Skill, luck and the short track record
Five years of monthly returns is a sample of sixty observations. That is rarely enough to distinguish a good process from a fortunate one — but it is enough to say how much uncertainty remains.
Philippe Trocellier — Founder — TP Advisory Services
This is a published outline, not a finished article. It sets out the argument and structure of a piece currently being written. The full text will replace it once complete.
Allocators are routinely asked to judge a process from a record too short to support the judgement. The honest response is not to refuse the question, but to quantify how much the record can actually establish.
Start with what the sample can support
The standard error of an estimated Sharpe ratio scales with the square root of the number of observations. For most real track records, the confidence interval is wide enough to contain both 'skilled' and 'lucky' — which is a finding, not a failure.
Remove what is not the manager's
- Strip out beta to the obvious market exposures.
- Strip out known factor tilts that can be replicated at low cost.
- What remains is the part of the record actually attributable to the process.
Look at behaviour, not only outcomes
Where the return series is too short to be conclusive, the decision series often is not. Turnover, position sizing, consistency of rule application and behaviour after losses are observable at a much higher frequency than performance.
Reporting the answer
The deliverable should state a range and the assumptions behind it, not a verdict. An allocator can act on 'the process is consistent, the record is too short to confirm it'. Nobody can act on a score with no error bar.