About these numbers

The Performance tab carries a notice, and it is not boilerplate:

Returns aren't accurate yet. Performance and risk figures are still being reworked. Values and holdings are correct; the return percentages may not be.

This page exists so the rest of this section is read in that light.

What you can rely on today

What you hold, and what it is worth. Positions, quantities, current values, cost basis and the totals built from them are correct. The holdings table and your net worth are not affected by this.

Dividends actually received. Payments that have happened are recorded from your transaction history rather than derived.

What is being reworked

Return percentages, and the risk measures computed from them. That includes everything in the risk grid: Sharpe, Sortino, Calmar, volatility, maximum drawdown, value at risk, beta, alpha, tracking error, information ratio and the capture ratios.

Every one of those is derived from a return series. If the return series is being corrected, so is everything downstream of it.

Why the pages exist anyway

Two reasons.

The measures themselves are not in question. What a Sharpe ratio means, why it penalises upside, and why a Calmar ratio is unstable when the drawdown is small are properties of the measures, not of our implementation. That understanding is worth having whenever the figures settle.

And these pages describe how Gylder calculates them, read off the code rather than a textbook. That is what lets you tell a genuine result from an artefact. Several documented behaviours, such as Sortino dividing by the count of down days, or capture ratios summing rather than compounding, are exactly the sort of thing that makes a figure look wrong when it is working as written.

How to read the section for now

Use the pages to understand what a measure is asking, and be cautious about acting on the specific number next to it.

Where a figure looks implausible, the documented formula is the fastest way to work out whether it is a real property of your portfolio, a known simplification, or part of what is being corrected.

Known reasons a figure can look wrong, and be right

Worth ruling these out first, because none of them are the rework:

• Short windows. Annualising a strong month projects it across a year. See annualised return.
• The risk-free rate is a fixed 3% assumption, not a live market rate. It feeds Sharpe, Sortino and alpha, and matters most when returns are near 3%.
• History is downsampled to weekly deep in the past, so long windows are approximations.
• Manual holdings are excluded from contribution-based return and risk, so a portfolio heavy in them has risk measures covering less than all of it.
• Benchmark-relative figures need benchmark data. When it is missing they show a dash rather than a number.

This page will be updated when the notice comes off the Performance tab.