-quarter data set, so the comparison table extends from five measures to eight without a single input changing. AA1.07 also owns active share, which is the measure the Sharpe-high, IR-low diagnosis above sends you to compute.