The distribution quality block on the Performance overview prints four figures. On the capture I am working from, Omega ratio reads 0.073 with the subtitle gains versus losses over RF, tail ratio reads 0.070 with the subtitle right tail over left tail, annual volatility reads 42.30% and Calmar reads -2.174.
The first two are the useful pair for a funding decision, and they are useful for a reason that has nothing to do with sophistication. Both have a non-arbitrary anchor at 1.0, both are scale-free, and both can be explained to a non-quant on a committee in one sentence each. That combination is rare and it is what makes them a gate rather than a discussion.
Two ratios that fail a strategy faster than a Sharpe does
A Sharpe compresses a distribution into its first two moments and then divides. That works when the distribution is roughly symmetric and falls apart exactly where funding decisions are hardest, which is on strategies with skew, with fat left tails, or with a short record where the second moment is poorly estimated.
Omega compares the mass of gains above a threshold with the mass of losses below it, using the whole distribution rather than two summary numbers. At 1.0 the two sides balance. Below 1.0 the losses below the threshold outweigh the gains above it. The panel here names the threshold as the risk-free rate, which is set in the field at the top of the same block and recorded in the page footer, 4.5% on this capture.
Tail ratio compares the right tail with the left tail. At 1.0 the tails are symmetric, above 1.0 the upside tail is larger, below 1.0 the downside tail dominates. It answers the question a committee actually asks about a new strategy, which is not how volatile is it but what does the bad case look like relative to the good one.
Readings of 0.073 and 0.070 are not marginal. Both are an order of magnitude below their break-even anchors, and on any pre-committed gate that is a no-fund without further discussion. That is the point of a gate. It ends the conversation early on the cases that should end early, and preserves attention for the ones that are genuinely close.
Where the break-even sits and what a threshold above it means
Be precise about which part of this is arithmetic and which part is preference. That 1.0 is the balance point for both ratios is definitional. Any threshold you set above 1.0 is a policy choice made by your committee, and it should be written down as a preference rather than presented as a finding.
I have seen sensible desks use very different thresholds for defensible reasons. A desk funding many small allocations can afford a threshold close to 1.0 and rely on diversification. A desk making a small number of concentrated allocations needs a much larger margin because it cannot average away a mistake. Neither is right in the abstract, and any article that hands you a specific number is inventing it.
What is not a preference is the requirement to fix the threshold before you look at the candidate. A threshold set after seeing the figure is not a gate, it is a rationalisation, and everyone in the room knows it.
The observation count a tail ratio actually needs
This is where most gates are built badly, and the arithmetic is unforgiving.
A tail ratio compares two percentile estimates. If the tails are defined at 5%, then on a sample of 87 observations there are roughly four observations beyond each cut. The ratio is being driven by four points on each side, and one of those points arriving or leaving moves it substantially.
Work backwards from how many observations you want in each tail. For at least ten observations in a 5% tail you need a sample of 200. For a 10% tail definition you need 100. Those are not thresholds anybody discovered, they are division, which is exactly why they belong in a policy document. Specify the tail definition and the minimum sample in the same sentence, because either one alone is meaningless.
Then require two stability checks before the figure counts.
- Jackknife the extremes. Recompute with the single largest loss removed, then with the two largest. If the verdict flips, you do not have an estimate, you have one trade with a story attached, and the correct next step is to examine that trade rather than the ratio.
- Split the sample in half by time and recompute on each half. Two halves that disagree on the sign of the verdict mean the ratio is describing a regime rather than the strategy. The rolling Sharpe panel in the screenshot is the visual version of this test, and a series that ranges from -5.5782 to 0.6361 across three months is telling you the halves will not agree.
Turning the pair into a written gate
A gate is a pre-commitment with six clauses, and the specific numbers in the middle two are yours to set.
Clause one, minimum observations, stated with the tail definition that goes with it. Clause two, minimum span in independent periods rather than in days, because 86 daily observations across 96 days is far less evidence than the count suggests. Clause three, Omega at or above your threshold, computed at a stated risk-free rate that is the same for every candidate. Clause four, tail ratio at or above your threshold. Clause five, both figures stable under the jackknife and the time split. Clause six, both computed on a return series with external flows removed and the treatment documented.
Clause six is the one that gets skipped and the one that invalidates everything above it when it is skipped. If the underlying series is a raw account balance, deposits and withdrawals are sitting in the distribution alongside trading outcomes, and the tails you are measuring may be cash movements.
Write the gate so that it produces one of three outcomes rather than a score. Fund, do not fund, or insufficient data. That third outcome is the one that keeps the process honest, because it is the correct answer far more often than either of the others and it is the one a scoring system will never give you.
What the gate cannot see
Both ratios describe the shape of a realised distribution. Four things that decide whether an allocation works are invisible to them.
Capacity, because a distribution measured at small size says nothing about the same strategy at ten times the notional. Crowding, because the historical distribution was generated in a period when whoever else runs this trade was positioned however they were positioned. Regime coverage, because a sample that contains no stress event will show clean tails for the entirely uninformative reason that nothing has happened yet. And path risk inside each position, which on this account is genuinely unavailable, since the Distributions tab shows the hold time histogram at n equal to zero and the excursion histograms empty with a note that maximum favourable and adverse excursion are only recorded on paper trades.
So the pair is a fast negative screen and a weak positive one. Failing it should end the process. Passing it should move the candidate to the part of the diligence where capacity and crowding get argued about properly, which is the expensive part, and the entire value of the gate is in making sure you only spend that money on candidates who deserve it.