Alpha, Fees and a Benchmark Nobody Checked
Three years ago you hired a large-cap value manager for a client's $22M taxable account. Gross annual returns were 9.8%, 4.2% and 14.1%; the client's 60/40 policy benchmark returned 8.9%, 3.5% and 15.0%. The fee is 0.65%. At the annual review the manager states: 'We beat the benchmark in two of three years and added 92 basis points cumulatively.' Your holdings-based style analysis shows the portfolio tracks a large-cap value index that returned 11.2%, 5.0% and 12.8% over the same periods. The client's board asks: 'Do we fire them?'
1. Test the manager's claim. What is the actual excess return?
Read the worked answer
Compound; do not count years. Gross cumulative: 1.098 × 1.042 × 1.141 = 1.3054 → +30.54%. Benchmark: 1.089 × 1.035 × 1.15 = 1.2962 → +29.62%. The '92 basis points' is a compounding artifact - the arithmetic sum of annual excesses is 0.90 + 0.70 − 0.90 = 0.70pp, which compounds to 0.92pp.
Annualized: manager 1.3054^(1/3) − 1 = 9.29%; benchmark 9.03% → +0.26% a year gross.
Net of the 0.65% fee the sign flips: approximately 8.64% versus 9.03%, i.e. −0.39% a year, or about −1.39pp cumulative. 'Two of three years' is a counting fallacy: the single miss came in the largest up year (14.1% versus 15.0%), so the shortfall sits in up-capture - the most expensive place to lose it, because it compounds against asset growth.
2. Is that excess distinguishable from noise, and is the benchmark even the right test?
Read the worked answer
Significance. Annual excesses: +0.90, +0.70, −0.90. Mean 0.233; standard deviation 0.987; IR = 0.24. The t-statistic is 0.24 × √3 = 0.41. Nothing here is distinguishable from zero: three annual observations cannot separate skill from luck, and a manager with genuine 1% alpha has roughly a one-in-four chance of producing this record. An evaluation that cannot fail is not an evaluation.
Benchmark validity. The manager runs large-cap value; the policy benchmark is 60/40. That benchmark embeds an asset-allocation decision the manager never makes, so the 0.26% is not alpha - it is mostly the equity risk premium plus noise. Against a value style index (11.2%, 5.0%, 12.8% = 31.71% cumulative, 9.61% annualized), the manager lags 0.32% a year gross and about 0.97% net. The accurate verdict is not 'beat the benchmark' but 'underperformed its own style' - and even that rests on one noisy observation set.
The honest position: both comparisons are weak, and the wrong benchmark guarantees this argument repeats every year.
3. Do you fire them? Set out the sequence of decisions.
Read the worked answer
Fix the measurement before making the decision.
- Correct the benchmark first. Replace the 60/40 policy benchmark with a style-appropriate large-cap value benchmark, or a formal style blend from holdings-based analysis, and write it into the IPS. Until that exists, no termination decision is defensible.
- Extend the window. Pre-specify a five-year rolling evaluation with a return criterion (net excess > 0), a risk criterion (IR > 0.3) and a style-drift test. Three years is not a sample.
- Price the exit. This is a $22M taxable account. If unrealized gains are, illustratively, 30% of the account ($6.6M), tax at 28.8% is about $1.9M - 8.6% of the account. At a 1%/yr alpha expectation, that is more than eight years of alpha paid up front. A merely mediocre manager is often cheaper to keep than to fire.
- Verdict: retain, formally placed on watch against the pre-specified criteria, while the benchmark is corrected. If a five-year record against the corrected benchmark shows negative net alpha, and transition cost is less than about three years of expected alpha, terminate.
Document all four steps for the board. Their question 'do we fire them?' is really 'how would we know?' - and today the answer is that you cannot.