The Risk and Statistics block on the Performance page has two controls in its header. One is a benchmark selector offering None, S&P 500, Nasdaq 100, Bitcoin and Total US Market. The other is a field labelled risk-free percent. Underneath them sit the Sharpe ratio, the Sortino ratio, the Calmar ratio, annual volatility, annual return and max drawdown, and below that a distribution quality row whose Omega tile carries the subtitle gains versus losses over RF.
The footer of that panel records what was in force at capture time. Observations 86, trades per year estimated at 327, days span 96, risk-free 4.5%. That footer is the most important line in the block and it is the first thing cropped out of any screenshot that gets pasted into a deck.
A control nobody re-reads after the first session
Assumption fields have a characteristic failure pattern. Somebody sets them once during onboarding, or accepts a default, and from then on every number downstream is quoted as though it were an observation rather than a function of a setting. The setting is invisible precisely because it is stable.
The discipline is to treat the field as part of the metric name. The figure in the Sharpe tile is not a Sharpe ratio, it is a Sharpe ratio at 4.5%, in the same way that a bond yield is not a number without a convention attached. Once you write it that way in your notes, the cross-period problem announces itself without any further analysis.
One thing worth establishing rather than assuming is which tiles consume the field. The panel labels Omega explicitly as being measured over the risk-free rate, and Sharpe and Sortino are excess-return constructions in standard usage, but the wiring of any particular implementation is not something to infer from a screenshot. Move the field, watch which figures change, and record the answer. It takes a minute and it converts a guess into a documented property of your reporting stack.
The arithmetic of a rate change, and where it bites hardest
For any ratio in the excess-return-over-risk family, the sensitivity to the rate assumption is simple. Change the annual rate by some amount and the ratio moves by roughly that amount divided by the annualised risk measure in the denominator.
On this panel the annual volatility reads 42.30%. At that level, moving the rate by 100 basis points moves a Sharpe-style ratio by about 0.024, which is immaterial. Run the same arithmetic on a market-neutral book at 6% volatility and 100 basis points is 0.17 of ratio, which is the difference between a mandate that clears an investment committee threshold and one that does not.
So the first rule is that the rate assumption matters in inverse proportion to volatility. High-volatility strategies are almost indifferent to it. Low-volatility, high-Sharpe strategies, which is to say the ones that get funded on the strength of their ratio, are the ones where a stale default is decisive. The managers most exposed to this error are the ones with the most to gain from it, which is a reason to be systematic rather than trusting.
Threshold-defined ratios behave differently and worse. Omega compares gains above a threshold with losses below it, and when the mass of the distribution sits near that threshold, small moves in the threshold reallocate observations from one side to the other. The sensitivity is not a smooth division by volatility, it depends on the local density. That is worth knowing before you compare an Omega across two periods computed at two different rates.
Why cross-period comparison breaks before cross-manager comparison does
Short rates have moved across a wide range over the last several years. A ratio computed on a 2021 window at a near-zero rate and a ratio computed on a current window at 4.5% are not two measurements of the same quantity, they are two different quantities that share a name.
There is a second and nastier version of the problem. If the rate is a single global setting rather than a stored property of each computation, then re-running an old period today reprices history. The figure you circulated last year is not reproducible unless somebody recorded the rate that produced it. This is not hypothetical bookkeeping. It is the reason a report from eighteen months ago will not match when a new analyst regenerates it, and the resulting hour of confusion always lands on the person who cannot explain the difference.
The artefact that fixes it already exists on this page. The footer prints the rate in force next to the observation count and the span. Capture that footer with every extract, store it with the figure, and the reproducibility problem disappears. Store the figure alone and you have built a number that cannot be checked.
The common-rate rule for comparing managers
When ratios arrive from outside, they arrive at whatever rate the sender chose, and the sender's choice was not made with your comparison in mind. Three rules make the comparison defensible.
- Fix one convention in writing and apply it to every candidate. The convention should name the instrument and the tenor, and it should specify whether the rate is the average over the measurement window or the rate at its start. Either is defensible. Mixing them is not.
- Never compare a self-reported ratio to one you computed. If the candidate cannot state the rate they used, do not adjust their number, recompute it from their return series at your rate. A ratio whose assumptions you cannot state is a ratio you cannot rank.
- Recompute the whole comparison set whenever the convention changes, and date-stamp the version. A ranking table that mixes vintages is worse than no table, because it looks authoritative.
The same logic applies internally across strategies in one book. Two sleeves compared on Sharpe at different rate assumptions will rank incorrectly whenever their volatilities differ, and the low-volatility sleeve is the one that gets misranked.
What travels with the number when it leaves the screen
The minimum disclosure attached to any ratio you circulate is five fields, and they belong in the same object as the figure rather than in a footnote.
The window with explicit dates. The observation count and the sampling frequency. The rate and its source. The definition of the input series, including whether external cash flows were removed. And whether the figure is gross or net of fees and financing.
This page supplies three of those five in its own footer, which is more than most reporting surfaces do. The remaining two are yours to establish and document once.
One caution to close on, because the temptation here is strong. The header strip on this page prints a Sharpe of 1.23 while the risk panel prints -4.246, and it is tempting to reach for the rate field as the explanation. Do not. The rate is one assumption among several that differ between any two implementations, alongside the input series, the window, the sampling frequency and the flow treatment. How much of a gap a rate change can and cannot account for on your own numbers is something you establish by moving the field and watching, not by reasoning about it from the outside. That experiment is a minute of work and it is the difference between a documented reporting stack and a plausible story.