Three-Year Small-Cap Record: Sharpe Comparison, Information Ratio and Statistical Significance
A small-cap manager has exactly 36 months of live history. Annualized return is 11.2%, the Russell 2000 returned 9.0%, and the risk-free rate averaged 3.0%. The portfolio's standard deviation is 18.5% versus 20.0% for the index, beta is 0.95, and tracking error is 4.2%. The manager's Sharpe ratio is 0.44; the index's is 0.30. The client's investment committee compares the manager with its large-cap manager (Sharpe 0.62) and with the S&P 500 (Sharpe 0.55), and moves to terminate, noting that 'the lowest Sharpe means the worst risk-adjusted performance'. The consultant notes that the policy benchmark is the Russell 2000 and asks for an information ratio and for a view on whether three years is enough to conclude skill. The manager's marketing deck instead shows a five-year simulated track record with a Sharpe ratio of 0.90.
1. Compute the information ratio and alpha, and assess whether the committee's Sharpe-ratio comparison is a valid basis for termination.
Read the worked answer
Alpha (CAPM, versus Russell 2000): required return = 3.0 + 0.95 × (9.0 − 3.0) = 8.70%. Alpha = 11.2 − 8.70 = +2.5% per year.
Information ratio: IR = alpha / tracking error = 2.5 / 4.2 = 0.60 (using active return relative to the policy benchmark, the active return is 11.2 − 9.0 = 2.2%, giving 2.2/4.2 = 0.52; the difference is the beta-adjusted versus unadjusted active return, and the report should state which is used).
Sharpe comparison: the committee's inference is invalid. The Sharpe ratio divides excess return by total volatility, which is dominated by the asset class, not by the manager. A small-cap portfolio and a large-cap portfolio have different return distributions, different factor exposures and different benchmark risk; ranking them by Sharpe is like comparing a bond fund and an equity fund. A lower Sharpe can simply mean a riskier asset class, not worse management. Within its own mandate, the manager's 0.44 versus the index's 0.30 is favourable, and the beta of 0.95 means it ran slightly less market risk than the index.
The correct comparison is against the policy benchmark, using active return and active risk: the IR of roughly 0.5–0.6 is respectable for a small-cap mandate. The committee should also look at Sortino (downside risk), up/down capture, style consistency and the process, not a single ratio computed across incomparable strategies.
2. Assess whether 36 months is enough to conclude skill, and what the standard error of the alpha estimate implies for the termination decision.
Read the worked answer
With three years of monthly data, the t-statistic on alpha is approximately IR × √years = 0.60 × √3 ≈ 1.04 (using the unadjusted active return, 0.52 × 1.73 ≈ 0.90). Either way it is around 1, well below the conventional 1.96 threshold for 95% confidence. In other words, a 2.5% alpha with 4.2% tracking error over three years is statistically indistinguishable from zero: the estimate is noisy, and the confidence interval spans negative values.
That does not mean the manager is unskilled; it means the sample cannot discriminate. Three years is a short window, and the problem is worse when returns are autocorrelated, when the strategy is concentrated, or when the period contains one style regime. A single strong year can drive the whole result. The committee should:
- Extend the evaluation to the longest available period, and examine rolling three-year IRs rather than one window.
- Check whether the alpha comes from a few positions or months (return concentration) and whether it survives after transaction costs.
- Look at process evidence: team stability, capacity, turnover, style drift.
- Pre-commit to a decision rule (for example, termination only after a stated number of years below a stated IR threshold) rather than reacting to a single Sharpe ranking.
Terminating on a Sharpe comparison across asset classes, without an IR-based test and without acknowledging the low power of a three-year sample, is a governance failure, not a performance conclusion.
3. How should the marketing deck's five-year simulated track record with a Sharpe of 0.90 be treated?
Read the worked answer
The simulated five-year record is not actual performance and cannot be presented as though it were. Under the GIPS standards, model or backtested results are not actual composite performance, cannot be linked to actual composite returns, and cannot be used to claim a five-year track record that the firm does not have. The firm's actual composite history is three years; that is what must be presented, with the since-inception date stated.
If the firm wants to show the simulation at all, it must be clearly labelled as hypothetical/supplemental, separated from the actual composite presentation, accompanied by the assumptions (universe, rebalancing, costs, survivorship, look-ahead bias) and a statement that the results do not represent actual trading and may not be reproducible. It must not be labelled 'GIPS-compliant', and the 0.90 Sharpe must not be compared with the live 0.44 as if the difference measured skill - simulated results typically exclude transaction costs, market impact and the behavioural drag of real trading.
The consultant should also ask whether the simulation was run with the current stock-selection rules applied to historical data, which is a classic source of look-ahead bias, and whether any live accounts were excluded from the simulation. Until those questions are answered, the deck's five-year Sharpe should be disregarded in the termination decision.