Sample institutional review with 10 variants

Where the data comes from

The 11 series (10 portfolios and the market) are generated with seed 20261010: 240 months, from 2006-01 to 2025-12. No month is a real month.

Their parameters come from summary statistics of the “10 Portfolios Formed on Momentum” (value-weighted) and of the market and momentum factors of the Kenneth R. French Data Library, from 1963-01 to 2025-12, downloaded on 2026-10-10: each decile's alpha, market and momentum exposure and residual volatility, and the mean and volatility of the market and of momentum.

We neither store nor show that data: each file carries the notice “Copyright 2026 Eugene F. Fama and Kenneth R. French”, is built from the CRSP database, and the library's page gives no licence to copy or redistribute it. That is why this sample uses synthetic series with parameters taken from those statistics.

It is a simplification: the deciles' residuals are generated independent of each other, and the market's tails (a Student's t with 6 degrees of freedom) and the momentum crashes were chosen by hand; they do not come from those files.

Kenneth R. French Data Library page

What was reviewed

DeclaredThe series: a synthetic series that imitates decile 10 of the momentum portfolios (the stocks with the highest return over the previous 12 months leaving out the last one), as a signal vendor would offer it. Monthly returns, as fractions.
MeasuredThe variants: the 10 decile portfolios, one column each, declared as the attempts the series was chosen from. Rigor counted 10 columns.
DeclaredThe benchmark: the market, with the cash rate added (total return).
DeclaredWho presents it: a vendor offering the signal to others, so the report speaks of what their clients will ask.

How it was measured

With Rigor's engine unchanged, the same as for any report on the site: verdict policy 2026-09-27-dependence-1, seed 12345 and 500 bootstrap resamples.
The variants' dates were compared with the series': they are unique, in order and the same.
No public data downloaded: in the alpha the cash rate is taken as zero, and the report says so.
The report uses its fund track record template (monthly returns someone presents to others): where it says “fund”, “manager” or “real history” it speaks of that template, not of the data, which in this sample is synthetic.
Every figure carries its tag: Measured (computed on the series), Declared (stated by whoever supplies the data) or Not measured (could not be computed from what was supplied).

Results and what each one means

Class C: there is a material weakness; we would not rely on this fund's track record until it is resolved.

Figure	Value	Evidence	What it means
Compound annual return	13.3%	Measured	How much the series grew per year, compounded, from 2006-01 to 2025-12.
Annual volatility	22.2%	Measured	How much the series moves in a typical year.
Annualised Sharpe	0.68	Measured	Mean return over its volatility, annualised, with no cash subtracted.
Maximum drawdown	-43.6%	Measured	The worst fall from an earlier peak.
Months under water	35	Measured	The longest stretch, in months, without regaining an earlier peak.
Probability that the true Sharpe is above zero (PSR)	0.998	Measured	As a single test, with the series' length, skew and tails. Passing needs 0.95 or more.
Deflated Sharpe (DSR) with 10 trials	0.912	Measured	The same probability, asking it to beat the best of 10 trials with no skill. With 0.95 or more it passes; between 0.5 and 0.95 it is weak.
Sharpe 10 trials with no skill would show	0.36	Measured	The best Sharpe to expect from pure luck among 10 trials with this length of history.
Years of history to leave that luck behind	5.8	Measured	With fewer years, a Sharpe like this one could come from picking the best of 10 trials.
Probability of overfitting (PBO, CSCV)	25.7%	Measured	In what share of the data's splits the best variant in sample falls below the median out of sample. Above 50% is a bad sign.
Benchmark's compound annual return	9.0%	Measured	The same measure for the market, over the same months.
Annual difference against the benchmark	+4.3%	Measured	The series' compound annual return minus the benchmark's.
Tracking error	11.4%	Measured	How far the series strays from the benchmark in a typical year.
Information ratio	0.45	Measured	The mean annual excess over the benchmark (arithmetic mean, not the compound difference above) per unit of tracking error.

The report's six questions

Question	What it measures	Result
Statistical significance	Does the result stand out from chance as a single test?	Pass
Number of settings tried	Does it still stand after counting the attempts it came from?	Weak
Costs	Does it hold up against trading costs?	Not measured
Out of sample	Does it hold on data not used to choose it?	Not measured
Data quality and trading pattern	Does the data have gaps, jumps or suspicious patterns?	Pass
Benchmark	Does it add anything over holding the benchmark?	Weak

Factor attribution

Rigor's engine splits the series' return against a single factor, the benchmark: how much the exposure to the market explains and how much is left as alpha.

Figure	Value	Evidence	What it means
Beta against the benchmark	1.25	Measured	How much the series moves for each 1 % of the market.
Share of the return the exposure to the benchmark explains	82.3%	Measured	What holding the market with that beta would give. With no cash rate, cash stays inside this exposure and the alpha.
Annual alpha	+2.7%	Measured	The return the exposure to the benchmark does not explain; its 95 % range runs from -2.8% to +8.1%.
Alpha's t-statistic	0.96	Measured	With t between -2 and 2, the alpha cannot be told apart from zero.

Not measured Attribution to several factors (size, value, momentum and others): in the extended review, computed apart from the Rigor report with the methodology agreed in the quote. The engine does not compute it yet, so there are no figures here.

What is not measured

Not measuredCosts: they need the gross and the net series, or the trades; without them the costs are not re-applied. Capacity and costs by scenario are in the extended review, computed apart from the Rigor report.
Not measuredOut of sample: the series does not say since which date its process has not changed; with that date declared, the report measures that stretch apart.
Not measuredHow the signal was built: the review reads the series supplied; it does not review the code, the source data or the execution, and it checks nothing against a broker or a custodian.
Not measuredThe future: no figure in this review is a prediction.

What a client would receive

These notes, written on their own series.
The full Rigor report, with all its sections.
The PDF with the notes and the report.
A 60-minute call and a re-run after 30 days.